Method: Uses three coordinated agents: User Agents for persona-driven behaviors, Dialog Agent for persuasion strategies, and Optimization Agent for evaluation and refinement. Validates through Chain-of-Attitude modeling and LLM persuasion assessment.

Result: Applied to real-world marketing, significantly improved small LLMs’ persuasion capacity, increasing organic traffic conversion rate by 22.4% (from 1.83% to 2.24%).

Conclusion: MADS enables cost-effective generation of training data and demonstrates clear business value by enhancing persuasion capabilities in practical applications.

Abstract: We propose MADS (Multi-Agent Dialogue Simulation), a scalable framework for generating persuasive multi-turn dialogues via agent self-play. MADS employs three coordinated agents: User Agents simulating diverse persona-driven behaviors, a Dialog Agent executing task-oriented persuasion strategies and an Optimization Agent evaluating and refining dialogue outcomes. We further validate its effectiveness through users’ Chain-of-Attitude (CoA) modeling and dedicated LLMs’ persuasion assessment. This approach enables low-cost generation of training data without human annotation, addressing key industry challenges such as lack of user data, cold-start evaluation difficulties, and prompt inefficiency. Applied to a real-world marketing scenario, MADS significantly improved the persuasion capacity of small LLMs, increasing the organic traffic conversion rate by 22.4% (from 1.83% to 2.24%) , demonstrating clear business value.

Method: Created a dialogue management model using information state approach with predefined slots and text components, identified critical circumstances, defined information states and dialogue moves, implemented using LLM in-context learning with database queries based on slots and entity ordering.

Result: Achieved maximal inform and success rates on MultiWOZ test conversations, showing improvement over previous methods.

Conclusion: The proposed model effectively incorporates goal-aware planning through information state management and LLM-based implementation, demonstrating superior task completion performance.

Abstract: Task oriented dialogue systems (TOD) complete particular tasks based on user preferences across natural language interactions. Considering the impressive performance of large language models (LLMs) in natural language processing (NLP) tasks, most of the latest TODs are centered on LLMs. While proactive planning is crucial for task completion, many existing TODs overlook effective goal-aware planning. This paper creates a model for managing task-oriented conversations, conceptualized centered on the information state approach to dialogue management. The created model incorporated constructive intermediate information in planning. Initially, predefined slots and text part informational components are created to model user preferences. Investigating intermediate information, critical circumstances are identified. Informational components corresponding to these circumstances are created. Possible configurations for these informational components lead to limited information states. Then, dialogue moves, which indicate movement between these information states and the procedures that must be performed in the movements, are created. Eventually, the update strategy is constructed. The created model is implemented leveraging in-context learning of LLMs. In this model, database queries are created centered on indicated predefined slots and the order of retrieved entities is indicated centered on text part. This mechanism enables passing the whole corresponding entities to the preferences in the order of congruency. Evaluations exploiting the complete test conversations of MultiWOZ, with no more than a domain in a conversation, illustrate maximal inform and success, and improvement compared with previous methods.

Method: Trained two neural network models (6 and 10 hidden layers, 500 epochs each) using 25 features, tested on 1000 MT outputs from 7 systems compared against human references.

Result: The developed metrics showed better human correlation compared to other available MT evaluation metrics.

Conclusion: Supervised learning-based approach is effective for Gujarati MT evaluation, providing better performance than existing methods.

Abstract: Machine Translation (MT) Evaluation is an integral part of the MT development life cycle. Without analyzing the outputs of MT engines, it is impossible to evaluate the performance of an MT system. Through experiments, it has been identified that what works for English and other European languages does not work well with Indian languages. Thus, In this paper, we have introduced a reference-based MT evaluation metric for Gujarati which is based on supervised learning. We have trained two versions of the metric which uses 25 features for training. Among the two models, one model is trained using 6 hidden layers with 500 epochs while the other model is trained using 10 hidden layers with 500 epochs. To test the performance of the metric, we collected 1000 MT outputs of seven MT systems. These MT engine outputs were compared with 1 human reference translation. While comparing the developed metrics with other available metrics, it was found that the metrics produced better human correlations.

Method: Develops a theoretical framework that classifies hallucinations based on Closed World vs Open World assumptions, distinguishing correctable from unavoidable hallucinations under open-world conditions.

Result: Shows that while hallucinations can be mitigated in closed-world settings with consistent training/test distributions, they become inevitable in open-world environments where the environment is unbounded.

Conclusion: Hallucinations should be approached not merely as engineering defects but as structural features to be tolerated and made compatible with human intelligence, especially in AGI contexts.

Abstract: Large Language Models (LLMs) exhibit impressive linguistic competence but also produce inaccurate or fabricated outputs, often called hallucinations''. Engineering approaches usually regard hallucination as a defect to be minimized, while formal analyses have argued for its theoretical inevitability. Yet both perspectives remain incomplete when considering the conditions required for artificial general intelligence (AGI). This paper reframes hallucination’’ as a manifestation of the generalization problem. Under the Closed World assumption, where training and test distributions are consistent, hallucinations may be mitigated. Under the Open World assumption, however, where the environment is unbounded, hallucinations become inevitable. This paper further develops a classification of hallucination, distinguishing cases that may be corrected from those that appear unavoidable under open-world conditions. On this basis, it suggests that ``hallucination’’ should be approached not merely as an engineering defect but as a structural feature to be tolerated and made compatible with human intelligence.

Method: Applied Llama 3.1 model to process unstructured regional trade agreement texts using zero-shot, one-shot, and few-shot prompting techniques with positive and negative examples.

Result: Evaluated performance using quantitative and qualitative metrics, discussing key insights and challenges in triple extraction from economic texts.

Conclusion: Emphasizes the significance of language models in economic applications and identifies potential future directions for knowledge extraction from legal trade documents.

Abstract: This study investigates the effectiveness of Large Language Models (LLMs) for the extraction of structured knowledge in the form of Subject-Predicate-Object triples. We apply the setup for the domain of Economics application. The findings can be applied to a wide range of scenarios, including the creation of economic trade knowledge graphs from natural language legal trade agreement texts. As a use case, we apply the model to regional trade agreement texts to extract trade-related information triples. In particular, we explore the zero-shot, one-shot and few-shot prompting techniques, incorporating positive and negative examples, and evaluate their performance based on quantitative and qualitative metrics. Specifically, we used Llama 3.1 model to process the unstructured regional trade agreement texts and extract triples. We discuss key insights, challenges, and potential future directions, emphasizing the significance of language models in economic applications.

Method: Leverages original ESC training data to guide response generation for logical coherence, then uses reinforcement learning to refine the reasoning process.

Result: Significantly improves both logical soundness and supportive quality of responses in emotional support conversations.

Conclusion: CARE advances development of empathetic, cognitively robust, and human-like emotional support systems by explicitly enhancing cognitive reasoning.

Abstract: Emotional Support Conversation (ESC) plays a vital role in alleviating psychological stress and providing emotional value through dialogue. While recent studies have largely focused on data augmentation and synthetic corpus construction, they often overlook the deeper cognitive reasoning processes that underpin effective emotional support. To address this gap, we propose \textbf{CARE}, a novel framework that strengthens reasoning in ESC without relying on large-scale synthetic data. CARE leverages the original ESC training set to guide models in generating logically coherent and supportive responses, thereby explicitly enhancing cognitive reasoning. Building on this foundation, we further employ reinforcement learning to refine and reinforce the reasoning process. Experimental results demonstrate that CARE significantly improves both the logical soundness and supportive quality of responses, advancing the development of empathetic, cognitively robust, and human-like emotional support systems.

Method: Uses Item-ID + Oral-language Mixture-of-Experts (IDIOMoE) that splits Feed Forward Network into text expert and item expert with token-type gating, treating item interactions as a native dialect in language space.

Result: Demonstrates strong recommendation performance on both public and proprietary datasets while preserving the text understanding capabilities of the pretrained LLM.

Conclusion: IDIOMoE successfully unifies collaborative filtering and LLMs by enabling collaborative signals to be understood in the same way as natural language, avoiding destructive interference between modalities.

Abstract: While collaborative filtering delivers predictive accuracy and efficiency, and Large Language Models (LLMs) enable expressive and generalizable reasoning, modern recommendation systems must bring these strengths together. Growing user expectations, such as natural-language queries and transparent explanations, further highlight the need for a unified approach. However, doing so is nontrivial. Collaborative signals are often token-efficient but semantically opaque, while LLMs are semantically rich but struggle to model implicit user preferences when trained only on textual inputs. This paper introduces Item-ID

• Oral-language Mixture-of-Experts Language Model (IDIOMoE), which treats item interaction histories as a native dialect within the language space, enabling collaborative signals to be understood in the same way as natural language. By splitting the Feed Forward Network of each block of a pretrained LLM into a separate text expert and an item expert with token-type gating, our method avoids destructive interference between text and catalog modalities. IDIOMoE demonstrates strong recommendation performance across both public and proprietary datasets, while preserving the text understanding of the pretrained model.

Method: Supervised fine-tuning on datasets spanning general knowledge, mathematics, and open-ended trivia, evaluating two metacognitive tasks: (1) single-question confidence estimation, and (2) pairwise confidence comparison. Assessed generalization to unseen domains including medical and legal reasoning.

Result: Fine-tuning improves calibration (alignment between stated confidence and accuracy) and discrimination (higher confidence for correct vs. incorrect responses) within and across domains, while leaving accuracy unchanged. However, improvements are task-specific and do not transfer between single-question calibration and pairwise comparison tasks.

Conclusion: Uncertainty communication in LLMs is trainable and generalizable, but different metacognitive skills do not naturally reinforce one another and must be developed together through multitask training to achieve broader gains across domains.

Abstract: Large language models (LLMs) are increasingly used in decision-making contexts, but when they present answers without signaling low confidence, users may unknowingly act on erroneous outputs. While prior work shows that LLMs maintain internal uncertainty signals, their explicit verbalized confidence is typically miscalibrated and poorly discriminates between correct and incorrect answers. Across two types of LLMs, we investigate whether supervised finetuning can improve models’ ability to communicate uncertainty and whether such improvements generalize across tasks and domains. We finetune the LLMs on datasets spanning general knowledge, mathematics, and open-ended trivia, and evaluate two metacognitive tasks: (1) single-question confidence estimation, where the model assigns a numeric certainty to its answer, and (2) pairwise confidence comparison, where the model selects which of two answers it is more likely to have correct. We assess generalization to unseen domains, including medical and legal reasoning. Results show that finetuning improves calibration (alignment between stated confidence and accuracy) and discrimination (higher confidence for correct vs. incorrect responses) within and across domains, while leaving accuracy unchanged. However, improvements are task-specific: training on single-question calibration does not transfer to pairwise comparison, and vice versa. In contrast, multitask finetuning on both forms of metacognition yields broader gains, producing lower calibration error and stronger discrimination in out-of-domain evaluations. These results show that while uncertainty communication in LLMs is trainable and generalizable, different metacognitive skills do not naturally reinforce one another and must be developed together through multitask training.

Method: Proposes a BERT-based pipeline fine-tuned with binary cross-entropy and pairwise ranking loss for automated CIU extraction and ordering from Cookie Theft picture descriptions.

Result: Achieves 93% median precision, 96% median recall in CIU detection, 24% sequence error rates. Features show strong Pearson correlations with ground truth and outperform dictionary-based baseline. Comparable to manual annotations in group difference evaluation.

Conclusion: The pipeline effectively characterizes visual narrative paths for cognitive impairment assessment, with implementation and models open-sourced to public.

Abstract: Current methods for automated assessment of cognitive-linguistic impairment via picture description often neglect the visual narrative path - the sequence and locations of elements a speaker described in the picture. Analyses of spatio-semantic features capture this path using content information units (CIUs), but manual tagging or dictionary-based mapping is labor-intensive. This study proposes a BERT-based pipeline, fine tuned with binary cross-entropy and pairwise ranking loss, for automated CIU extraction and ordering from the Cookie Theft picture description. Evaluated by 5-fold cross-validation, it achieves 93% median precision, 96% median recall in CIU detection, and 24% sequence error rates. The proposed method extracts features that exhibit strong Pearson correlations with ground truth, surpassing the dictionary-based baseline in external validation. These features also perform comparably to those derived from manual annotations in evaluating group differences via ANCOVA. The pipeline is shown to effectively characterize visual narrative paths for cognitive impairment assessment, with the implementation and models open-sourced to public.

Method: Three approaches were evaluated: 1) classical supervised ML with embeddings and dimensionality reduction, 2) fine-tuning eight BERT variants for domain/skill alignment, and 3) ensemble learning with majority voting and stacking.

Result: DeBERTa-v3-base achieved highest F1 score (0.950) for domain alignment, RoBERTa-large achieved highest F1 score (0.869) for skill alignment. Ensemble models didn’t surpass best language models. Dimension reduction helped linear classifiers but not language models.

Conclusion: Fine-tuned language models demonstrated superior performance for automated item alignment to content standards compared to classical ML and ensemble methods.

Abstract: Accurate alignment of items to content standards is critical for valid score interpretation in large-scale assessments. This study evaluates three automated paradigms for aligning items with four domain and nineteen skill labels. First, we extracted embeddings and trained multiple classical supervised machine learning models, and further investigated the impact of dimensionality reduction on model performance. Second, we fine-tuned eight BERT model and its variants for both domain and skill alignment. Third, we explored ensemble learning with majority voting and stacking with multiple meta-models. The DeBERTa-v3-base achieved the highest weighted-average F1 score of 0.950 for domain alignment while the RoBERTa-large yielded the highest F1 score of 0.869 for skill alignment. Ensemble models did not surpass the best-performing language models. Dimension reduction enhanced linear classifiers based on embeddings but did not perform better than language models. This study demonstrated different methods in automated item alignment to content standards.}

Method: LST dynamically aggregates speech tokens into latent speech patches that serve as higher-level units, which can align with textual units or encapsulate common speech sequences like silences for better compute efficiency.

Result: LST outperforms vanilla approaches on speech-to-speech and text-to-text benchmarks in both data- and compute-controlled settings, achieving 6.5% absolute gain in speech accuracy on HellaSwag under compute-controlled training and 5.3% under data-controlled training.

Conclusion: LST enables more effective representational alignment and steeper scaling laws for speech-text models, demonstrating improved data efficiency and performance across speech and text tasks.

Abstract: Auto-regressive speech-text models are typically pre-trained on a large number of interleaved sequences of text tokens and raw speech encoded as speech tokens using vector quantization. These models have demonstrated state-of-the-art performance in speech-to-speech understanding and generation benchmarks, together with promising scaling laws, primarily enabled by the representational alignment between text and speech. Nevertheless, they suffer from shortcomings, partly owing to the disproportionately longer sequences of speech tokens in contrast to textual tokens. This results in a large compute imbalance between modalities during pre-training as well as during inference, and a potential hindrance to effectively aligning speech and text, ultimately translating to several orders of magnitude slower scaling laws. We introduce the Latent Speech-Text Transformer (LST), which makes pre-training speech-text models more data-efficient by dynamically and inexpensively aggregating speech tokens into latent speech patches. These patches serve as higher-level units that can either align with corresponding textual units to aid capability transfer or even encapsulate common speech sequences like silences to be more compute-efficient. We show that LST outperforms vanilla approaches on speech-to-speech as well as text-to-text benchmarks in both data- and compute-controlled settings, the former indicating more effective representational alignment and the latter indicating steeper scaling laws for speech-text models. On HellaSwag story completion, LST achieves 6.5% absolute gain in speech accuracy under compute-controlled training and 5.3% under data-controlled training, while also improving text performance. We will release our models, code, and the evaluation data to facilitate further research.

Method: Proposed Sub-CP framework leverages submodular objectives to control block diversity, supporting flexible selection strategies where each block can range from globally diverse to locally coherent.

Result: Extensive experiments across diverse tasks on multiple datasets show that Sub-CP consistently improves performance across model scales.

Conclusion: Sub-CP provides fine-grained control over semantic structure while enabling precomputation, offering a better approach for efficient in-context learning.

Abstract: In-context learning (ICL) enables efficient few-shot learning in large language models (LLMs) without training, but suffers from the quadratic input complexity of transformers, limiting the maximum number of exemplars. While various efficient ICL approaches partition the context into blocks to process (e.g., ensembling, compression, cross-attention), they often ignore the information redundancy or under-representation caused by different partition strategies, leading to suboptimal performance. To tackle this problem, we propose Sub-CP, a block-aware context selection framework that leverages submodular objectives to control block diversity. Sub-CP supports a flexible spectrum of selection strategies, allowing each block to range from globally diverse to locally coherent. This allows fine-grained control over semantic structure while enabling precomputation. Extensive experiments across diverse tasks on multiple datasets show that Sub-CP consistently improves performance across model scales.

Method: Proposed TKTO method eliminates the need for paired data and directly targets token-level units, automatically providing fine-grained alignment signals without requiring token-level annotations.

Result: TKTO improves Japanese TTS accuracy by 39%, reduces CER by 54%, and automatically assigns 12.8 times stronger reward to targeted tokens.

Conclusion: TKTO enables more data-efficient training and provides fine-grained token-level alignment for improved TTS performance, particularly for challenging languages like Japanese.

Abstract: Aligning text-to-speech (TTS) system outputs with human feedback through preference optimization has been shown to effectively improve the robustness and naturalness of language model-based TTS models. Current approaches primarily require paired desirable and undesirable samples at the utterance level. However, such pairs are often limited in TTS output data, and utterance-level formulation prevents fine-grained token-level optimization needed for accurate pronunciation alignment. In this study, we propose TKTO that eliminates the need for paired data, enabling a more data-efficient training paradigm, and directly targets token-level units, automatically providing fine-grained alignment signals without token-level annotations. TKTO improves the challenging Japanese TTS accuracy by 39% and reduces CER by 54%, automatically assigning 12.8 times stronger reward to targeted tokens.

Method: Hybrid semantic search system combining lightweight typo-tolerant lexical retrieval, embedding-based vector similarity, and constrained LLM re-ranking, leveraging existing infrastructure with intelligent caching and graceful degradation.

Result: Proposed framework includes required resources, phased implementation strategy, offline evaluation protocol (Hit@K, Precision@K, Recall@K, MRR), and online measurement methodology with query success metrics.

Conclusion: The approach ensures trustworthiness through low false-positive rates, evolvability for terminological changes, and economic efficiency while addressing search challenges in Head Start programs.

Abstract: Head Start programs utilizing GoEngage face significant challenges when new or rotating staff attempt to locate appropriate Tasks (modules) on the platform homepage. These difficulties arise from domain-specific jargon (e.g., IFPA, DRDP), system-specific nomenclature (e.g., Application Pool), and the inherent limitations of lexical search in handling typos and varied word ordering. We propose a pragmatic hybrid semantic search system that synergistically combines lightweight typo-tolerant lexical retrieval, embedding-based vector similarity, and constrained large language model (LLM) re-ranking. Our approach leverages the organization’s existing Task Repository and Knowledge Base infrastructure while ensuring trustworthiness through low false-positive rates, evolvability to accommodate terminological changes, and economic efficiency via intelligent caching, shortlist generation, and graceful degradation mechanisms. We provide a comprehensive framework detailing required resources, a phased implementation strategy with concrete milestones, an offline evaluation protocol utilizing curated test cases (Hit@K, Precision@K, Recall@K, MRR), and an online measurement methodology incorporating query success metrics, zero-result rates, and dwell-time proxies.

Method: Treat parallel reasoning as set-of-next-token-prediction, incorporate set-based global loss into SFT using self-supervised bipartite matching between global forking tokens and unique reasoning traces.

Result: SSFT preserves reasoning modes and produces emergent global forking tokens, consistently outperforming SFT on multiple reasoning benchmarks under both Pass@1 and Cons@k metrics.

Conclusion: Set-based supervised fine-tuning effectively addresses the diversity-accuracy tradeoff in parallel reasoning by preserving unique reasoning modes through global loss and bipartite matching.

Abstract: Although LLMs have demonstrated improved performance by scaling parallel test-time compute, doing so relies on generating reasoning paths that are both diverse and accurate. For challenging problems, the forking tokens that trigger diverse yet correct reasoning modes are typically deep in the sampling tree. Consequently, common strategies to encourage diversity, such as temperature scaling, encounter a worsened trade-off between diversity and accuracy. Motivated by this challenge, we treat parallel reasoning as a set-of-next-token-prediction problem, and incorporate a set-based global loss into Supervised Fine-Tuning (SFT) using self-supervised bipartite matching between our global forking tokens and unique reasoning traces. We observe that, while naive fine-tuning with multiple reasoning traces collapses these unique reasoning modes, our proposed method, Set Supervised Fine-Tuning (SSFT), preserves these modes and produces emergent global forking tokens. Experiments on multiple reasoning benchmarks show that our SSFT consistently outperforms SFT under both Pass@1 and Cons@k metrics.

Method: Controlled empirical study using Pythia model suite (410M-12B parameters) across five diverse tasks, evaluating models after 1-3 training iterations with synthetic data proportions ranging from 0-50%.

Result: Models maintain stable performance with up to 20% synthetic data but degradation accelerates beyond 30%; larger models (6.9B-12B) show greater robustness than smaller models (410M-1.4B); calibration degradation precedes accuracy loss; reasoning tasks degrade faster than retrieval tasks.

Conclusion: Current best practices (like STaR and Self-Instruct with >80% external data) operate within safe regimes; practical guidance provided for synthetic data budgets based on model scale and task requirements.

Abstract: Synthetic data generated by large language models has become integral to modern NLP training pipelines, from bootstrapping reasoning capabilities to augmenting instruction-following datasets. While recent work demonstrates successful applications maintaining high external data ratios, systematic understanding of how synthetic data proportion affects model behavior across different scales remains limited. This paper presents a controlled empirical study examining model performance, calibration, and output characteristics when trained on varying synthetic-to-external data ratios. Using the Pythia model suite (410M-12B parameters) across five diverse tasks, we evaluate models after one to three training iterations with synthetic data proportions ranging from 0-50%. Our key findings include: models maintain stable performance with up to 20% synthetic data, but degradation accelerates beyond 30%; larger models (6.9B-12B) show greater robustness to synthetic data than smaller models (410M-1.4B); calibration degradation precedes accuracy loss, providing an early warning signal; and task characteristics matter, with reasoning tasks degrading faster than retrieval tasks under synthetic data training. Importantly, we find that current best practices, such as those employed in STaR and Self-Instruct systems that maintain greater than 80% external data, operate well within safe regimes identified by our experiments. We provide practical guidance for practitioners on synthetic data budgets based on model scale and task requirements, alongside detailed comparison with concurrent work including Shumailov et al.’s model collapse findings.

Method: Proposed a curiosity-driven LLM-as-a-judge approach that learns personalized creative judgments using the Torrance Test of Creative Thinking benchmark with human expert annotations.

Result: Showed improvements over baseline supervised fine-tuning across various evaluation metrics (Pearson correlation, Cohen’s, F1 values), especially when annotators disagree.

Conclusion: The method enables models of various sizes to learn nuanced creative judgments of different individuals, making it particularly useful for subjective evaluations with annotator disagreement.

Abstract: Modern large language models (LLMs) excel at objective tasks such as evaluating mathematical reasoning and factual accuracy, yet they falter when faced with the nuanced, subjective nature of assessing creativity. In this work, we propose a novel curiosity-driven LLM-as-a-judge for evaluating creative writing which is personlized to each individual’s creative judgments. We use the Torrance Test of Creative Thinking(TTCW) benchmark introduced in Chakrabarty et al. (2024), which has stories annotated by expert humans across various subjective dimensions like Originality, to test our hypothesis. We show that our method enables models across various sizes, to learn the nuanced creative judgments of different individuals, by showing improvements over baseline supervised finetuning(SFT) method across various evaluation metrics like Pearson correlation, Cohen’s and F1 values. Our method is especially useful in subjective evaluations where not all the annotators agree with each other.

Method: The authors categorize existing research along dimensions including linguistic description levels, models included, genres analyzed, languages analyzed, and prompting approaches, using this scheme to present findings and expose current research trends.

Result: AI-generated text is more likely to contain formal and impersonal style (more nouns, determiners, adpositions; fewer adjectives/adverbs), lower lexical diversity, smaller vocabulary size, and repetitive text. Current research is concentrated on English data and GPT models.

Conclusion: Current research remains heavily focused on English and GPT models, highlighting the need for broader cross-linguistic and cross-model investigation, as well as addressing prompt sensitivity issues through multiple prompt wordings in future studies.

Abstract: Large language models (LLMs) are solidifying their position in the modern world as effective tools for the automatic generation of text. Their use is quickly becoming commonplace in fields such as education, healthcare, and scientific research. There is a growing need to study the linguistic features present in AI-generated text, as the increasing presence of such texts has profound implications in various disciplines such as corpus linguistics, computational linguistics, and natural language processing. Many observations have already been made, however a broader synthesis of the findings made so far is required to provide a better understanding of the topic. The present survey paper aims to provide such a synthesis of extant research. We categorize the existing works along several dimensions, including the levels of linguistic description, the models included, the genres analyzed, the languages analyzed, and the approach to prompting. Additionally, the same scheme is used to present the findings made so far and expose the current trends followed by researchers. Among the most-often reported findings is the observation that AI-generated text is more likely to contain a more formal and impersonal style, signaled by the increased presence of nouns, determiners, and adpositions and the lower reliance on adjectives and adverbs. AI-generated text is also more likely to feature a lower lexical diversity, a smaller vocabulary size, and repetitive text. Current research, however, remains heavily concentrated on English data and mostly on text generated by the GPT model family, highlighting the need for broader cross-linguistic and cross-model investigation. In most cases authors also fail to address the issue of prompt sensitivity, leaving much room for future studies that employ multiple prompt wordings in the text generation phase.

Method: Evaluated nine text-only models and three multimodal audio-text models on DementiaBank speech corpus using various adaptations: in-context learning with different demonstration selection policies, reasoning-augmented prompting, parameter-efficient fine-tuning, and multimodal integration.

Result: Class-centroid demonstrations achieved highest in-context learning performance, reasoning improved smaller models, token-level fine-tuning produced best scores, and adding classification heads improved underperforming models. Multimodal models performed well but didn’t surpass top text-only models.

Conclusion: Model adaptation strategies critically influence speech-based dementia detection, and properly adapted open-weight models can match or exceed commercial systems.

Abstract: Over half of US adults with Alzheimer disease and related dementias remain undiagnosed, and speech-based screening offers a scalable detection approach. We compared large language model adaptation strategies for dementia detection using the DementiaBank speech corpus, evaluating nine text-only models and three multimodal audio-text models on recordings from DementiaBank speech corpus. Adaptations included in-context learning with different demonstration selection policies, reasoning-augmented prompting, parameter-efficient fine-tuning, and multimodal integration. Results showed that class-centroid demonstrations achieved the highest in-context learning performance, reasoning improved smaller models, and token-level fine-tuning generally produced the best scores. Adding a classification head substantially improved underperforming models. Among multimodal models, fine-tuned audio-text systems performed well but did not surpass the top text-only models. These findings highlight that model adaptation strategies, including demonstration selection, reasoning design, and tuning method, critically influence speech-based dementia detection, and that properly adapted open-weight models can match or exceed commercial systems.

Method: Created WebDetective benchmark with hint-free multi-hop questions paired with a controlled Wikipedia sandbox ensuring full traceability of model actions, and developed a holistic evaluation framework that separates search sufficiency, knowledge utilization, and refusal behavior. Evaluated 25 state-of-the-art models and developed EvidenceLoop agentic workflow with verification loops and systematic evidence tracking.

Result: Evaluation revealed systematic weaknesses across all architectures: models struggle with knowledge utilization despite having sufficient evidence and demonstrate near-absent appropriate refusal when evidence is lacking. EvidenceLoop workflow improved both search and synthesis capabilities, demonstrating that WebDetective’s diagnostic framework can guide concrete architectural improvements.

Conclusion: Today’s systems excel at executing given reasoning paths but fail when required to discover them. WebDetective establishes itself as a critical tool for developing genuinely autonomous reasoning systems rather than pattern-following agents.

Abstract: RAG (Retrieval-Augmented Generation) systems and web agents are increasingly evaluated on multi-hop deep search tasks, yet current practice suffers from two major limitations. First, most benchmarks leak the reasoning path in the question text, allowing models to follow surface cues rather than discover reasoning chains autonomously. Second, evaluation is typically reduced to a single pass rate, which collapses diverse behaviours into one score and obscures whether failures stem from inadequate search, poor knowledge use, or inappropriate refusal. To address these issues, we present WebDetective, a benchmark of hint-free multi-hop questions paired with a controlled Wikipedia sandbox that ensures full traceability of model actions, and a holistic evaluation framework that separates search sufficiency, knowledge utilisation, and refusal behaviour. Our evaluation of 25 state-of-the-art models reveals systematic weaknesses across all architectures: models struggle with knowledge utilisation despite having sufficient evidence and demonstrate near-absent appropriate refusal when evidence is lacking. These patterns expose a fundamental gap: today’s systems excel at executing given reasoning paths but fail when required to discover them. We develop an agentic workflow, EvidenceLoop, that explicitly targets the challenges our benchmark identifies, incorporating verification loops and systematic evidence tracking that improve both search and synthesis capabilities. This baseline demonstrates that WebDetective’s diagnostic framework can guide concrete architectural improvements, establishing our benchmark as a critical tool for developing genuinely autonomous reasoning systems rather than pattern-following agents.

Method: LiRA uses a multi-agent collaborative workflow with specialized agents for content outlining, subsection writing, editing, and reviewing to emulate human literature review processes.

Result: LiRA outperforms baselines like AutoSurvey and MASS-Survey in writing and citation quality on SciReviewGen and ScienceDirect datasets, while maintaining competitive similarity to human-written reviews. It also shows robustness to reviewer model variation.

Conclusion: Agentic LLM workflows like LiRA can improve the reliability and usability of automated scientific writing even without domain-specific tuning, demonstrating significant potential for literature review automation.

Abstract: The rapid growth of scientific publications has made it increasingly difficult to keep literature reviews comprehensive and up-to-date. Though prior work has focused on automating retrieval and screening, the writing phase of systematic reviews remains largely under-explored, especially with regard to readability and factual accuracy. To address this, we present LiRA (Literature Review Agents), a multi-agent collaborative workflow which emulates the human literature review process. LiRA utilizes specialized agents for content outlining, subsection writing, editing, and reviewing, producing cohesive and comprehensive review articles. Evaluated on SciReviewGen and a proprietary ScienceDirect dataset, LiRA outperforms current baselines such as AutoSurvey and MASS-Survey in writing and citation quality, while maintaining competitive similarity to human-written reviews. We further evaluate LiRA in real-world scenarios using document retrieval and assess its robustness to reviewer model variation. Our findings highlight the potential of agentic LLM workflows, even without domain-specific tuning, to improve the reliability and usability of automated scientific writing.

Method: Proposed NLD-LLM framework using diverse transformer models (Qwen, DeepSeek, Phi, LLaMA, Mistral) with comprehensive prompt design strategy including standardized formatting, clear task guidance, and iterative refinement process. Evaluated using semantic and structural metrics.

Result: Prompt engineering significantly impacts model effectiveness, with smaller models often performing competitively when supported by well-crafted prompts.

Conclusion: Well-designed prompts can enable smaller language models to achieve competitive performance in natural language description tasks, highlighting the importance of prompt engineering over model size alone.

Abstract: Natural Language Description (NLD) is a Natural Language Processing (NLP) task that requires models to generate structured and meaningful outputs from natural language inputs. In this work, we propose NLD-LLM, a systematic NLP framework to evaluate the performance of language models to generate accurate and concise source code descriptions. This framework incorporates a diverse set of transformer models, including Qwen, DeepSeek, Phi, LLaMA, and Mistral, spanning various sizes, architectures, and training approaches. Central to NLD-LLM is a comprehensive prompt design strategy that includes standardized formatting, clear task guidance, and NLD prompting, ensuring fair and consistent evaluation. Additionally, we apply an iterative refinement process to improve output’s quality and assess the model’s adaptability. Using semantic and structural metrics, our analysis demonstrates that prompt engineering significantly impacts the effectiveness of the model such that smaller models often performing competitively when supported by well-crafted prompts.

Method: Analysis of LLM capabilities compared to human cognitive limitations, focusing on memory differences. Proposes creating models with human-like memory constraints.

Result: LLMs’ superhuman prediction ability stems from superior long-term memory (facts/training examples) and short-term memory (text context), causing them to underestimate human reading difficulty.

Conclusion: Need to develop models with human-like memory constraints and collect better human data to properly evaluate progress in cognitive modeling.

Abstract: When people listen to or read a sentence, they actively make predictions about upcoming words: words that are less predictable are generally read more slowly than predictable ones. The success of large language models (LLMs), which, like humans, make predictions about upcoming words, has motivated exploring the use of these models as cognitive models of human linguistic prediction. Surprisingly, in the last few years, as language models have become better at predicting the next word, their ability to predict human reading behavior has declined. This is because LLMs are able to predict upcoming words much better than people can, leading them to predict lower processing difficulty in reading than observed in human experiments; in other words, mainstream LLMs are ‘superhuman’ as models of language comprehension. In this position paper, we argue that LLMs’ superhumanness is primarily driven by two factors: compared to humans, LLMs have much stronger long-term memory for facts and training examples, and they have much better short-term memory for previous words in the text. We advocate for creating models that have human-like long-term and short-term memory, and outline some possible directions for achieving this goal. Finally, we argue that currently available human data is insufficient to measure progress towards this goal, and outline human experiments that can address this gap.

Method: Multi-stage information extraction pipeline using large language models with iterative extraction and source tracking to capture 47 features spanning composition, processing, microstructure, and properties from experimental materials reports.

Result: Achieved F1 scores around 0.96 at both feature and tuple levels. Improved microstructure category F1 by 10.0% (feature) and 13.7% (tuple), reduced missed materials from 49 to 13 out of 396 (miss rate from 12.4% to 3.3%) in MPEA studies.

Conclusion: The pipeline enables scalable literature mining with high precision, minimal omissions, and zero false positives, providing trustworthy datasets for machine learning while generalizing to diverse material classes.

Abstract: Data-driven materials discovery requires large-scale experimental datasets, yet most of the information remains trapped in unstructured literature. Existing extraction efforts often focus on a limited set of features and have not addressed the integrated composition-processing-microstructure-property relationships essential for understanding materials behavior, thereby posing challenges for building comprehensive databases. To address this gap, we propose a multi-stage information extraction pipeline powered by large language models, which captures 47 features spanning composition, processing, microstructure, and properties exclusively from experimentally reported materials. The pipeline integrates iterative extraction with source tracking to enhance both accuracy and reliability. Evaluations at the feature level (independent attributes) and tuple level (interdependent features) yielded F1 scores around 0.96. Compared with single-pass extraction without source tracking, our approach improved F1 scores of microstructure category by 10.0% (feature level) and 13.7% (tuple level), and reduced missed materials from 49 to 13 out of 396 materials in 100 articles on precipitate-containing multi-principal element alloys (miss rate reduced from 12.4% to 3.3%). The pipeline enables scalable and efficient literature mining, producing databases with high precision, minimal omissions, and zero false positives. These datasets provide trustworthy inputs for machine learning and materials informatics, while the modular design generalizes to diverse material classes, enabling comprehensive materials information extraction.

Method: Created SynCED-EnDe with 1,000 gold-labeled and 8,000 silver-labeled sentence pairs from 2024-2025 sources (StackExchange, GOV.UK), balanced 50/50 between error and non-error cases, with explicit error subclasses and auxiliary judgments.

Result: Benchmark experiments with XLM-R and related encoders show substantial performance gains over WMT21 due to balanced labels and refined annotations.

Conclusion: SynCED-EnDe serves as a community resource to advance safe deployment of MT in information retrieval and conversational assistants, particularly for emerging contexts like wearable AI devices.

Abstract: Critical Error Detection (CED) in machine translation aims to determine whether a translation is safe to use or contains unacceptable deviations in meaning. While the WMT21 English-German CED dataset provided the first benchmark, it is limited in scale, label balance, domain coverage, and temporal freshness. We present SynCED-EnDe, a new resource consisting of 1,000 gold-labeled and 8,000 silver-labeled sentence pairs, balanced 50/50 between error and non-error cases. SynCED-EnDe draws from diverse 2024-2025 sources (StackExchange, GOV.UK) and introduces explicit error subclasses, structured trigger flags, and fine-grained auxiliary judgments (obviousness, severity, localization complexity, contextual dependency, adequacy deviation). These enrichments enable systematic analyses of error risk and intricacy beyond binary detection. The dataset is permanently hosted on GitHub and Hugging Face, accompanied by documentation, annotation guidelines, and baseline scripts. Benchmark experiments with XLM-R and related encoders show substantial performance gains over WMT21 due to balanced labels and refined annotations. We envision SynCED-EnDe as a community resource to advance safe deployment of MT in information retrieval and conversational assistants, particularly in emerging contexts such as wearable AI devices.

Method: Proposes Directed Decoding Map (DDM) to capture structural relationships between decoding steps, overcoming redundancy in model confidence, and Gaussian-Trajectory Attribution (GTA) that fits cell-wise Gaussian distributions at each decoding position to compute attribution scores based on trajectory likelihood.

Result: Extensive experiments validate the utility of the proposed methods, showing improved performance over directly using model confidence for attribution tasks.

Conclusion: The decoding mechanism of dLLMs can be effectively leveraged for model attribution through structural analysis of decoding trajectories, providing a powerful tool for distinguishing between different models and model versions.

Abstract: Discrete Diffusion Large Language Models (dLLMs) have recently emerged as a competitive paradigm for non-autoregressive language modeling. Their distinctive decoding mechanism enables faster inference speed and strong performance in code generation and mathematical tasks. In this work, we show that the decoding mechanism of dLLMs not only enhances model utility but also can be used as a powerful tool for model attribution. A key challenge in this problem lies in the diversity of attribution scenarios, including distinguishing between different models as well as between different checkpoints or backups of the same model. To ensure broad applicability, we identify two fundamental problems: what information to extract from the decoding trajectory, and how to utilize it effectively. We first observe that relying directly on per-step model confidence yields poor performance. This is mainly due to the bidirectional decoding nature of dLLMs: each newly decoded token influences the confidence of other decoded tokens, making model confidence highly redundant and washing out structural signal regarding decoding order or dependencies. To overcome this, we propose a novel information extraction scheme called the Directed Decoding Map (DDM), which captures structural relationships between decoding steps and better reveals model-specific behaviors. Furthermore, to make full use of the extracted structural information during attribution, we propose Gaussian-Trajectory Attribution (GTA), where we fit a cell-wise Gaussian distribution at each decoding position for each target model, and define the likelihood of a trajectory as the attribution score: if a trajectory exhibits higher log-likelihood under the distribution of a specific model, it is more likely to have been generated by that model. Extensive experiments under different settings validate the utility of our methods.

Method: A strictly causal thinking paradigm that reasons incrementally during audio streaming with no lookahead, amortizing reasoning during listening windows to avoid additional latency.

Result: Experiments show consistent improvements in response quality through both objective metrics and human evaluations, with robust handling of conversational dynamics.

Conclusion: Chronological Thinking effectively bridges the gap between human-like conversational thinking and full-duplex spoken dialogue systems, improving response quality while maintaining real-time performance.

Abstract: Recent advances in spoken dialogue language models (SDLMs) reflect growing interest in shifting from turn-based to full-duplex systems, where the models continuously perceive user speech streams while generating responses. This simultaneous listening and speaking design enables real-time interaction and the agent can handle dynamic conversational behaviors like user barge-in. However, during the listening phase, existing systems keep the agent idle by repeatedly predicting the silence token, which departs from human behavior: we usually engage in lightweight thinking during conversation rather than remaining absent-minded. Inspired by this, we propose Chronological Thinking, a on-the-fly conversational thinking mechanism that aims to improve response quality in full-duplex SDLMs. Specifically, chronological thinking presents a paradigm shift from conventional LLM thinking approaches, such as Chain-of-Thought, purpose-built for streaming acoustic input. (1) Strictly causal: the agent reasons incrementally while listening, updating internal hypotheses only from past audio with no lookahead. (2) No additional latency: reasoning is amortized during the listening window; once the user stops speaking, the agent halts thinking and begins speaking without further delay. Experiments demonstrate the effectiveness of chronological thinking through both objective metrics and human evaluations show consistent improvements in response quality. Furthermore, chronological thinking robustly handles conversational dynamics and attains competitive performance on full-duplex interaction metrics.

Method: Analyzed FinMA’s model architecture, instruction tuning using Financial Instruction Tuning (FIT) dataset, and evaluation under FLARE benchmark for financial tasks.

Result: FinMA performs well in sentiment analysis and classification tasks, but struggles with numerical reasoning, entity recognition, and summarization.

Conclusion: This research advances understanding of how financial LLMs should be designed and evaluated to effectively support finance-related decision-making processes.

Abstract: This research explores the strengths and weaknesses of domain-adapted Large Language Models (LLMs) in the context of financial natural language processing (NLP). The analysis centers on FinMA, a model created within the PIXIU framework, which is evaluated for its performance in specialized financial tasks. Recognizing the critical demands of accuracy, reliability, and domain adaptation in financial applications, this study examines FinMA’s model architecture, its instruction tuning process utilizing the Financial Instruction Tuning (FIT) dataset, and its evaluation under the FLARE benchmark. Findings indicate that FinMA performs well in sentiment analysis and classification, but faces notable challenges in tasks involving numerical reasoning, entity recognition, and summarization. This work aims to advance the understanding of how financial LLMs can be effectively designed and evaluated to assist in finance-related decision-making processes.

Method: Tested various delimiters across leading model families (Llama, Qwen, Gemma) on MMLU, analyzed attention head scores to understand the mechanism, and explored methods to improve robustness including specifying delimiters in prompts.

Result: Performance varied dramatically (±23%) based on delimiter choice, with the ability to manipulate model rankings by changing a single character. Good delimiters steer attention toward key tokens. Specifying delimiters in prompts improved robustness.

Conclusion: LLMs are surprisingly brittle to delimiter choice across topics and model families, with no improvement at scale. Practical recommendations include using specific delimiters and explicitly stating them in prompts for better robustness.

Abstract: Common Large Language model (LLM) evaluations rely on demonstration examples to steer models’ responses to the desired style. While the number of examples used has been studied and standardized, the choice of how to format examples is less investigated. In evaluation protocols and real world usage, users face the choice how to separate in-context examples: use a comma? new line? semi-colon? hashtag? etc.? Surprisingly, we find this seemingly minor choice can dramatically alter model response quality. Across leading model families (Llama, Qwen, Gemma), performance on MMLU for example can vary by $\pm 23%$ depending on the choice of delimiter. In fact, one can manipulate model rankings to put any model in the lead by only modifying the single character separating examples. We find LLMs’ brittleness pervades topics, model families, and doesn’t improve with scale. By probing attention head scores, we find that good-performing delimiters steer attention towards key tokens in the input. Finally, we explore methods to improve LLMs’ robustness to the choice of delimiter. We find specifying the selected delimiter in the prompt boosts robustness and offer practical recommendations for the best-performing delimiters to select.

Method: Created DeliberationBank dataset with opinion data from 3,000 participants and summary judgments from 4,500 participants, then trained DeliberationJudge (fine-tuned DeBERTa) to rate summaries from individual perspectives.

Result: DeliberationJudge is more efficient and better aligned with human judgments than LLM judges. Evaluation of 18 LLMs revealed persistent weaknesses, especially underrepresentation of minority positions.

Conclusion: The framework provides a scalable and reliable way to evaluate deliberation summarization, helping ensure AI systems are more representative and equitable for policymaking.

Abstract: Large-scale public deliberations generate thousands of free-form contributions that must be synthesized into representative and neutral summaries for policy use. While LLMs have been shown as a promising tool to generate summaries for large-scale deliberations, they also risk underrepresenting minority perspectives and exhibiting bias with respect to the input order, raising fairness concerns in high-stakes contexts. Studying and fixing these issues requires a comprehensive evaluation at a large scale, yet current practice often relies on LLMs as judges, which show weak alignment with human judgments. To address this, we present DeliberationBank, a large-scale human-grounded dataset with (1) opinion data spanning ten deliberation questions created by 3,000 participants and (2) summary judgment data annotated by 4,500 participants across four dimensions (representativeness, informativeness, neutrality, policy approval). Using these datasets, we train DeliberationJudge, a fine-tuned DeBERTa model that can rate deliberation summaries from individual perspectives. DeliberationJudge is more efficient and more aligned with human judgements compared to a wide range of LLM judges. With DeliberationJudge, we evaluate 18 LLMs and reveal persistent weaknesses in deliberation summarization, especially underrepresentation of minority positions. Our framework provides a scalable and reliable way to evaluate deliberation summarization, helping ensure AI systems are more representative and equitable for policymaking.

Method: Systematic taxonomy and controlled reproduction of hallucination types through prompt engineering, mapping hallucinations into vector space using embeddings, and analyzing with unsupervised learning in reduced-dimensional representations.

Result: Quantitative evaluation shows consistent correlation between hallucination severity and spatial divergence from correct outputs, demonstrating that simple classification algorithms can reliably distinguish hallucinations.

Conclusion: The approach provides theoretical and empirical evidence for effective hallucination detection within a single LLM, offering a lightweight framework to improve model reliability.

Abstract: This work introduces a novel methodology for the automatic detection of hallucinations generated during large language model (LLM) inference. The proposed approach is based on a systematic taxonomy and controlled reproduction of diverse hallucination types through prompt engineering. A dedicated hallucination dataset is subsequently mapped into a vector space using an embedding model and analyzed with unsupervised learning techniques in a reduced-dimensional representation of hallucinations with veridical responses. Quantitative evaluation of inter-centroid distances reveals a consistent correlation between the severity of informational distortion in hallucinations and their spatial divergence from the cluster of correct outputs. These findings provide theoretical and empirical evidence that even simple classification algorithms can reliably distinguish hallucinations from accurate responses within a single LLM, thereby offering a lightweight yet effective framework for improving model reliability.

Method: EAD implements temperature annealing during generation - starting with high temperature to encourage semantic diversity on early tokens, then gradually lowering temperature to maintain sample quality and policy alignment.

Result: EAD significantly improves sample efficiency and consistently outperforms fixed-temperature sampling across various RLVR algorithms and model sizes, while being lightweight and plug-and-play.

Conclusion: Aligning exploration with the natural dynamics of sequential generation (explore-at-the-beginning, exploit-at-the-end) offers a robust approach to improving LLM reasoning capabilities.

Abstract: Reinforcement learning with verifiable rewards (RLVR) is a powerful paradigm for enhancing the reasoning capabilities of large language models (LLMs), yet its success hinges on effective exploration. An ideal exploration strategy must navigate two fundamental challenges: it must preserve sample quality while also ensuring training stability. While standard fixed-temperature sampling is simple, it struggles to balance these competing demands, as high temperatures degrade sample quality and low temperatures limit discovery. In this work, we propose a simpler and more effective strategy, Exploratory Annealed Decoding (EAD), grounded in the insight that exploration is most impactful on early tokens which define a sequence’s semantic direction. EAD implements an intuitive explore-at-the-beginning, exploit-at-the-end strategy by annealing the sampling temperature from high to low during generation. This dynamic schedule encourages meaningful, high-level diversity at the start, then gradually lowers the temperature to preserve sample quality and keep the sampling distribution close to the target policy, which is essential for stable training. We demonstrate that EAD is a lightweight, plug-and-play method that significantly improves sample efficiency, consistently outperforming fixed-temperature sampling across various RLVR algorithms and model sizes. Our work suggests that aligning exploration with the natural dynamics of sequential generation offers a robust path to improving LLM reasoning.

Method: Created Camellia benchmark with 19,530 manually annotated entities and 2,173 masked contexts from social media, evaluated 4 multilingual LLM families on cultural context adaptation, sentiment association, and entity extractive QA tasks.

Result: LLMs struggle with cultural adaptation across all Asian languages, show distinct cultural biases in sentiment association, and have performance gaps in entity extraction between cultures due to poor context understanding.

Conclusion: LLMs exhibit significant cultural biases in Asian languages, with performance varying based on access to culturally-relevant data and distinct bias patterns across model families, highlighting the need for better cultural adaptation.

Abstract: As Large Language Models (LLMs) gain stronger multilingual capabilities, their ability to handle culturally diverse entities becomes crucial. Prior work has shown that LLMs often favor Western-associated entities in Arabic, raising concerns about cultural fairness. Due to the lack of multilingual benchmarks, it remains unclear if such biases also manifest in different non-Western languages. In this paper, we introduce Camellia, a benchmark for measuring entity-centric cultural biases in nine Asian languages spanning six distinct Asian cultures. Camellia includes 19,530 entities manually annotated for association with the specific Asian or Western culture, as well as 2,173 naturally occurring masked contexts for entities derived from social media posts. Using Camellia, we evaluate cultural biases in four recent multilingual LLM families across various tasks such as cultural context adaptation, sentiment association, and entity extractive QA. Our analyses show a struggle by LLMs at cultural adaptation in all Asian languages, with performance differing across models developed in regions with varying access to culturally-relevant data. We further observe that different LLM families hold their distinct biases, differing in how they associate cultures with particular sentiments. Lastly, we find that LLMs struggle with context understanding in Asian languages, creating performance gaps between cultures in entity extraction.

Method: Systematic evaluation of 3 Llama Guards and 2 GPT-oss models in RAG systems, testing how inserting benign documents into guardrail context affects judgments, and analyzing effects of retrieved documents, user query, and LLM-generated response components.

Result: Inserting benign documents alters guardrail judgments in approximately 11% of input cases and 8% of output cases, making them unreliable. Tested mitigation methods only provided minor improvements.

Conclusion: Current guardrails have a significant context-robustness gap, motivating the need for training and evaluation protocols that are robust to retrieval and query composition.

Abstract: With the increasing adoption of large language models (LLMs), ensuring the safety of LLM systems has become a pressing concern. External LLM-based guardrail models have emerged as a popular solution to screen unsafe inputs and outputs, but they are themselves fine-tuned or prompt-engineered LLMs that are vulnerable to data distribution shifts. In this paper, taking Retrieval Augmentation Generation (RAG) as a case study, we investigated how robust LLM-based guardrails are against additional information embedded in the context. Through a systematic evaluation of 3 Llama Guards and 2 GPT-oss models, we confirmed that inserting benign documents into the guardrail context alters the judgments of input and output guardrails in around 11% and 8% of cases, making them unreliable. We separately analyzed the effect of each component in the augmented context: retrieved documents, user query, and LLM-generated response. The two mitigation methods we tested only bring minor improvements. These results expose a context-robustness gap in current guardrails and motivate training and evaluation protocols that are robust to retrieval and query composition.

Method: Created WeatherArchive-Bench with two tasks: WeatherArchive-Retrieval (locating relevant passages from over 1M archival news segments) and WeatherArchive-Assessment (evaluating LLMs’ ability to classify societal vulnerability and resilience indicators).

Result: Experiments show dense retrievers often fail on historical terminology, while LLMs frequently misinterpret vulnerability and resilience concepts, revealing limitations in reasoning about complex societal indicators.

Conclusion: The benchmark provides insights for designing more robust climate-focused RAG systems and highlights key limitations in current approaches to analyzing historical weather archives.

Abstract: Historical archives on weather events are collections of enduring primary source records that offer rich, untapped narratives of how societies have experienced and responded to extreme weather events. These qualitative accounts provide insights into societal vulnerability and resilience that are largely absent from meteorological records, making them valuable for climate scientists to understand societal responses. However, their vast scale, noisy digitized quality, and archaic language make it difficult to transform them into structured knowledge for climate research. To address this challenge, we introduce WeatherArchive-Bench, the first benchmark for evaluating retrieval-augmented generation (RAG) systems on historical weather archives. WeatherArchive-Bench comprises two tasks: WeatherArchive-Retrieval, which measures a system’s ability to locate historically relevant passages from over one million archival news segments, and WeatherArchive-Assessment, which evaluates whether Large Language Models (LLMs) can classify societal vulnerability and resilience indicators from extreme weather narratives. Extensive experiments across sparse, dense, and re-ranking retrievers, as well as a diverse set of LLMs, reveal that dense retrievers often fail on historical terminology, while LLMs frequently misinterpret vulnerability and resilience concepts. These findings highlight key limitations in reasoning about complex societal indicators and provide insights for designing more robust climate-focused RAG systems from archival contexts. The constructed dataset and evaluation framework are publicly available at https://anonymous.4open.science/r/WeatherArchive-Bench/.

Method: RS uses neural networks to predict the residual error in similarity scores from interpretable systems, maintaining interpretability while improving accuracy.

Result: Evaluation across four datasets shows RS matches state-of-the-art performance while providing faithful and interpretable predictions.

Conclusion: RS enables high-performance authorship verification with maintained interpretability, addressing the explainability gap in neural methods.

Abstract: Responsible use of Authorship Verification (AV) systems not only requires high accuracy but also interpretable solutions. More importantly, for systems to be used to make decisions with real-world consequences requires the model’s prediction to be explainable using interpretable features that can be traced to the original texts. Neural methods achieve high accuracies, but their representations lack direct interpretability. Furthermore, LLM predictions cannot be explained faithfully – if there is an explanation given for a prediction, it doesn’t represent the reasoning process behind the model’s prediction. In this paper, we introduce Residualized Similarity (RS), a novel method that supplements systems using interpretable features with a neural network to improve their performance while maintaining interpretability. Authorship verification is fundamentally a similarity task, where the goal is to measure how alike two documents are. The key idea is to use the neural network to predict a similarity residual, i.e. the error in the similarity predicted by the interpretable system. Our evaluation across four datasets shows that not only can we match the performance of state-of-the-art authorship verification models, but we can show how and to what degree the final prediction is faithful and interpretable.

Method: Critical analysis of various approaches including sub-quadratic attention variants, recurrent neural networks, state space models, and hybrid architectures, evaluating compute/memory complexity and benchmark results.

Result: The paper provides a comprehensive assessment of alternative architectures that could potentially challenge pure-attention Transformers’ dominance.

Conclusion: The dominance of pure-attention Transformers may soon be challenged by more efficient architectures that overcome the quadratic complexity bottleneck.

Abstract: Transformers have dominated sequence processing tasks for the past seven years – most notably language modeling. However, the inherent quadratic complexity of their attention mechanism remains a significant bottleneck as context length increases. This paper surveys recent efforts to overcome this bottleneck, including advances in (sub-quadratic) attention variants, recurrent neural networks, state space models, and hybrid architectures. We critically analyze these approaches in terms of compute and memory complexity, benchmark results, and fundamental limitations to assess whether the dominance of pure-attention transformers may soon be challenged.

Method: Systematic experiments across 5 LLMs on math, QA, and coding tasks, testing performance with perfect retrieval, whitespace replacement, and forced attention to relevant tokens. Also tested placing evidence immediately before questions.

Result: Performance degraded 13.9%-85% with increasing input length even with perfect retrieval. Degradation occurred even with minimal distraction and forced attention. The mitigation strategy improved GPT-4o performance by up to 4%.

Conclusion: Input length alone harms LLM performance independently of retrieval quality, revealing a previously unrealized limitation. A simple recitation-based mitigation strategy can help transform long-context tasks into short-context ones.

Abstract: Large language models (LLMs) often fail to scale their performance on long-context tasks performance in line with the context lengths they support. This gap is commonly attributed to retrieval failures – the models’ inability to identify relevant information in the long inputs. Accordingly, recent efforts often focus on evaluating and improving LLMs’ retrieval performance: if retrieval is perfect, a model should, in principle, perform just as well on a long input as it does on a short one – or should it? This paper presents findings that the answer to this question may be negative. Our systematic experiments across 5 open- and closed-source LLMs on math, question answering, and coding tasks reveal that, even when models can perfectly retrieve all relevant information, their performance still degrades substantially (13.9%–85%) as input length increases but remains well within the models’ claimed lengths. This failure occurs even when the irrelevant tokens are replaced with minimally distracting whitespace, and, more surprisingly, when they are all masked and the models are forced to attend only to the relevant tokens. A similar performance drop is observed when all relevant evidence is placed immediately before the question. Our findings reveal a previously-unrealized limitation: the sheer length of the input alone can hurt LLM performance, independent of retrieval quality and without any distraction. They motivate our simple, model-agnostic mitigation strategy that transforms a long-context task into a short-context one by prompting the model to recite the retrieved evidence before attempting to solve the problem. On RULER, we observe a consistent improvement of GPT-4o up to 4% on an already strong baseline.

Method: Graph-based methods to build cross-lingual mental health ontologies that capture culturally embedded expressions of distress, align them across languages, and link with clinical terminology.

Result: Framework enables culturally valid representations for AI systems in mental health care, addressing gaps in healthcare communication for multilingual contexts.

Conclusion: CL-PDE framework provides more inclusive and patient-centric NLP tools for mental health care by grounding AI systems in culturally appropriate representations across Indian languages.

Abstract: Mental health communication in India is linguistically fragmented, culturally diverse, and often underrepresented in clinical NLP. Current health ontologies and mental health resources are dominated by diagnostic frameworks centered on English or Western culture, leaving a gap in representing patient distress expressions in Indian languages. We propose cross-linguistic graphs of patient stress expressions (CL-PDE), a framework for building cross-lingual mental health ontologies through graph-based methods that capture culturally embedded expressions of distress, align them across languages, and link them with clinical terminology. Our approach addresses critical gaps in healthcare communication by grounding AI systems in culturally valid representations, allowing more inclusive and patient-centric NLP tools for mental health care in multilingual contexts.

Method: Applied Direct Preference Optimization (DPO) to adapt Mistral-7B using 8,838 heart failure nursing notes from MIMIC-III and 21,210 preference pairs derived from expert-verified GPT outputs, model generations, and original notes.

Result: BLEU increased by 84% (0.173 to 0.318), BERTScore improved by 7.6% (0.828 to 0.891), expert ratings rose across accuracy (+14.4), completeness (+14.5), logical consistency (+14.1), readability (+11.1), and structural clarity (+6.0).

Conclusion: DPO can align lightweight clinical language models with expert standards, supporting privacy-preserving, AI-assisted documentation within EHR systems to reduce administrative burden and improve ICU patient safety.

Abstract: Nursing documentation in intensive care units (ICUs) provides essential clinical intelligence but often suffers from inconsistent terminology, informal styles, and lack of standardization, challenges that are particularly critical in heart failure care. This study applies Direct Preference Optimization (DPO) to adapt Mistral-7B, a locally deployable language model, using 8,838 heart failure nursing notes from the MIMIC-III database and 21,210 preference pairs derived from expert-verified GPT outputs, model generations, and original notes. Evaluation across BLEU, ROUGE, BERTScore, Perplexity, and expert qualitative assessments demonstrates that DPO markedly enhances documentation quality. Specifically, BLEU increased by 84% (0.173 to 0.318), BERTScore improved by 7.6% (0.828 to 0.891), and expert ratings rose across accuracy (+14.4 points), completeness (+14.5 points), logical consistency (+14.1 points), readability (+11.1 points), and structural clarity (+6.0 points). These results indicate that DPO can align lightweight clinical language models with expert standards, supporting privacy-preserving, AI-assisted documentation within electronic health record systems to reduce administrative burden and improve ICU patient safety.

Method: A multi-agent system powered by Llama-3.3 70B Instruct model that decomposes structural analysis into subtasks: Problem Analysis Agent extracts parameters, Geometry Agent derives node coordinates and element connectivity, Translation Agent converts to OpenSeesPy code, Model Validation Agent performs consistency checks, and Load Agent applies load conditions.

Result: The system achieves accuracy over 80% in most cases across 10 repeated trials on 20 benchmark problems, outperforming Gemini-2.5 Pro and ChatGPT-4o models.

Conclusion: The developed LLM-based multi-agent system successfully automates finite element modeling for 2D frames, demonstrating high accuracy and outperforming other large language models in structural engineering tasks.

Abstract: Large language models (LLMs) have recently been used to empower autonomous agents in engineering, significantly improving automation and efficiency in labor-intensive workflows. However, their potential remains underexplored in structural engineering, particularly for finite element modeling tasks requiring geometric modeling, complex reasoning, and domain knowledge. To bridge this gap, this paper develops a LLM-based multi-agent system to automate finite element modeling of 2D frames. The system decomposes structural analysis into subtasks, each managed by a specialized agent powered by the lightweight Llama-3.3 70B Instruct model. The workflow begins with a Problem Analysis Agent, which extracts geometry, boundary, and material parameters from the user input. Next, a Geometry Agent incrementally derives node coordinates and element connectivity by applying expert-defined rules. These structured outputs are converted into executable OpenSeesPy code by a Translation Agent and refined by a Model Validation Agent through consistency checks. Then, a Load Agent applies load conditions into the assembled structural model. Experimental evaluations on 20 benchmark problems demonstrate that the system achieves accuracy over 80% in most cases across 10 repeated trials, outperforming Gemini-2.5 Pro and ChatGPT-4o models.

Method: Uses selective distillation with three trust metrics: Self-Consistency (stability across generations), Class Entailment Alignment (semantic coherence with patent classes), and LLM Agreement Scoring (rationale-label plausibility), integrated into a unified trust score.

Result: Outperforms label-based learning and conventional distillation on USPTO-2M dataset in accuracy, stability, and interpretability.

Conclusion: Establishes a reliable paradigm for leveraging reasoning-aware trust indicators in patent analytics.

Abstract: Large language models (LLMs) increasingly generate natural language rationales to enhance interpretability, but these often contain logical errors, label mismatches, and domain-specific misalignments. Directly using such rationales as supervision risks propagating noise and undermining training stability. To address this challenge, we introduce Self-Filtered Distillation, a framework specifically tailored for patent classification, which treats LLM-generated rationales as trust signals rather than ground-truth supervision. The framework employs selective distillation guided by three unsupervised trust metrics: (1) Self-Consistency, which measures the stability of LLM-generated rationales across multiple generations; (2) Class Entailment Alignment, which assesses semantic coherence with patent-specific class definitions; and (3) LLM Agreement Scoring, which validates rationale-label plausibility. These metrics are integrated into a unified trust score that primarily weights training samples while optionally filtering out extremely low-trust cases, enabling reasoning-aware supervision. Experiments on the USPTO-2M dataset, a widely used benchmark for patent classification, show that our method outperforms label-based learning and conventional distillation in accuracy, stability, and interpretability, establishing a reliable paradigm for leveraging reasoning-aware trust indicators in patent analytics.

Method: Created SimulatorArena benchmark with 909 annotated human-LLM conversations on math tutoring and document creation tasks, evaluating simulators on message behavior matching and rating alignment.

Result: Profile-conditioned simulators achieved Spearman’s ρ of 0.7 on both tasks, closely aligning with human judgments. Used best simulators to benchmark 18 assistants including latest LLMs.

Conclusion: Profile-conditioned simulators provide practical, scalable alternative to human evaluation for assessing LLM performance in interactive applications.

Abstract: Large language models (LLMs) are increasingly used in interactive applications, and human evaluation remains the gold standard for assessing their performance in multi-turn conversations. Since human studies are costly, time-consuming, and hard to reproduce, recent work explores using LLMs to simulate users for automatic assistant evaluation. However, there is no benchmark or systematic study to evaluate whether these simulated users are reliable stand-ins for real users. To address this, we introduce SimulatorArena, a benchmark of 909 annotated human-LLM conversations on two interactive tasks – math tutoring and document creation. SimulatorArena evaluates simulators based on how closely their messages match human behavior and how well their assistant ratings align with human judgments. Experiments on various simulator methods show that simulators conditioned on user profiles, capturing traits like background and message styles, align closely with human judgments. They reach Spearman’s $\rho$ of 0.7 on both tasks, providing a practical, scalable alternative to human evaluation. Using the best simulator for each task, we benchmark 18 assistants, including the latest LLMs such as GPT-5, Claude 4.1 Opus, and Gemini 2.5 Pro.

Method: Convert QA instances into knowledge graphs encoding queries, entities, and agents; train heterogeneous GNN to propagate information across node types and produce task-aware routing distributions; use soft supervision and weighted aggregation of agent outputs.

Result: Extensive experiments show tAgentRouter consistently outperforms single-agent and ensemble baselines, while generalizing across benchmarks and LLM backbones.

Conclusion: Graph-supervised multi-agent routing is effective and robust for question answering, capturing complementary strengths of diverse agents through principled collaboration schemes.

Abstract: Large language models (LLMs) and agent-based frameworks have advanced rapidly, enabling diverse applications. Yet, with the proliferation of models and agentic strategies, practitioners face substantial uncertainty in selecting the best configuration for a downstream task. Prior studies show that different agents and backbones exhibit complementary strengths, and that larger models are not always superior, underscoring the need for adaptive routing mechanisms. Existing approaches to agent routing, however, often emphasize cost efficiency while overlooking the fine-grained contextual and relational structure inherent in QA tasks. In this paper, we propose tAgentRouter, a framework that formulates multi-agent QA as a knowledge-graph-guided routing problem supervised by empirical performance signals. Specifically, we convert QA instance into a knowledge graph that jointly encodes queries, contextual entities, and agents, and then train a heterogeneous graph neural network (GNN) to propagate information across node types and produce task-aware routing distributions over agents. By leveraging soft supervision and weighted aggregation of agent outputs, AgentRouter learns principled collaboration schemes that capture the complementary strengths of diverse agents. Extensive experiments demonstrate that our framework consistently outperforms single-agent and ensemble baselines, while generalizing across benchmarks and LLM backbones. These results highlight the effectiveness and robustness of graph-supervised multi-agent routing for question answering.

Method: Created SoNLI dataset with dialogue transcripts featuring complex social nuances, paired with inferences, likelihood scores, and human explanations. Evaluated models through multi-step counterfactual reasoning.

Result: The paper introduces SoNLI as a benchmark to assess weaknesses in current models’ theory-of-mind abilities regarding social inference.

Conclusion: SoNLI provides a framework to identify and address limitations in models’ social reasoning capabilities, particularly for understanding sarcasm and irony in dialogue.

Abstract: Making theory-of-mind inferences from human dialogue is a strong indicator of a model’s underlying social abilities, which are fundamental for adept AI assistants. However, large language and reasoning models struggle to understand sophisticated social phenomena in transcript data, such as sarcasm and irony. To assess the weaknesses of current models and to identify their solutions, we introduce SocialNLI (SoNLI) – the first social dialogue inference dataset. SoNLI consists of a collection of dialogue transcripts hand-picked to center complex social nuances like irony and sarcasm, paired with inferences, corresponding likelihood scores, and human-written explanations. We explore social inference analysis as a facet of theory-of-mind, and evaluate LLM and reasoning model theory-of-mind ability through multi-step counterfactual reasoning.

Method: Fully vectorized GPU implementation using PyTorch with memory-efficient counting via torch.unique to create compact batch-specific n-gram dictionaries, avoiding traditional hashing-based vectorization memory costs.

Result: Achieved speedups of over 13x on NVIDIA T4 GPUs and exceeding 40x on NVIDIA A100 GPUs, transforming evaluation from a significant bottleneck to negligible training overhead.

Conclusion: TensorBLEU provides a practical solution for large-vocabulary models and accelerates research in RL-based model fine-tuning by making per-sentence BLEU computation efficient on GPUs.

Abstract: Modern natural language processing models have achieved unprecedented scale, yet the tools for their evaluation often remain a computational bottleneck, limiting the pace of research. This is particularly acute for in-training evaluation metrics, such as per-sentence reward signals in Reinforcement Learning, which must operate efficiently on batches of token IDs directly on the GPU. In this paper, we introduce TensorBLEU, a novel implementation of the BLEU metric designed from the ground up for this specific use case. Our approach is fully vectorized for GPU-accelerated, per-sentence computation within PyTorch and introduces a memory-efficient counting mechanism. By creating a compact, batch-specific dictionary of n-grams using \texttt{torch.unique}, our method avoids the prohibitive memory costs of traditional hashing-based vectorization, making it practical for large-vocabulary models. We benchmark TensorBLEU against NLTK, the standard library for token-ID-based BLEU calculation on the CPU. Experiments show that TensorBLEU provides speedups of over 13x on consumer-grade GPUs (NVIDIA T4) and exceeding 40x on data-center-class hardware (NVIDIA A100). This performance transforms a significant bottleneck into a negligible part of the training loop. By clearly defining its role as a “Token-ID BLEU” for development purposes and open-sourcing our implementation, we provide a powerful tool for accelerating research in areas like RL-based model fine-tuning.

Method: Augmented widely used planning benchmarks with manually annotated, fine-grained natural language constraints across four formal categories, tested over 4 LLMs, 3 formal languages, 5 methods, and 4 datasets.

Result: Introduction of constraints consistently halves performance and significantly challenges robustness to problem complexity and lexical shift.

Conclusion: Current LLMs have limited capability in handling constrained planning tasks, highlighting the need for more realistic benchmarks and improved constraint handling methods.

Abstract: LLMs have been widely used in planning, either as planners to generate action sequences end-to-end, or as formalizers to represent the planning domain and problem in a formal language that can derive plans deterministically. However, both lines of work rely on standard benchmarks that only include generic and simplistic environmental specifications, leading to potential overestimation of the planning ability of LLMs and safety concerns in downstream tasks. We bridge this gap by augmenting widely used planning benchmarks with manually annotated, fine-grained, and rich natural language constraints spanning four formally defined categories. Over 4 state-of-the-art reasoning LLMs, 3 formal languages, 5 methods, and 4 datasets, we show that the introduction of constraints not only consistently halves performance, but also significantly challenges robustness to problem complexity and lexical shift.

Method: Multi-level knowledge distillation framework that transfers knowledge from a black box teacher LLM to multiple downstream models (encoder for classification, decoder for explanation) using both data and logit level insights, combined with post-training techniques and prompt engineering.

Result: LANTERN significantly improves task-specific metrics for job person fit and explanation. Online evaluations show 0.24% increase in apply rate and 0.28% increase in qualified applications.

Conclusion: LANTERN successfully addresses the challenges of deploying LLMs for domain-specific tasks by creating efficient, specialized models through knowledge distillation, leading to measurable improvements in real-world job seeking platform performance.

Abstract: Large language models (LLMs) have achieved strong performance across a wide range of natural language processing tasks. However, deploying LLMs at scale for domain specific applications, such as job-person fit and explanation in job seeking platforms, introduces distinct challenges. At LinkedIn, the job person fit task requires analyzing a candidate’s public profile against job requirements to produce both a fit assessment and a detailed explanation. Directly applying open source or finetuned LLMs to this task often fails to yield high quality, actionable feedback due to the complexity of the domain and the need for structured outputs. Moreover, the large size of these models leads to high inference latency and limits scalability, making them unsuitable for online use. To address these challenges, we introduce LANTERN, a novel LLM knowledge distillation framework tailored specifically for job person fit tasks. LANTERN involves modeling over multiple objectives, an encoder model for classification purpose, and a decoder model for explanation purpose. To better distill the knowledge from a strong black box teacher model to multiple downstream models, LANTERN incorporates multi level knowledge distillation that integrates both data and logit level insights. In addition to introducing the knowledge distillation framework, we share our insights on post training techniques and prompt engineering, both of which are crucial for successfully adapting LLMs to domain specific downstream tasks. Extensive experimental results demonstrate that LANTERN significantly improves task specific metrics for both job person fit and explanation. Online evaluations further confirm its effectiveness, showing measurable gains in job seeker engagement, including a 0.24% increase in apply rate and a 0.28% increase in qualified applications.

Method: Create reasoning prototypes by clustering activation differences between Chain-of-Thought and neutral prompts, then project input’s hidden state onto these prototypes to form instance-specific steering vectors during inference.

Result: PDS consistently improves accuracy on GSM8K, AQuA-RAT, and BIG-Bench tasks without fine-tuning or prompt engineering, with gains persisting even when CoT is explicitly suppressed.

Conclusion: Dynamic, prototype-guided steering serves as a lightweight alternative to training-time approaches for enhancing LLM reasoning by strengthening latent reasoning processes rather than inducing superficial behavioral changes.

Abstract: Despite impressive breadth, LLMs still rely on explicit reasoning instructions or static, one-fits-all steering methods, leaving a gap for adaptive, instruction-free reasoning amplification. We present Prototype-Based Dynamic Steering (PDS), a test-time method that amplifies large language model (LLM) reasoning without adding or altering instructions. We introduce “reasoning prototypes” by clustering activation differences between Chain-of-Thought (CoT) and neutral prompts. At inference, an input’s hidden state is projected onto these prototypes to form an instance-specific steering vector. Evaluated on GSM8K, AQuA-RAT, and BIG-Bench tasks, PDS consistently improves accuracy without fine-tuning or prompt engineering. Notably, the gains persist even when CoT is explicitly suppressed to improve cost-efficiency, indicating that the intervention strengthens latent reasoning processes rather than inducing a superficial behavioral shift. These results position dynamic, prototype-guided steering as a lightweight alternative to training-time approaches for enhancing LLM reasoning.

Method: Developed CAM with incremental overlapping clustering algorithm for structured memory development, supporting hierarchical summarization and online batch integration, with adaptive memory exploration during inference.

Result: CAM demonstrates dual advantages in both performance and efficiency across diverse long-text reading comprehension tasks including question answering, query-based summarization, and claim verification.

Conclusion: The constructivist approach provides a systematic blueprint for robust and efficient memory systems in LLM-based reading comprehension, addressing current limitations in handling long-form documents.

Abstract: Current Large Language Models (LLMs) are confronted with overwhelming information volume when comprehending long-form documents. This challenge raises the imperative of a cohesive memory module, which can elevate vanilla LLMs into autonomous reading agents. Despite the emergence of some heuristic approaches, a systematic design principle remains absent. To fill this void, we draw inspiration from Jean Piaget’s Constructivist Theory, illuminating three traits of the agentic memory – structured schemata, flexible assimilation, and dynamic accommodation. This blueprint forges a clear path toward a more robust and efficient memory system for LLM-based reading comprehension. To this end, we develop CAM, a prototype implementation of Constructivist Agentic Memory that simultaneously embodies the structurality, flexibility, and dynamicity. At its core, CAM is endowed with an incremental overlapping clustering algorithm for structured memory development, supporting both coherent hierarchical summarization and online batch integration. During inference, CAM adaptively explores the memory structure to activate query-relevant information for contextual response, akin to the human associative process. Compared to existing approaches, our design demonstrates dual advantages in both performance and efficiency across diverse long-text reading comprehension tasks, including question answering, query-based summarization, and claim verification.

Method: Built structured Knowledge Graph from OMIn dataset and integrated it into retrieval-augmented generation pipeline, evaluated using locally deployable LLMs (Gemma-3, Phi-4, Mistral-Nemo) with stronger models (GPT-4o, Llama-3.3) as judges.

Result: KEO significantly improves global sensemaking by revealing patterns and system-level insights, while text-chunk RAG remains effective for fine-grained procedural tasks requiring localized retrieval.

Conclusion: KG-augmented LLMs show promise for secure, domain-specific QA and have potential in high-stakes reasoning applications.

Abstract: We present Knowledge Extraction on OMIn (KEO), a domain-specific knowledge extraction and reasoning framework with large language models (LLMs) in safety-critical contexts. Using the Operations and Maintenance Intelligence (OMIn) dataset, we construct a QA benchmark spanning global sensemaking and actionable maintenance tasks. KEO builds a structured Knowledge Graph (KG) and integrates it into a retrieval-augmented generation (RAG) pipeline, enabling more coherent, dataset-wide reasoning than traditional text-chunk RAG. We evaluate locally deployable LLMs (Gemma-3, Phi-4, Mistral-Nemo) and employ stronger models (GPT-4o, Llama-3.3) as judges. Experiments show that KEO markedly improves global sensemaking by revealing patterns and system-level insights, while text-chunk RAG remains effective for fine-grained procedural tasks requiring localized retrieval. These findings underscore the promise of KG-augmented LLMs for secure, domain-specific QA and their potential in high-stakes reasoning.

Method: Hybrid One-Bit KV Cache (H1B-KV) represents key vectors using 1-bit binary sketches for hardware-friendly bitwise attention, and compresses value vectors with 4-bit quantization. This holistic approach enables a 7B LLM to handle 8k-token context with under 60MB cache.

Result: After lightweight finetuning, H1B-KV matches full-precision performance on perplexity benchmarks and complex downstream tasks (GSM8K, MMLU, HumanEval). It achieves 70x memory reduction and significantly outperforms leading methods (KIVI, SparseLLM, Loki) in quality-per-byte.

Conclusion: H1B-KV establishes itself as a robust solution for deploying LLMs in memory-constrained environments by providing comprehensive KV cache compression without sacrificing context or performance.

Abstract: Autoregressive decoding in large language models (LLMs) requires caching a growing list of past key-value (KV) pairs, making long-context inference a memory-bound problem. While recent methods have explored quantizing the cache, evicting tokens, or using binary sketches for keys (e.g., Loki), these approaches often provide an incomplete solution by leaving one component (like values) uncompressed or by discarding context information. This paper introduces the Hybrid One-Bit KV Cache (H1B-KV), a comprehensive compression scheme that radically reduces memory usage without sacrificing context. H1B-KV represents each key vector using a 1-bit binary sketch, enabling hardware-friendly bitwise attention, and further compresses value vectors using 4-bit quantization. This holistic, hybrid approach allows a 7-billion parameter LLM to handle an 8k-token context with under 60 MB of cache memory - a 70x reduction. We demonstrate that after a lightweight finetuning, H1B-KV matches full-precision performance not only on perplexity benchmarks but also on complex downstream tasks like mathematical reasoning (GSM8K), multi-task understanding (MMLU), and code generation (HumanEval). Our results show H1B-KV significantly outperforms leading quantization (KIVI), token eviction (SparseLLM), and key-only sketching (Loki) methods in quality-per-byte, establishing it as a robust solution for deploying LLMs in memory-constrained environments.

Method: Used mean reward of N sampled responses as a proxy for prompt difficulty, analyzed performance differences between easy and difficult prompts, and explored selective removal strategies.

Result: Difficult prompts show substantially inferior optimization performance, incorporating them degrades overall performance, and the performance gap closes with increased model capacity. Selective removal of challenging prompts enhances performance.

Conclusion: Prompt difficulty significantly impacts self-play optimization, and strategic prompt selection based on difficulty can improve alignment outcomes for language models.

Abstract: Self-play preference optimization has emerged as a prominent paradigm for aligning large language models (LLMs). It typically involves a language model to generate on-policy responses for prompts and a reward model (RM) to guide the selection of chosen and rejected responses, which can be further trained with direct preference optimization (DPO). However, the role of prompts remains underexplored, despite being a core component in this pipeline. In this work, we investigate how prompts of varying difficulty influence self-play preference optimization. We first use the mean reward of $N$ sampled responses of a prompt as a proxy for its difficulty. We find that difficult prompts exhibit substantially inferior self-play optimization performance in comparison to easy prompts for language models. Moreover, incorporating difficult prompts into training fails to enhance overall performance and, in fact, leads to slight degradation compared to training on easy prompts alone. We also observe that the performance gap between difficult and easy prompts closes as the model capacity increases, suggesting that difficulty interacts with the model capacity. Building on these findings, we explore strategies to mitigate the negative effect of difficult prompts on final performance. We demonstrate that selectively removing an appropriate portion of challenging prompts enhances overall self-play performance, while also reporting failed attempts and lessons learned.

Method: Pareto-Guided Singular Value Decomposition (PGSVD) - a zero-shot pipeline that improves activation-aware compression via Pareto-guided rank selection and alternating least-squares implementation.

Result: PGSVD applied to both LLM and VLM shows better accuracy at the same compression levels and inference speedup.

Conclusion: The proposed PGSVD framework effectively addresses the memory and computing challenges of LLMs and VLMs while maintaining performance.

Abstract: Large language models (LLM) and vision-language models (VLM) have achieved state-of-the-art performance, but they impose significant memory and computing challenges in deployment. We present a novel low-rank compression framework to address this challenge. First, we upper bound the change of network loss via layer-wise activation-based compression errors, filling a theoretical gap in the literature. We then formulate low-rank model compression as a bi-objective optimization and prove that a single uniform tolerance yields surrogate Pareto-optimal heterogeneous ranks. Based on our theoretical insights, we propose Pareto-Guided Singular Value Decomposition (PGSVD), a zero-shot pipeline that improves activation-aware compression via Pareto-guided rank selection and alternating least-squares implementation. We apply PGSVD to both LLM and VLM, showing better accuracy at the same compression levels and inference speedup.

Method: Introduces EvoPresent framework with PresAesth - a multi-task reinforcement learning aesthetic model that provides aesthetic scoring, defect adjustment, and comparative feedback for iterative self-improvement. Also creates EvoPresent Benchmark with 650 AI conference papers and 2,000 slide pairs for evaluation.

Result: Findings show: (i) High-quality feedback is essential for agent self-improvement, (ii) Automated generation pipelines exhibit trade-off between visual design and content construction, (iii) Multi-task RL training shows stronger generalization in aesthetic awareness tasks.

Conclusion: EvoPresent successfully addresses the core challenge of academic presentation generation by providing reliable aesthetic evaluation and feedback mechanisms, enabling self-improvement even with limited training data.

Abstract: The promotion of academic papers has become an important means of enhancing research visibility. However, existing automated methods struggle limited storytelling, insufficient aesthetic quality, and constrained self-adjustment, making it difficult to achieve efficient and engaging dissemination. At the heart of those challenges is a simple principle: \emph{there is no way to improve it when you cannot evaluate it right}. To address this, we introduce \textbf{EvoPresent}, a self-improvement agent framework that unifies coherent narratives, aesthetic-aware designs, and realistic presentation delivery via virtual characters. Central to EvoPresent is \textbf{PresAesth}, a multi-task reinforcement learning (RL) aesthetic model that provides reliable aesthetic scoring, defect adjustment, and comparative feedback, enabling iterative self-improvement even under limited aesthetic training data. To systematically evaluate the methods, we introduce \textbf{EvoPresent Benchmark}, a comprehensive benchmark comprising: \textit{Presentation Generation Quality}, built on 650 top-tier AI conference papers with multimodal resources (slides, videos and scripts) to assess both content and design; and \textit{Aesthetic Awareness}, consisting of 2,000 slide pairs with varying aesthetic levels, supporting joint training and evaluation on scoring, defect adjustment, and comparison. Our findings highlight that (i) High-quality feedback is essential for agent self-improvement, while initial capability alone does not guarantee effective self-correction. (ii) Automated generation pipelines exhibit a trade-off between visual design and content construction. (iii) Multi-task RL training shows stronger generalization in aesthetic awareness tasks.

Method: Identifies key entities as initial nodes, generates reasoning child nodes, and refines process through historical error analysis and real-time feedback using depth-first search with node generation.

Result: Achieved 54.47% F1 score on HotpotQA and 70.05% on StrategyQA, surpassing best baselines by 5.03% and 7.25% respectively.

Conclusion: FGDIP demonstrates versatility and potential to enhance language agents in multi-hop reasoning tasks through dynamic adaptive strategies.

Abstract: Recent advancements in language agents have led to significant improvements in multi-hop reasoning tasks. However, existing approaches often struggle with handling open-domain problems, which require massive information retrieval due to their reliance on a fixed sequence of actions. To address this, we propose Feedback-Guided Dynamic Interactive Planning (FGDIP), a novel framework tailored to enhance reasoning in LLMs by utilizing dynamic and adaptive strategies for information exploration in open-domain multi-hop reasoning tasks. Our approach begins by identifying key entities relevant to the problem, which serve as the initial nodes in the reasoning process. From these initial nodes, we then generate reasoning child nodes with the process being refined through a combination of historical error analysis and real-time feedback, which allows the framework to dynamically adjust and optimize its reasoning strategies. By integrating depth-first search with an innovative node generation technique, our framework adapts based on both prior error paths and concurrently generated nodes at the same hierarchical level. This dynamic strategy effectively expands the search space while ensuring the reasoning process systematically converges toward accurate solutions. Experimental results show that FGDIP achieved up to 54.47% F1 score on the HotpotQA dataset and 70.05% on the StrategyQA dataset, surpassing the best baseline by 5.03% and 7.25% respectively, highlighting its versatility and potential to enhance language agents in multi-hop reasoning tasks.

Method: Two-step training process: 1) Synthesize plans from advanced LLMs using homologous consensus filtering and fine-tune, 2) Rule-based reinforcement learning with executor capability gain reward.

Result: Achieves state-of-the-art performance on three long-horizon agent tasks, reduces training costs by 8x compared to RL baselines, and requires no manual effort or extra training data.

Conclusion: EAGLET provides an efficient and effective solution for enhancing LLM agents’ planning capabilities in long-horizon tasks.

Abstract: Agents based on large language models (LLMs) struggle with brainless trial-and-error and generating hallucinatory actions due to a lack of global planning in long-horizon tasks. In this paper, we introduce a plan-and-execute framework and propose EAGLET, an efficient and effective planner training method to enhance the executor agent’s planning abilities without human effort. Specifically, we train a plug-and-play global planner through a two-step process: we first synthesize high-quality plans from an advanced LLM using our proposed homologous consensus filtering strategy, and apply fine-tuning as a cold start. Moreover, we further improve the planner with a rule-based reinforcement learning stage using a novel executor capability gain reward, ensuring it can handle task instructions of varying difficulty. Experiments on three long-horizon agent tasks show that executor agents equipped with our planner outperform existing methods, achieving new state-of-the-art performance. Meanwhile, EAGLET reduces training costs by 8x compared to RL-based baselines, and it does not require manual effort or extra training data, offering an efficient and effective solution.

Method: Proposes a multi-agent debate framework where multiple MLLM agents iteratively debate, verify, and update each other’s responses through multiple rounds to refine attribute value inferences.

Result: Experiments on ImplicitAVE dataset show significant accuracy improvements with just a few debate rounds, particularly for attributes with initially low performance. Various debate configurations were evaluated.

Conclusion: Multi-agent debate strategies effectively address limitations of single-agent approaches and provide a scalable solution for implicit AVE in multimodal e-commerce applications.

Abstract: Implicit Attribute Value Extraction (AVE) is essential for accurately representing products in e-commerce, as it infers lantent attributes from multimodal data. Despite advances in multimodal large language models (MLLMs), implicit AVE remains challenging due to the complexity of multidimensional data and gaps in vision-text understanding. In this work, we introduce \textsc{\modelname}, a multi-agent debate framework that employs multiple MLLM agents to iteratively refine inferences. Through a series of debate rounds, agents verify and update each other’s responses, thereby improving inference performance and robustness. Experiments on the ImplicitAVE dataset demonstrate that even a few rounds of debate significantly boost accuracy, especially for attributes with initially low performance. We systematically evaluate various debate configurations, including identical or different MLLM agents, and analyze how debate rounds affect convergence dynamics. Our findings highlight the potential of multi-agent debate strategies to address the limitations of single-agent approaches and offer a scalable solution for implicit AVE in multimodal e-commerce.

Method: CoSpaDi uses structured sparse factorization with a dense dictionary and column-sparse coefficient matrix, forming a union-of-subspaces representation. It leverages calibration data to minimize functional reconstruction error rather than weight approximation error.

Result: CoSpaDi consistently outperforms state-of-the-art low-rank methods across Llama and Qwen models at 20-50% compression ratios, achieving better accuracy and perplexity without fine-tuning.

Conclusion: Structured sparse dictionary learning is a powerful alternative to low-rank approaches for LLM compression, offering greater expressiveness and better model fidelity while enabling efficient deployment through sparse matrix operations and quantization compatibility.

Abstract: Post-training compression of large language models (LLMs) largely relies on low-rank weight approximation, which represents each column of a weight matrix in a shared low-dimensional subspace. While this is a computationally efficient strategy, the imposed structural constraint is rigid and can lead to a noticeable model accuracy drop. In this work, we propose CoSpaDi (Compression via Sparse Dictionary Learning), a novel training-free compression framework that replaces low-rank decomposition with a more flexible structured sparse factorization in which each weight matrix is represented with a dense dictionary and a column-sparse coefficient matrix. This formulation enables a union-of-subspaces representation: different columns of the original weight matrix are approximated in distinct subspaces spanned by adaptively selected dictionary atoms, offering greater expressiveness than a single invariant basis. Crucially, CoSpaDi leverages a small calibration dataset to optimize the factorization such that the output activations of compressed projection layers closely match those of the original ones, thereby minimizing functional reconstruction error rather than mere weight approximation. This data-aware strategy preserves better model fidelity without any fine-tuning under reasonable compression ratios. Moreover, the resulting structured sparsity allows efficient sparse-dense matrix multiplication and is compatible with post-training quantization for further memory and latency gains. We evaluate CoSpaDi across multiple Llama and Qwen models under per-layer and per-group settings at 20-50% compression ratios, demonstrating consistent superiority over state-of-the-art data-aware low-rank methods both in accuracy and perplexity. Our results establish structured sparse dictionary learning as a powerful alternative to conventional low-rank approaches for efficient LLM deployment.

Method: Systematic data collection pipeline yielding largest validated African multi-modal dataset (19B tokens text + 12,628h speech), fine-tuning models, and structured research program for capacity building.

Result: Substantial improvements over baselines: +23.69 ChrF++, +0.33 COMET, +15.34 BLEU points across 31 languages; competitive performance against Google Translate; mentored 15 early-career researchers.

Conclusion: The initiative successfully addresses the African language technology gap through comprehensive data collection, model development, and sustainable capacity building, though continued development is needed in some areas.

Abstract: Despite representing nearly one-third of the world’s languages, African languages remain critically underserved by modern NLP technologies, with 88% classified as severely underrepresented or completely ignored in computational linguistics. We present the African Languages Lab (All Lab), a comprehensive research initiative that addresses this technological gap through systematic data collection, model development, and capacity building. Our contributions include: (1) a quality-controlled data collection pipeline, yielding the largest validated African multi-modal speech and text dataset spanning 40 languages with 19 billion tokens of monolingual text and 12,628 hours of aligned speech data; (2) extensive experimental validation demonstrating that our dataset, combined with fine-tuning, achieves substantial improvements over baseline models, averaging +23.69 ChrF++, +0.33 COMET, and +15.34 BLEU points across 31 evaluated languages; and (3) a structured research program that has successfully mentored fifteen early-career researchers, establishing sustainable local capacity. Our comparative evaluation against Google Translate reveals competitive performance in several languages while identifying areas that require continued development.

Method: Progressive code-switching from target language to English within demonstrations and instructions to scaffold reasoning process and facilitate latent reasoning in English.

Result: Consistent outperformance of X-ICL baselines with 3.1%p gains in target languages and 1.9%p in unseen languages, with even larger improvements (14.7% and 5.3%) in low-resource settings across 4 LLMs, 6 datasets, and 10 languages.

Conclusion: Code-switching serves as a principled and robust approach to overcome translation barriers during inference, moving LLMs toward more equitable and effective multilingual systems.

Abstract: While large language models (LLMs) exhibit strong multilingual abilities, their reliance on English as latent representations creates a translation barrier, where reasoning implicitly depends on internal translation into English. When this process fails, performance in non-English languages deteriorates sharply, limiting the inclusiveness of LLM-based applications. Existing cross-lingual in-context learning (X-ICL) methods primarily leverage monolingual demonstrations, often failing to mitigate this barrier and instead reinforcing it. In this work, we introduce code-switching in-context learning (CSICL), a simple yet effective prompting strategy that progressively transitions from a target language to English within demonstrations and instruction to facilitate their latent reasoning in English. By explicitly scaffolding the reasoning process through controlled code-switching, CSICL acts as an implicit linguistic bridge that enhances cross-lingual alignment and reduces reliance on the translation barrier. We conduct extensive experiments across 4 LLMs, 6 datasets, and 10 languages, spanning both knowledge-intensive and reasoning-oriented domains. Our results demonstrate that CSICL consistently outperforms X-ICL baselines, achieving gains of 3.1%p and 1.9%p in both target and unseen languages, respectively. The improvement is even more pronounced in low-resource settings, with gains of 14.7% in target and 5.3% in unseen languages. These findings establish code-switching as a principled and robust approach for overcoming the translation barrier during inference, moving LLMs toward more equitable and effective multilingual systems.

Method: Models RAG as a Markov Decision Process (MDP) with decision-making and execution components, introduces efficient pruning strategy for data expansion optimization, and uses comprehensive process-level policy optimization.

Result: Achieves 6.2% average absolute performance improvement across six datasets, significantly outperforms existing baselines, and pruning strategy improves data construction efficiency by nearly 6x.

Conclusion: DecEx-RAG provides an efficient solution for process-supervised RAG training, enhancing autonomous task decomposition, dynamic retrieval, and high-quality answer generation capabilities of LLMs.

Abstract: Agentic Retrieval-Augmented Generation (Agentic RAG) enhances the processing capability for complex tasks through dynamic retrieval and adaptive workflows. Recent advances (e.g., Search-R1) have shown that outcome-supervised reinforcement learning demonstrate strong performance. However, this approach still suffers from inefficient exploration, sparse reward signals, and ambiguous global reward feedback. To address these challenges, we propose DecEx-RAG, which models RAG as a Markov Decision Process (MDP) incorporating decision-making and execution, while introducing an efficient pruning strategy to optimize data expansion. Through comprehensive process-level policy optimization, DecEx-RAG significantly enhances the autonomous task decomposition, dynamic retrieval, and high-quality answer generation capabilities of large language models (LLMs). Experiments show that DecEx-RAG achieves an average absolute performance improvement of $6.2%$ across six datasets, significantly outperforming existing baselines. Moreover, the pruning strategy improves data construction efficiency by nearly $6 \times$, providing an efficient solution for process-supervised RAG training. The code is available at https://github.com/sdsxdxl/DecEx-RAG.

Method: Compute similarity scores using adaptable criteria and NLP techniques (Bag-of-Words, sequential, surface approaches) on entities from eight semantic artifacts including Wikidata, then use LLM to validate mapping suggestions.

Result: Multi-source synonym sets with standardized labels per entity, ready for integration into IVOA Vocabularies and OntoPortal-Astro platform.

Conclusion: The approach enables creation of FAIR-compliant mappings that will enhance astronomical data interoperability through standardized entity resolution.

Abstract: This ongoing work focuses on the development of a methodology for generating a multi-source mapping of astronomical observation facilities. To compare two entities, we compute scores with adaptable criteria and Natural Language Processing (NLP) techniques (Bag-of-Words approaches, sequential approaches, and surface approaches) to map entities extracted from eight semantic artifacts, including Wikidata and astronomy-oriented resources. We utilize every property available, such as labels, definitions, descriptions, external identifiers, and more domain-specific properties, such as the observation wavebands, spacecraft launch dates, funding agencies, etc. Finally, we use a Large Language Model (LLM) to accept or reject a mapping suggestion and provide a justification, ensuring the plausibility and FAIRness of the validated synonym pairs. The resulting mapping is composed of multi-source synonym sets providing only one standardized label per entity. Those mappings will be used to feed our Name Resolver API and will be integrated into the International Virtual Observatory Alliance (IVOA) Vocabularies and the OntoPortal-Astro platform.

Method: Derived spectral bounds for InfoNCE gradients, used effective rank as anisotropy proxy, designed spectrum-aware batch selection with greedy builder, and implemented in-batch whitening.

Result: Greedy-64 reduced time-to-67.5% top-1 accuracy by 15% vs random on ImageNet-100, with similar gains on CIFAR-10. In-batch whitening reduced 50-step gradient variance by 1.37×.

Conclusion: Spectrum-aware batch selection and whitening effectively accelerate contrastive learning by optimizing gradient properties, with theoretical bounds matching empirical results.

Abstract: We derive non-asymptotic spectral bands that bound the squared InfoNCE gradient norm via alignment, temperature, and batch spectrum, recovering the (1/\tau^{2}) law and closely tracking batch-mean gradients on synthetic data and ImageNet. Using effective rank (R_{\mathrm{eff}}) as an anisotropy proxy, we design spectrum-aware batch selection, including a fast greedy builder. On ImageNet-100, Greedy-64 cuts time-to-67.5% top-1 by 15% vs.\ random (24% vs.\ Pool–P3) at equal accuracy; CIFAR-10 shows similar gains. In-batch whitening promotes isotropy and reduces 50-step gradient variance by (1.37\times), matching our theoretical upper bound.

Method: Two methods: optimal transport-based informative attention to improve learning focal information, and accumulative joint entropy reduction on named entities to enhance informative salience.

Result: Achieves better ROUGE scores on CNN/Daily Mail dataset and competitive results on XSum. Human evaluation confirms better informativeness performance over strong baseline.

Conclusion: The proposed approach effectively improves informativeness in abstractive text summarization, with analysis providing insights into the evaluation results.

Abstract: Abstractive text summarization is integral to the Big Data era, which demands advanced methods to turn voluminous and often long text data into concise but coherent and informative summaries for efficient human consumption. Despite significant progress, there is still room for improvement in various aspects. One such aspect is to improve informativeness. Hence, this paper proposes a novel learning approach consisting of two methods: an optimal transport-based informative attention method to improve learning focal information in reference summaries and an accumulative joint entropy reduction method on named entities to enhance informative salience. Experiment results show that our approach achieves better ROUGE scores compared to prior work on CNN/Daily Mail while having competitive results on XSum. Human evaluation of informativeness also demonstrates the better performance of our approach over a strong baseline. Further analysis gives insight into the plausible reasons underlying the evaluation results.

Method: Proposed MoNE which performs neuron-granular expert selection using simple top-k selection within each expert, requiring no additional routing parameters or inter-expert communication.

Result: MoNE matches traditional MoE performance while activating only 50% of MoE-layer parameters, and consistently outperforms traditional MoE at equal activated parameter counts.

Conclusion: MoNE is a practical approach for improving parameter utilization and inference efficiency in MoE-like models with negligible latency overhead.

Abstract: In this work, we first explore whether the parameters activated by the MoE layer remain highly sparse at inference. We perform a sparsification study on several representative MoE models. For each expert, we rank parameters by the magnitude of their activations from the gate projection and progressively prune the activated subset. Pruning up to 60% of parameters within that subset causes only negligible task-performance degradation; substantial drops occur only after more than 90% are removed. We further decompose experts into neuron-granular MoE and visualize their activation values, finding that most neuron activations are near zero. This observation motivates us to select only high-activation neuron experts during pretraining. Based on this insight, we propose Mixture of Neuron Experts (MoNE). MoNE achieves neuron-granular expert selection by only applying a simple top-k selection within each expert, incurs negligible latency, and requires no additional routing parameters or inter-expert communication. Extensive experiments demonstrate that MoNE matches traditional MoE performance while activating only 50% of the MoE-layer parameters, and it consistently outperforms traditional MoE when compared at equal numbers of activated parameters. These results suggest that MoNE is a practical approach to improving parameter utilization and inference efficiency in MoE-like models.

Method: Two-stage rollouts with adaptive unlearning: first stage generates half trajectories, then lightweight unlearning suppresses sampled responses to force second stage to explore different output regions.

Result: Outperforms GRPO with average relative gains of 24.3% on Qwen2.5-3B, 33.0% on Llama3.2-3B-Instruct, and 10.4% on Qwen3-8B-Base across five reasoning benchmarks.

Conclusion: EEPO’s sample-then-forget mechanism effectively disrupts the self-reinforcing loop and promotes wider exploration, addressing the exploration-exploitation challenge in RLVR for LLMs.

Abstract: Balancing exploration and exploitation remains a central challenge in reinforcement learning with verifiable rewards (RLVR) for large language models (LLMs). Current RLVR methods often overemphasize exploitation, leading to entropy collapse, diminished exploratory capacity, and ultimately limited performance gains. Although techniques that increase policy stochasticity can promote exploration, they frequently fail to escape dominant behavioral modes. This creates a self-reinforcing loop-repeatedly sampling and rewarding dominant modes-that further erodes exploration. We introduce Exploration-Enhanced Policy Optimization (EEPO), a framework that promotes exploration via two-stage rollouts with adaptive unlearning. In the first stage, the model generates half of the trajectories; it then undergoes a lightweight unlearning step to temporarily suppress these sampled responses, forcing the second stage to explore different regions of the output space. This sample-then-forget mechanism disrupts the self-reinforcing loop and promotes wider exploration during rollouts. Across five reasoning benchmarks, EEPO outperforms GRPO, achieving average relative gains of 24.3% on Qwen2.5-3B, 33.0% on Llama3.2-3B-Instruct, and 10.4% on Qwen3-8B-Base.

Method: Targeted post-training on curated, high-quality French data and strategic model merging to enhance performance in both French and English.

Result: Luth models outperform all open-source counterparts of comparable size on multiple French benchmarks while retaining their original English capabilities. Strategic model merging further enhances performance in both languages.

Conclusion: Luth establishes a new state of the art for French SLMs and provides a robust baseline for future French-language research, demonstrating that targeted adaptation can effectively address language performance gaps.

Abstract: The landscape of Large Language Models (LLMs) remains predominantly English-centric, resulting in a significant performance gap for other major languages, such as French, especially in the context of Small Language Models (SLMs). Existing multilingual models demonstrate considerably lower performance in French compared to English, and research on efficient adaptation methods for French remains limited. To address this, we introduce \textbf{Luth}, a family of French-specialized SLMs: through targeted post-training on curated, high-quality French data, our models outperform all open-source counterparts of comparable size on multiple French benchmarks while retaining their original English capabilities. We further show that strategic model merging enhances performance in both languages, establishing Luth as a new state of the art for French SLMs and a robust baseline for future French-language research.

Method: Use continual pre-training with large-scale unlabeled business conversation data to adapt LLMs for downstream summarization tasks.

Result: Substantial gains in both in-domain and out-of-domain summarization benchmarks, with strong generalization and robustness maintained.

Conclusion: Continual pre-training is an effective approach for adapting LLMs to conversational summarization, with practical guidelines provided for industrial applications.

Abstract: Large language models (LLMs) have achieved impressive performance in text summarization, yet their performance often falls short when applied to specialized domains %or conversational data that differ from their original pre-training distribution. While fine-tuning can improve summarization quality, it typically relies on costly and scarce high-quality labeled data. In this work, we explore continual pre-training as a scalable, self-supervised approach to adapt LLMs for downstream summarization tasks, particularly in the context of noisy real-world conversation transcripts. We conduct extensive experiments using large-scale, unlabeled business conversation data to investigate whether continual pre-training enhances model capabilities in conversational summarization. Our results demonstrate that continual pre-training yields substantial gains in both in-domain and out-of-domain summarization benchmarks, while maintaining strong generalization and robustness. We also analyze the effects of data selection strategies, providing practical guidelines for applying continual pre-training in summarization-focused industrial applications.

Method: Created and annotated a multilingual benchmark dataset capturing Swiss and EU privacy law obligations. Used a novel GPT-5-based method for large-scale compliance assessment of privacy policies to measure the impact of the 2023 Swiss privacy law revision.

Result: 18% of local websites explicitly referenced generators. Generator use was associated with substantially higher compliance levels - up to 15 percentage points higher than policies without generator use. Compliance increases were observed indicating the revision had an effect.

Conclusion: The findings contribute to debates on LLMs for cross-lingual legal analysis, the Brussels Effect of EU regulations, and the role of automated tools in improving compliance and contractual quality.

Abstract: It has become increasingly challenging for firms to comply with a plethora of novel digital regulations. This is especially true for smaller businesses that often lack both the resources and know-how to draft complex legal documents. Instead of seeking costly legal advice from attorneys, firms may turn to cheaper alternative legal service providers such as automated contract generators. While these services have a long-standing presence, there is little empirical evidence on their prevalence and output quality. We address this gap in the context of a 2023 Swiss privacy law revision. To enable a systematic evaluation, we create and annotate a multilingual benchmark dataset that captures key compliance obligations under Swiss and EU privacy law. Using this dataset, we validate a novel GPT-5-based method for large-scale compliance assessment of privacy policies, allowing us to measure the impact of the revision. We observe compliance increases indicating an effect of the revision. Generators, explicitly referenced by 18% of local websites, are associated with substantially higher levels of compliance, with increases of up to 15 percentage points compared to privacy policies without generator use. These findings contribute to three debates: the potential of LLMs for cross-lingual legal analysis, the Brussels Effect of EU regulations, and, crucially, the role of automated tools in improving compliance and contractual quality.

Method: Proposes Integrated Gradient (IG) score to detect context noise, and Context Denoising Training (CDT) strategy to improve attention on critical tokens and reinforce their influence on predictions.

Result: CDT substantially boosts model attention on critical tokens and improves predictions. An 8B model trained with CDT achieves performance (50.92) comparable to GPT-4o (51.00) across four tasks.

Conclusion: Context Denoising Training is an effective strategy that enhances long-context model performance by mitigating contextual noise and improving attention mechanisms.

Abstract: Long-context models (LCMs) have demonstrated great potential in processing long sequences, facilitating many real-world applications. The success of LCMs can be attributed to their ability to locate implicit critical information within the context for further prediction. However, recent research reveals that LCMs are often susceptible to contextual noise, i.e., irrelevant tokens, that can mislead model attention. In this paper, we conduct a fine-grained analysis of the context noise and propose an effective metric, the Integrated Gradient (IG) score, to detect and quantify the noise information within the context. Our findings reveal that even simple mitigation of detected context noise can substantially boost the model’s attention on critical tokens and benefit subsequent predictions. Building on this insight, we propose Context Denoising Training (CDT), a straightforward yet effective training strategy that improves attention on critical tokens while reinforcing their influence on model predictions. Extensive experiments across four tasks, under both context window scaling and long-context alignment settings, demonstrate the superiority of CDT. Notably, when trained with CDT, an open-source 8B model can achieve performance (50.92) comparable to GPT-4o (51.00).

Method: Evaluated LLMs’ sensitivity to harmful content under extended context by varying type (explicit vs implicit), position (beginning, middle, end), prevalence (0.01-0.50 of prompt), and context length (600-6000 tokens) across harmful content categories using LLaMA-3, Qwen-2.5, and Mistral models.

Result: Across all models and harmful content categories, similar patterns emerged: performance peaks at moderate harmful prevalence (0.25) but declines when content is very sparse or dominant; recall decreases with increasing context length; harmful sentences at the beginning are detected more reliably; and explicit content is more consistently recognized than implicit.

Conclusion: This provides the first systematic view of how LLMs prioritize and calibrate harmful content in long contexts, highlighting both emerging strengths and remaining challenges for safety-critical use.

Abstract: Large language models (LLMs) increasingly support applications that rely on extended context, from document processing to retrieval-augmented generation. While their long-context capabilities are well studied for reasoning and retrieval, little is known about their behavior in safety-critical scenarios. We evaluate LLMs’ sensitivity to harmful content under extended context, varying type (explicit vs. implicit), position (beginning, middle, end), prevalence (0.01-0.50 of the prompt), and context length (600-6000 tokens). Across harmful content categories such as toxic, offensive, and hate speech, with LLaMA-3, Qwen-2.5, and Mistral, we observe similar patterns: performance peaks at moderate harmful prevalence (0.25) but declines when content is very sparse or dominant; recall decreases with increasing context length; harmful sentences at the beginning are generally detected more reliably; and explicit content is more consistently recognized than implicit. These findings provide the first systematic view of how LLMs prioritize and calibrate harmful content in long contexts, highlighting both their emerging strengths and the challenges that remain for safety-critical use.

Method: Conducted targeted replications using a broader set of LLMs and more test items to test the robustness of LSZ’s findings.

Result: Found that prompt language has minimal effect on outputs, contradicting LSZ’s claims about substantial cultural shifts.

Conclusion: The reported cultural tendencies are not stable traits but fragile artifacts of specific models and task design, challenging the notion that LLMs encode grounded cultural beliefs.

Abstract: In a recent study, Lu, Song, and Zhang (2025) (LSZ) propose that large language models (LLMs), when prompted in different languages, display culturally specific tendencies. They report that the two models (i.e., GPT and ERNIE) respond in more interdependent and holistic ways when prompted in Chinese, and more independent and analytic ways when prompted in English. LSZ attribute these differences to deep-seated cultural patterns in the models, claiming that prompt language alone can induce substantial cultural shifts. While we acknowledge the empirical patterns they observed, we find their experiments, methods, and interpretations problematic. In this paper, we critically re-evaluate the methodology, theoretical framing, and conclusions of LSZ. We argue that the reported “cultural tendencies” are not stable traits but fragile artifacts of specific models and task design. To test this, we conducted targeted replications using a broader set of LLMs and a larger number of test items. Our results show that prompt language has minimal effect on outputs, challenging LSZ’s claim that these models encode grounded cultural beliefs.

Method: Propose a prompt optimization framework that modifies task instruction prompts using turn-by-turn feedback and experience replay for prompt rewriting, enabling long-term planning without parameter updates.

Result: Significant improvement in multi-turn tasks like text-to-SQL and task-oriented dialogue, with generalization across different LLM-based agents and ability to leverage diverse LLMs as meta-prompting agents.

Conclusion: The method warrants future research in reinforcement learning-inspired parameter-free optimization methods for LLMs.

Abstract: Large language models (LLMs) have achieved remarkable success in a wide range of natural language processing tasks and can be adapted through prompting. However, they remain suboptimal in multi-turn interactions, often relying on incorrect early assumptions and failing to track user goals over time, which makes such tasks particularly challenging. Prior works in dialogue systems have shown that long-term planning is essential for handling interactive tasks. In this work, we propose a prompt optimisation framework inspired by reinforcement learning, which enables such planning to take place by only modifying the task instruction prompt of the LLM-based agent. By generating turn-by-turn feedback and leveraging experience replay for prompt rewriting, our proposed method shows significant improvement in multi-turn tasks such as text-to-SQL and task-oriented dialogue. Moreover, it generalises across different LLM-based agents and can leverage diverse LLMs as meta-prompting agents. This warrants future research in reinforcement learning-inspired parameter-free optimisation methods.

Method: Introduced a four-part framework to categorize how benchmarks frame culture (knowledge, preference, performance, bias), qualitatively examined 20 cultural benchmarks, and identified six recurring methodological issues.

Result: Identified six methodological issues including treating countries as cultures, overlooking within-culture diversity, and relying on oversimplified survey formats. Proposed concrete improvements based on anthropological methods.

Conclusion: Cultural benchmarks should move beyond static recall tasks and incorporate real-world narratives, involve cultural communities in design, and evaluate models in context to better capture complex cultural situations.

Abstract: Cultural evaluation of large language models has become increasingly important, yet current benchmarks often reduce culture to static facts or homogeneous values. This view conflicts with anthropological accounts that emphasize culture as dynamic, historically situated, and enacted in practice. To analyze this gap, we introduce a four-part framework that categorizes how benchmarks frame culture, such as knowledge, preference, performance, or bias. Using this lens, we qualitatively examine 20 cultural benchmarks and identify six recurring methodological issues, including treating countries as cultures, overlooking within-culture diversity, and relying on oversimplified survey formats. Drawing on established anthropological methods, we propose concrete improvements: incorporating real-world narratives and scenarios, involving cultural communities in design and validation, and evaluating models in context rather than isolation. Our aim is to guide the development of cultural benchmarks that go beyond static recall tasks and more accurately capture the responses of the models to complex cultural situations.

Method: Uses chain-of-thought with two scoring methods (log-probabilities and direct ratings) plus model-as-judge peer review on World Values Survey (55 countries, 19 topics) and PEW Global Attitudes Survey (39 countries, 8 topics).

Result: Top models align closely with survey responses (Pearson’s r≈0.90 on WVS), but show significant regional bias: Western regions average r=0.82 vs non-Western regions r=0.61 (0.21 gap). Peer review flagged 348 conflicts and peer agreement correlates with survey alignment.

Conclusion: Shows progress toward culture-aware AI but highlights significant regional biases that remain as open challenges for cross-regional deployment.

Abstract: We present EvalMORAAL, a transparent chain-of-thought (CoT) framework that uses two scoring methods (log-probabilities and direct ratings) plus a model-as-judge peer review to evaluate moral alignment in 20 large language models. We assess models on the World Values Survey (55 countries, 19 topics) and the PEW Global Attitudes Survey (39 countries, 8 topics). With EvalMORAAL, top models align closely with survey responses (Pearson’s r approximately 0.90 on WVS). Yet we find a clear regional difference: Western regions average r=0.82 while non-Western regions average r=0.61 (a 0.21 absolute gap), indicating consistent regional bias. Our framework adds three parts: (1) two scoring methods for all models to enable fair comparison, (2) a structured chain-of-thought protocol with self-consistency checks, and (3) a model-as-judge peer review that flags 348 conflicts using a data-driven threshold. Peer agreement relates to survey alignment (WVS r=0.74, PEW r=0.39, both p<.001), supporting automated quality checks. These results show real progress toward culture-aware AI while highlighting open challenges for use across regions.

Method: Used linear probes on final-token representations of LLMs and identified specific attention heads in the final Transformer layer that show opposite activation patterns for simple and difficult problems.

Result: Problem difficulty can be linearly modeled from LLM representations, and specific attention heads achieve difficulty perception through opposite activation patterns. LLMs can serve as automatic difficulty annotators.

Conclusion: Difficulty perception in LLMs is structurally organized and present, offering practical applications for benchmark construction and curriculum learning while reducing reliance on human labeling.

Abstract: Large language models (LLMs) are increasingly deployed on complex reasoning tasks, yet little is known about their ability to internally evaluate problem difficulty, which is an essential capability for adaptive reasoning and efficient resource allocation. In this work, we investigate whether LLMs implicitly encode problem difficulty in their internal representations. Using a linear probe on the final-token representations of LLMs, we demonstrate that the difficulty level of math problems can be linearly modeled. We further locate the specific attention heads of the final Transformer layer: these attention heads have opposite activation patterns for simple and difficult problems, thus achieving perception of difficulty. Our ablation experiments prove the accuracy of the location. Crucially, our experiments provide practical support for using LLMs as automatic difficulty annotators, potentially substantially reducing reliance on costly human labeling in benchmark construction and curriculum learning. We also uncover that there is a significant difference in entropy and difficulty perception at the token level. Our study reveals that difficulty perception in LLMs is not only present but also structurally organized, offering new theoretical insights and practical directions for future research.

Method: Take existing planning environments, impose temporal constraints on states, translate constrained problems into natural language, and test LLMs on solving them. The benchmark is extensible to new environments.

Result: Performance of state-of-the-art LLMs (GPT-5, o3, R1) deteriorates as the degree of constrainedness in planning tasks increases.

Conclusion: LLMs struggle with constrained planning tasks, highlighting the need for specialized benchmarks like LexiCon to evaluate their real-world applicability.

Abstract: Owing to their reasoning capabilities, large language models (LLMs) have been evaluated on planning tasks described in natural language. However, LLMs have largely been tested on planning domains without constraints. In order to deploy them in real-world settings where adherence to constraints, in particular safety constraints, is critical, we need to evaluate their performance on constrained planning tasks. We introduce LexiCon – a natural language-based (Lexi) constrained (Con) planning benchmark, consisting of a suite of environments, that can be used to evaluate the planning capabilities of LLMs in a principled fashion. The core idea behind LexiCon is to take existing planning environments and impose temporal constraints on the states. These constrained problems are then translated into natural language and given to an LLM to solve. A key feature of LexiCon is its extensibility. That is, the set of supported environments can be extended with new (unconstrained) environment generators, for which temporal constraints are constructed automatically. This renders LexiCon future-proof: the hardness of the generated planning problems can be increased as the planning capabilities of LLMs improve. Our experiments reveal that the performance of state-of-the-art LLMs, including reasoning models like GPT-5, o3, and R1, deteriorates as the degree of constrainedness of the planning tasks increases.

Method: Built from 70k real-world PDF pages across eight domains, extracts and links evidence from text, tables, and figures, then generates 1,600 multimodal QA pairs spanning various query types with 20% expert validation.

Result: Multimodal text-image fusion RAG systems consistently outperform both unimodal and jointly multimodal embedding-based retrieval, showing that neither text nor images alone are sufficient and current multimodal embeddings remain inadequate.

Conclusion: The benchmark enables apples-to-apples comparison across different paradigms, reveals when visual context complements textual evidence, uncovers systematic failure modes, and offers actionable guidance for developing more robust MM-RAG pipelines.

Abstract: Multimodal retrieval-augmented generation (MM-RAG) is a key approach for applying large language models (LLMs) and agents to real-world knowledge bases, yet current evaluations are fragmented, focusing on either text or images in isolation or on simplified multimodal setups that fail to capture document-centric multimodal use cases. In this paper, we introduce UniDoc-Bench, the first large-scale, realistic benchmark for MM-RAG built from 70k real-world PDF pages across eight domains. Our pipeline extracts and links evidence from text, tables, and figures, then generates 1,600 multimodal QA pairs spanning factual retrieval, comparison, summarization, and logical reasoning queries. To ensure reliability, 20% of QA pairs are validated by multiple annotators and expert adjudication. UniDoc-Bench supports apples-to-apples comparison across four paradigms: (1) text-only, (2) image-only, (3) multimodal text-image fusion, and (4) multimodal joint retrieval – under a unified protocol with standardized candidate pools, prompts, and evaluation metrics. Our experiments show that multimodal text-image fusion RAG systems consistently outperform both unimodal and jointly multimodal embedding-based retrieval, indicating that neither text nor images alone are sufficient and that current multimodal embeddings remain inadequate. Beyond benchmarking, our analysis reveals when and how visual context complements textual evidence, uncovers systematic failure modes, and offers actionable guidance for developing more robust MM-RAG pipelines.

Method: Generated a full 8-permutation paradigm of refined parasitic gap stimuli and evaluated GPT-2 using systematic Wilcox-style wh-effect analysis with direct minimal pair comparisons.

Result: GPT-2 succeeded across all four tested conditions, showing robust knowledge of filler-gap licensing principles even in complex parasitic gap environments.

Conclusion: The choice of evaluation metric is critical for assessing LLMs’ syntactic competence, with direct minimal pair approaches offering greater diagnostic transparency than DiD-style metrics.

Abstract: Recent studies probing the Argument from the Poverty of the Stimulus (APS) have applied Large Language Models (LLMs) to test the learnability of complex syntax through surprisal-based metrics. However, divergent conclusions raise questions concerning the insights these metrics offer. While Wilcox et al. (2024) used direct minimal pair comparisons (the “wh-effect”) to demonstrate that models successfully generalise knowledge of filler-gap dependencies, Lan et al. (2024) used a Difference-in-Differences (DiD) metric and found that models largely fail on parasitic gaps (PGs). This paper argues that the direct minimal pair approach offers greater diagnostic transparency. We demonstrate this by generating a full 8-permutation paradigm of refined PG stimuli and evaluating the GPT-2 model used in previous studies with a systematic Wilcox-style wh-effect analysis. Our results show that GPT-2 succeeds across all four tested conditions, indicating robust knowledge of filler-gap licensing principles even in complex PG environments. This finding, which contrasts with the more ambiguous results from DiD-style metrics, suggests that the choice of evaluation metric is critical for assessing an LLM’s syntactic competence.

Method: MASA implements a multi-A, single-B structure where multiple specialized A experts are asymmetrically shared across layers to capture diverse features, integrated by layer-specific B matrices while maintaining parameter efficiency.

Result: MASA achieves 59.62% average accuracy on MMLU benchmark, outperforming standard LoRA by 1.08 points (1.84% relative improvement) with comparable 0.52% learnable parameters.

Conclusion: The multi-A shared adaptation architecture effectively improves downstream task adaptation ability by enriching feature representation while maintaining parameter efficiency, demonstrating superior performance across multiple domains and tasks.

Abstract: Low-Rank Adaptation (LoRA) has emerged as a dominant method in Parameter-Efficient Fine-Tuning (PEFT) for large language models, which augments the transformer layer with one down-projection $A$ and one up-projection $B$. However, LoRA’s reliance on a single down-projection matrix ($A$) creates a representational bottleneck, as this solitary feature extractor is inherently insufficient for capturing the diverse signals required by complex tasks. This motivates our architectural shift to focus on enriching the feature adaptation to improve the downstream task adaptation ability. We propose MASA (Multi-$A$ Shared Adaptation), an architecture that implements a multi-$A$, single-$B$ structure where the multi-$A$ expert ensemble is asymmetrically shared across layers to ensure parameter efficiency. In MASA, these specialized experts capture diverse features, which are then integrated by a single, layer-specific $B$-matrix. The effectiveness and versatility of our method are validated through a comprehensive suite of experiments spanning multi-domain generalization, single-domain specialization, and multi-task reasoning. For example, on the MMLU benchmark, MASA achieves an average accuracy of 59.62%, outperforming the standard LoRA by 1.08 points (a relative improvement of 1.84%) with comparable learnable parameters of 0.52%.

Method: 1) Establish baseline on previously used stimuli (filtered and unfiltered), 2) Generate new refined dataset using Gemini 2.5 Pro Preview with linguistically-informed templates to mitigate confounds, then evaluate GPT-2 performance.

Result: Preliminary findings show GPT-2 demonstrates notably improved performance on refined stimuli compared to baselines.

Conclusion: Stimulus quality significantly influences outcomes in surprisal-based evaluations of LLM syntactic competency.

Abstract: Recent studies employing Large Language Models (LLMs) to test the Argument from the Poverty of the Stimulus (APS) have yielded contrasting results across syntactic phenomena. This paper investigates the hypothesis that characteristics of the stimuli used in recent studies, including lexical ambiguities and structural complexities, may confound model performance. A methodology is proposed for re-evaluating LLM competence on syntactic prediction, focusing on GPT-2. This involves: 1) establishing a baseline on previously used (both filtered and unfiltered) stimuli, and 2) generating a new, refined dataset using a state-of-the-art (SOTA) generative LLM (Gemini 2.5 Pro Preview) guided by linguistically-informed templates designed to mitigate identified confounds. Our preliminary findings indicate that GPT-2 demonstrates notably improved performance on these refined PG stimuli compared to baselines, suggesting that stimulus quality significantly influences outcomes in surprisal-based evaluations of LLM syntactic competency.

Method: Constructed Chinese Data-Text Pair (CDTP) dataset with aligned text-triple pairs, enabling evaluation across Knowledge Graph Completion, Triple-to-Text generation, and Question Answering tasks.

Result: Created a benchmark with 7 million aligned text pairs and 15 million triples spanning four domains, supporting fine-grained evaluation and multi-task fine-tuning for Chinese LLMs.

Conclusion: CB-ECLLM addresses the gap in Chinese LLM evaluation by providing structured datasets and comprehensive benchmarks, with open-source codebase for reproducible research and future directions.

Abstract: Large Language Models (LLMs) have achieved remarkable success across a wide range of natural language processing tasks. However, Chinese LLMs face unique challenges, primarily due to the dominance of unstructured free text and the lack of structured representations in Chinese corpora. While existing benchmarks for LLMs partially assess Chinese LLMs, they are still predominantly English-centric and fail to address the unique linguistic characteristics of Chinese, lacking structured datasets essential for robust evaluation. To address these challenges, we present a Comprehensive Benchmark for Evaluating Chinese Large Language Models (CB-ECLLM) based on the newly constructed Chinese Data-Text Pair (CDTP) dataset. Specifically, CDTP comprises over 7 million aligned text pairs, each consisting of unstructured text coupled with one or more corresponding triples, alongside a total of 15 million triples spanning four critical domains. The core contributions of CDTP are threefold: (i) enriching Chinese corpora with high-quality structured information; (ii) enabling fine-grained evaluation tailored to knowledge-driven tasks; and (iii) supporting multi-task fine-tuning to assess generalization and robustness across scenarios, including Knowledge Graph Completion, Triple-to-Text generation, and Question Answering. Furthermore, we conduct rigorous evaluations through extensive experiments and ablation studies to assess the effectiveness, Supervised Fine-Tuning (SFT), and robustness of the benchmark. To support reproducible research, we offer an open-source codebase and outline potential directions for future investigations based on our insights.

Method: Proposed Asymmetric Importance Sampling Policy Optimization (ASPO) flips IS ratios for positive-advantage tokens and incorporates soft dual-clipping mechanism for stability.

Result: ASPO significantly mitigates premature convergence, improves training stability, and enhances final performance on coding and mathematical reasoning benchmarks over GRPO-based baselines.

Conclusion: Correcting IS ratios in LLM RL is critical, and ASPO provides effective solution for balanced token weighting and improved training dynamics.

Abstract: Recent Large Language Model (LLM) post-training methods rely on token-level clipping mechanisms during Reinforcement Learning (RL). However, we identify a fundamental flaw in this Outcome-Supervised RL (OSRL) paradigm: the Importance Sampling (IS) ratios of positive-advantage tokens are mismatched, leading to unbalanced token weighting for positive and negative tokens. This mismatch suppresses the update of low-probability tokens while over-amplifying already high-probability ones. To address this, we propose Asymmetric Importance Sampling Policy Optimization (ASPO), which uses a simple yet effective strategy that flips the IS ratios of positive-advantage tokens, aligning their update direction with the learning dynamics of negative ones. AIS further incorporates a soft dual-clipping mechanism to stabilize extreme updates while maintaining gradient flow. Comprehensive experiments on coding and mathematical reasoning benchmarks demonstrate that ASPO significantly mitigates premature convergence, improves training stability, and enhances final performance over strong GRPO-based baselines. Our analysis provides new insights into the role of token-level weighting in OSRL and highlights the critical importance of correcting IS in LLM RL. The code and models of ASPO are available at https://github.com/wizard-III/Archer2.0.

Method: Introduces Spectrum Suite (compiled from >40 data sources, >90 tasks) to evaluate distributional modeling, and proposes Spectrum Tuning as a post-training method to improve steerability and coverage.

Result: Post-training helps elicit underlying capabilities but hurts in-context steerability. Spectrum Tuning improves over both pretrained and instruction-tuned models on steerability, output space coverage, and distributional alignment.

Conclusion: Spectrum Tuning effectively mitigates the negative effects of standard post-training, enhancing models’ ability to flexibly steer to diverse distributions while maintaining their underlying capabilities.

Abstract: Language model post-training has enhanced instruction-following and performance on many downstream tasks, but also comes with an often-overlooked cost on tasks with many possible valid answers. We characterize three desiderata for conditional distributional modeling: in-context steerability, valid output space coverage, and distributional alignment, and document across three model families how current post-training can reduce these properties. In particular, we disambiguate between two kinds of in-context learning: ICL for eliciting existing underlying knowledge or capabilities, and in-context steerability, where a model must use in-context information to override its priors and steer to a novel data generating distribution. To better evaluate and improve these desiderata, we introduce Spectrum Suite, a large-scale resource compiled from >40 data sources and spanning >90 tasks requiring models to steer to and match diverse distributions ranging from varied human preferences to numerical distributions and more. We find that while current post-training techniques help elicit underlying capabilities and knowledge, they hurt models’ ability to flexibly steer in-context. To mitigate these issues, we propose Spectrum Tuning, a post-training method using Spectrum Suite to improve steerability and distributional coverage. We find that Spectrum Tuning often improves over pretrained models and their instruction-tuned counterparts, enhancing steerability, spanning more of the output space, and improving distributional alignment on held-out datasets.

Method: Studied scaling trends of distilling competitive coding skills on two small non-reasoning LLMs, fine-tuned at different distillation stages on the same data to analyze learning phases.

Result: Identified a valley pattern: performance first decreases then increases sharply with more data. Found that easier coding questions benefit small models more than harder ones, and data correctness doesn’t impact distillation results.

Conclusion: This work advances understanding of code reasoning distillation training dynamics, challenging intuitive assumptions about data quality and quantity effects.

Abstract: Distilling the thinking traces of a Large Language Model (LLM) with reasoning capabilities into a smaller model has been proven effective. Yet, there is a scarcity of work done on how model performances scale with the quantity of distillation data. In this work, we study the scaling trend of distilling competitive coding skills on two small non-reasoning LLMs. We validate the hypothesis that there is a $\textit{valley of code reasoning}$: downstream performance on competitive coding first drops as data quantity increases, then it steadily increases in a sharper-than-log-linear fashion. Having identified the trend, we further fine-tune the models at two different distillation stages on the same data to ground conclusions on their respective learning phases. We learn that across stages in the low and medium-low data regimes, small models benefit significantly from easier coding questions than from harder ones. We also find that, surprisingly, the correctness of outputs in training data makes no difference to distillation outcomes. Our work represents a step forward in understanding the training dynamics of code reasoning distillation outside intuition

Method: Proposed Distributional Semantics Tracing (DST) framework that integrates interpretability techniques to create causal maps of model reasoning. Identified a ‘commitment layer’ where hallucinations become irreversible and analyzed conflicts between associative (System 1-like) and contextual (System 2-like) computational pathways.

Result: Found that hallucinations become inevitable at a specific commitment layer where internal representations irreversibly diverge from factuality. Demonstrated a strong negative correlation (ρ = -0.863) between contextual pathway coherence and hallucination rates, showing these failures are predictable consequences of internal semantic weakness.

Conclusion: Provides a mechanistic account of how, when, and why hallucinations occur within Transformer architecture, revealing that hallucinations stem from predictable conflicts between fast associative and slow contextual reasoning pathways.

Abstract: Large Language Models (LLMs) are prone to hallucination, the generation of plausible yet factually incorrect statements. This work investigates the intrinsic, architectural origins of this failure mode through three primary contributions.First, to enable the reliable tracing of internal semantic failures, we propose \textbf{Distributional Semantics Tracing (DST)}, a unified framework that integrates established interpretability techniques to produce a causal map of a model’s reasoning, treating meaning as a function of context (distributional semantics). Second, we pinpoint the model’s layer at which a hallucination becomes inevitable, identifying a specific \textbf{commitment layer} where a model’s internal representations irreversibly diverge from factuality. Third, we identify the underlying mechanism for these failures. We observe a conflict between distinct computational pathways, which we interpret using the lens of dual-process theory: a fast, heuristic \textbf{associative pathway} (akin to System 1) and a slow, deliberate \textbf{contextual pathway} (akin to System 2), leading to predictable failure modes such as \textit{Reasoning Shortcut Hijacks}. Our framework’s ability to quantify the coherence of the contextual pathway reveals a strong negative correlation ($\rho = -0.863$) with hallucination rates, implying that these failures are predictable consequences of internal semantic weakness. The result is a mechanistic account of how, when, and why hallucinations occur within the Transformer architecture.

Method: Train tokenizers monolingually and then align their vocabularies exhaustively using bilingual dictionaries or word-to-word translation, ensuring consistent indices for semantically equivalent words.

Result: Models trained with parallel tokenizers outperform conventional multilingual baselines across sentiment analysis, hate speech detection, emotion classification, and sentence embedding similarity tasks on thirteen low-resource languages.

Conclusion: Rethinking tokenization is essential for advancing multilingual representation learning, especially in low-resource settings, as parallel tokenizers enforce shared semantic space across languages while improving fertility balance.

Abstract: Tokenization defines the foundation of multilingual language models by determining how words are represented and shared across languages. However, existing methods often fail to support effective cross-lingual transfer because semantically equivalent words are assigned distinct embeddings. For example, “I eat rice” in English and “Ina cin shinkafa” in Hausa are typically mapped to different vocabulary indices, preventing shared representations and limiting cross-lingual generalization. We introduce parallel tokenizers. This new framework trains tokenizers monolingually and then aligns their vocabularies exhaustively using bilingual dictionaries or word-to-word translation, ensuring consistent indices for semantically equivalent words. This alignment enforces a shared semantic space across languages while naturally improving fertility balance. To assess their effectiveness, we pretrain a transformer encoder from scratch on thirteen low-resource languages and evaluate it on sentiment analysis, hate speech detection, emotion classification, and sentence embedding similarity. Across all tasks, models trained with parallel tokenizers outperform conventional multilingual baselines, confirming that rethinking tokenization is essential for advancing multilingual representation learning–especially in low-resource settings.

Method: Proposes Trace Credit to quantify token convergence potential by accumulating historical logits, and CreditDecoding algorithm that fuses current logits with Trace Credit to accelerate confidence convergence.

Result: Achieves 5.48x speedup and 0.48 performance improvement over LLaDA-8B-Instruct, and 4.11x speedup with 0.15 improvement over LLaDA-MoE-Instruct on eight benchmarks.

Conclusion: CreditDecoding effectively scales to long sequences and is orthogonal to mainstream inference optimizations, making it a readily integrable and versatile solution for diffusion LLM acceleration.

Abstract: Diffusion large language models (dLLMs) generate text through iterative denoising steps, achieving parallel decoding by denoising only high-confidence positions at each step. However, existing approaches often repetitively remask tokens due to initially low confidence scores, leading to redundant iterations and limiting overall acceleration. Through the analysis of dLLM decoding traces, we observe that the model often determines the final prediction for a token several steps before the decoding step. To leverage this historical information and avoid redundant steps, we introduce the concept of Trace Credit, which quantifies each token’s convergence potential by accumulating historical logits. Furthermore, we propose CreditDecoding, a training-free parallel decoding algorithm that accelerates the confidence convergence of correct but underconfident tokens by fusing current logits with Trace Credit. This process significantly reduces redundant iterations and enhances decoding robustness. On eight benchmarks, CreditDecoding achieves a 5.48 times speedup and a 0.48 performance improvement over LLaDA-8B-Instruct, and a 4.11 times speedup with a 0.15 performance improvement over LLaDA-MoE-Instruct. Importantly, CreditDecoding scales effectively to long sequences and is orthogonal to mainstream inference optimizations, making it a readily integrable and versatile solution.

Method: RoSE trains a small model on outputs from a candidate LLM generator, then evaluates it on synthetic examples from other candidate LLMs, with the mean performance as the final RoSE score.

Result: RoSE identifies the optimal generator more often than intrinsic heuristics across six LLMs, eleven languages, and three tasks, coming within 0.76 percentage points of the optimal baseline and achieving positive correlation with human test data performance.

Conclusion: RoSE is an effective proxy metric for selecting the best LLM generator for synthetic data without requiring human annotations, outperforming intrinsic heuristics and closely matching optimal generator performance.

Abstract: LLMs are powerful generators of synthetic data, which are used for training smaller, specific models. This is especially valuable for low-resource languages, where human-labelled data is scarce but LLMs can still produce high-quality text. However, LLMs differ in how useful their outputs are for training. Selecting the best LLM as a generator is challenging because extrinsic evaluation requires costly human annotations (which are often unavailable for low-resource languages), while intrinsic metrics correlate poorly with downstream performance. We introduce Round robin Synthetic data Evaluation (RoSE), a proxy metric for selecting the best LLM generator without human test sets. RoSE trains a small model on the outputs of a candidate generator (LLM) and then evaluates it on generated synthetic examples from all other candidate LLMs. The final RoSE score is the mean performance of this small model. Across six LLMs, eleven languages, and three tasks (sentiment, topic, intent), RoSE identifies the optimal generator more often than any other intrinsic heuristics. RoSE outperforms intrinsic heuristics and comes within 0.76 percentage points of the optimal generator baseline. This result is measured in terms of downstream performance, obtained by training a small model on the chosen generator’s outputs (optimal vs. proxy metric selected) and evaluating it on human-labelled test data. Additionally, RoSE is the only metric to achieve a positive correlation with performance on human test data.

Method: Apply smooth and Hadamard transformations to suppress outliers in key cache, enabling comprehensive codebook coverage of original data distribution. Design optimized CUDA kernel that fuses computation with dequantization.

Result: Outperforms existing quantization baselines across long-context understanding and mathematical reasoning tasks. With 2-bit quantization, achieves performance comparable to full precision, with 2.7× speedup in large-batch self-attention and 8.3× reduction in single-batch end-to-end latency on Llama-3.1-8B.

Conclusion: VecInfer enables aggressive KV cache compression while maintaining performance and enabling efficient inference through outlier suppression and optimized kernel design.

Abstract: The Key-Value (KV) cache introduces substantial memory overhead during large language model (LLM) inference. Although existing vector quantization (VQ) methods reduce KV cache usage and provide flexible representational capacity across bit-widths, they suffer severe performance degradation at ultra-low bit-widths due to key cache outliers that hinder effective codebook utilization. To address this challenge, we propose VecInfer, a novel VQ method for aggressive KV cache compression while enabling efficient inference. By applying smooth and Hadamard transformations, VecInfer suppresses outliers in the key cache, enabling the codebook to comprehensively cover the original data distribution and thereby reducing quantization difficulty. To facilitate efficient deployment, we design an optimized CUDA kernel that fuses computation with dequantization to minimize memory access overhead. Extensive evaluations demonstrate that VecInfer consistently outperforms existing quantization baselines across both long-context understanding and mathematical reasoning tasks. With only 2-bit quantization, VecInfer achieves performance comparable to full precision, while delivering up to $\mathbf{2.7\times}$ speedup in large-batch self-attention computation and $\mathbf{8.3\times}$ reduction in single-batch end-to-end latency on Llama-3.1-8B with a 196k sequence length.

Method: Extensive experiments on nine models and ten binding tasks, developing a causal model combining positional, lexical, and reflexive mechanisms to estimate next token distributions.

Result: The positional mechanism becomes noisy and unreliable in middle positions as entity count increases, while lexical and reflexive mechanisms compensate. The combined causal model achieves 95% agreement with actual model behavior and generalizes to longer, more natural text inputs.

Conclusion: Language models employ a mix of positional, lexical, and reflexive mechanisms for entity binding and retrieval, with this combination providing a more complete understanding of in-context reasoning capabilities.

Abstract: A key component of in-context reasoning is the ability of language models (LMs) to bind entities for later retrieval. For example, an LM might represent “Ann loves pie” by binding “Ann” to “pie”, allowing it to later retrieve “Ann” when asked “Who loves pie?” Prior research on short lists of bound entities found strong evidence that LMs implement such retrieval via a positional mechanism, where “Ann” is retrieved based on its position in context. In this work, we find that this mechanism generalizes poorly to more complex settings; as the number of bound entities in context increases, the positional mechanism becomes noisy and unreliable in middle positions. To compensate for this, we find that LMs supplement the positional mechanism with a lexical mechanism (retrieving “Ann” using its bound counterpart “pie”) and a reflexive mechanism (retrieving “Ann” through a direct pointer). Through extensive experiments on nine models and ten binding tasks, we uncover a consistent pattern in how LMs mix these mechanisms to drive model behavior. We leverage these insights to develop a causal model combining all three mechanisms that estimates next token distributions with 95% agreement. Finally, we show that our model generalizes to substantially longer inputs of open-ended text interleaved with entity groups, further demonstrating the robustness of our findings in more natural settings. Overall, our study establishes a more complete picture of how LMs bind and retrieve entities in-context.

Method: Created RECODE-H benchmark with 102 tasks, structured instructions, unit tests, and a five-level feedback hierarchy. Developed ReCodeAgent framework for iterative code generation with feedback integration.

Result: Experiments with leading LLMs (GPT-5, Claude-Sonnet-4, DeepSeek-V3.1, Gemini 2.5) show substantial performance gains with richer feedback, though challenges remain in complex research code generation.

Conclusion: RECODE-H establishes a foundation for developing adaptive, feedback-driven LLM agents in scientific research implementation.

Abstract: Large language models (LLMs) show the promise in supporting scientific research implementation, yet their ability to generate correct and executable code remains limited. Existing works largely adopt one-shot settings, ignoring the iterative and feedback-driven nature of realistic workflows of scientific research development. To address this gap, we present RECODE-H, a benchmark of 102 tasks from research papers and repositories that evaluates LLM agents through multi-turn interactions with LLM-simulated human feedback. It includes structured instructions,unit tests, and a five-level feedback hierarchy to reflect realistic researcher-agent collaboration. We further present ReCodeAgent, a framework that integrates feedback into iterative code generation. Experiments with leading LLMs, including GPT-5, Claude-Sonnet-4, DeepSeek-V3.1, and Gemini 2.5, show substantial performance gains with richer feedback, while also highlighting ongoing challenges in the generation of complex research code. RECODE-H establishes a foundation for developing adaptive, feedback-driven LLM agents in scientific research implementation

Method: Client-server architecture using Real-time Transport Protocol (RTP) for low-latency communication. Developed BRDialect ASR by fine-tuning IndicWav2Vec model on ten Bengali regional dialects.

Result: BRDialect outperforms baseline ASR models by 12.41-33.98% on RegSpeech12 dataset. System operates at 24 kbps bandwidth with average end-to-end delay of 4.9 seconds.

Conclusion: BanglaTalk enables cost-effective, interactive real-time speech technology for Bengali speakers across diverse dialects, promoting inclusive and accessible speech assistance.

Abstract: Real-time speech assistants are becoming increasingly popular for ensuring improved accessibility to information. Bengali, being a low-resource language with a high regional dialectal diversity, has seen limited progress in developing such systems. Existing systems are not optimized for real-time use and focus only on standard Bengali. In this work, we present BanglaTalk, the first real-time speech assistance system for Bengali regional dialects. BanglaTalk follows the client-server architecture and uses the Real-time Transport Protocol (RTP) to ensure low-latency communication. To address dialectal variation, we introduce a dialect-aware ASR system, BRDialect, developed by fine-tuning the IndicWav2Vec model in ten Bengali regional dialects. It outperforms the baseline ASR models by 12.41-33.98% on the RegSpeech12 dataset. Furthermore, BanglaTalk can operate at a low bandwidth of 24 kbps while maintaining an average end-to-end delay of 4.9 seconds. Low bandwidth usage and minimal end-to-end delay make the system both cost-effective and interactive for real-time use cases, enabling inclusive and accessible speech technology for the diverse community of Bengali speakers.

Method: Two-component framework: (1) cognitive science-inspired reasoning mechanism for relation extraction, (2) RL optimization with novel reward function that promotes important relation keywords from an LLM-constructed dictionary.

Result: Achieves 24.65% F1 on One-shot NYT29 with Qwen2.5-15B-Instruct, with RL optimization providing +23.46% absolute improvement. Human evaluation shows 54% relative increase in explanation quality ratings.

Conclusion: CogRE successfully enhances both accuracy and explainability in relation extraction through cognitive-structured reasoning and RL optimization, generating high-quality relational keywords that align with human judgments.

Abstract: This paper introduces a framework for relation extraction (RE) that enhances both accuracy and explainability. The framework has two key components: (i) a reasoning mechanism that formulates relation extraction as a series of text-processing steps inspired by cognitive science, and (ii) an optimization process driven by reinforcement learning (RL) with a novel reward function designed to improve both task accuracy and explanation quality. We call our approach CogRE. Our framework addresses the lack of supervision for language-based explanations in traditional RE by promoting outputs that include important relation keywords. These keywords are drawn from a high-quality dictionary that is automatically constructed using an LLM. We evaluate our approach for the task of one-shot RE using two LLMs and two RE datasets. Our experiments show that CogRE improves explanation quality by addressing two common failure patterns in one-shot RE: poor attention focus and limited one-shot learning capability. For example, our cognitive-structured reasoning with Qwen2.5-15B-Instruct on One-shot NYT29 achieves 24.65% F1, surpassing prior reasoning-based designs. Optimizing this approach with RL using our reward further improves performance by +23.46% (absolute). Finally, human evaluation shows that our best model generates relational keywords closely aligned with gold labels, increasing human explanation quality ratings by 54% (relative).

Method: Developed MathVC with a meta planning controller that monitors CPS stages (sense-making, team organization, planning, execution, validation) and predicts next speaker, combined with persona simulation using task schemas and error-injected persona schemas based on teacher-specified misconceptions.

Result: Evaluation with 14 U.S. middle schoolers showed constructive interactions, shared solutions, and gains in engagement, motivation, and confidence through diverse perspectives, immediate scaffolding, and human-like fallibility.

Conclusion: LLM-based technologies can effectively simulate peers for collaboration to support learning, providing insights into enhancing collaborative problem solving in mathematics education.

Abstract: Collaborative problem solving (CPS) is essential in mathematics education, fostering deeper learning through the exchange of ideas. Yet, classrooms often lack the resources, time, and peer dynamics needed to sustain productive CPS. Recent advancements in Large Language Models (LLMs) offer a promising avenue to enhance CPS in mathematical education. We designed and developed MathVC, a multi-persona LLM simulated virtual classroom platform to facilitate CPS in mathematics. MathVC combines a meta planning controller that monitors CPS stages-sense-making, team organization, planning, execution, validation, and predicts the next speaker, with a persona simulation stack that encodes mathematical thinking via a task schema and error-injected persona schemas seeded from teacher-specified misconceptions. We evaluated MathVC with 14 U.S. middle schoolers. Students reported constructive interaction and reaching shared solutions, describing gains in engagement, motivation, and confidence through diverse perspectives, immediate scaffolding, and human-like fallibility. Our findings also provide insights into simulating peers via LLM-based technologies for collaboration to support learning.

Method: Developed SciKnowEval dataset with 28K multi-level questions spanning four scientific domains, then evaluated 20 leading open-source and proprietary LLMs using this benchmark across five progressive levels of scientific understanding.

Result: Proprietary models often achieve state-of-the-art performance, but substantial challenges remain particularly in scientific reasoning and real-world application.

Conclusion: SciKnowEval serves as a standard benchmark for evaluating scientific capabilities in LLMs and a catalyst for advancing more capable and reliable scientific language models.

Abstract: Large language models (LLMs) are playing an increasingly important role in scientific research, yet there remains a lack of comprehensive benchmarks to evaluate the breadth and depth of scientific knowledge embedded in these models. To address this gap, we introduce SciKnowEval, a large-scale dataset designed to systematically assess LLMs across five progressive levels of scientific understanding: memory, comprehension, reasoning, discernment, and application. SciKnowEval comprises 28K multi-level questions and solutions spanning biology, chemistry, physics, and materials science. Using this benchmark, we evaluate 20 leading open-source and proprietary LLMs. The results show that while proprietary models often achieve state-of-the-art performance, substantial challenges remain – particularly in scientific reasoning and real-world application. We envision SciKnowEval as a standard benchmark for evaluating scientific capabilities in LLMs and as a catalyst for advancing more capable and reliable scientific language models.

Method: Artificially introduced varying levels of noise into diverse datasets and systematically evaluated LLMs’ robustness against corrupted text variations. Also tested on real-world benchmarks mimicking common errors.

Result: Generative LLMs are robust to noisy perturbations, contrary to popular belief. They achieve state-of-the-art performance on Grammar Error Correction and Lexical Semantic Change benchmarks with minimal prompting.

Conclusion: LLMs demonstrate strong resilience to text noise, making them suitable for real-world applications where clean data is rare. The work releases a human-annotated dataset and code for reproducibility.

Abstract: Having a clean dataset has been the foundational assumption of most natural language processing (NLP) systems. However, properly written text is rarely found in real-world scenarios and hence, oftentimes invalidates the aforementioned foundational assumption. Recently, Large language models (LLMs) have shown impressive performance, but can they handle the inevitable noise in real-world data? This work tackles this critical question by investigating LLMs’ resilience against morphological variations in text. To that end, we artificially introduce varying levels of noise into a diverse set of datasets and systematically evaluate LLMs’ robustness against the corrupt variations of the original text. Our findings show that contrary to popular beliefs, generative LLMs are quiet robust to noisy perturbations in text. This is a departure from pre-trained models like BERT or RoBERTa whose performance has been shown to be sensitive to deteriorating noisy text. Additionally, we test LLMs’ resilience on multiple real-world benchmarks that closely mimic commonly found errors in the wild. With minimal prompting, LLMs achieve a new state-of-the-art on the benchmark tasks of Grammar Error Correction (GEC) and Lexical Semantic Change (LSC). To empower future research, we also release a dataset annotated by humans stating their preference for LLM vs. human-corrected outputs along with the code to reproduce our results.

Method: Two-step framework: 1) LLM generates candidate labels from dataset and merges similar ones, 2) LLM assigns the most appropriate label to each text sample using in-context learning without fine-tuning.

Result: Achieves comparable or superior performance to state-of-the-art embedding-based clustering techniques while significantly reducing computational complexity and resource requirements.

Conclusion: Demonstrates the transformative potential of LLMs in simplifying and enhancing text clustering tasks with minimal human intervention.

Abstract: Text clustering serves as a fundamental technique for organizing and interpreting unstructured textual data, particularly in contexts where manual annotation is prohibitively costly. With the rapid advancement of Large Language Models (LLMs) and their demonstrated effectiveness across a broad spectrum of NLP tasks, an emerging body of research has begun to explore their potential in the domain of text clustering. However, existing LLM-based approaches still rely on fine-tuned embedding models and sophisticated similarity metrics, rendering them computationally intensive and necessitating domain-specific adaptation. To address these limitations, we propose a novel framework that reframes text clustering as a classification task by harnessing the in-context learning capabilities of LLMs. Our framework eliminates the need for fine-tuning embedding models or intricate clustering algorithms. It comprises two key steps: first, the LLM is prompted to generate a set of candidate labels based on the dataset and then merges semantically similar labels; second, it assigns the most appropriate label to each text sample. By leveraging the advanced natural language understanding and generalization capabilities of LLMs, the proposed approach enables effective clustering with minimal human intervention. Experimental results on diverse datasets demonstrate that our framework achieves comparable or superior performance to state-of-the-art embedding-based clustering techniques, while significantly reducing computational complexity and resource requirements. These findings underscore the transformative potential of LLMs in simplifying and enhancing text clustering tasks. We make our code available to the public for utilization at https://github.com/ECNU-Text-Computing/Text-Clustering-via-LLM. We also provide the supplementary Appendix within the repository.

Method: Used a uniform Retrieval Augmented Generation framework to generate responses from 24 LLMs on real patient data from Indian medical chatbot interactions in Indian English and 4 Indic languages, with evaluation using both automated techniques and human evaluators on four specific metrics.

Result: Models showed significant performance variations, instruction-tuned Indic models didn’t always perform well on Indic language queries, factual correctness was generally lower for Indic queries compared to English queries, and code-mixed and culturally relevant queries posed challenges.

Conclusion: Current LLMs have limitations in handling Indic languages and cultural contexts in real-world medical applications, highlighting the need for better evaluation frameworks and model development that accounts for linguistic and cultural nuances.

Abstract: Assessing the capabilities and limitations of large language models (LLMs) has garnered significant interest, yet the evaluation of multiple models in real-world scenarios remains rare. Multilingual evaluation often relies on translated benchmarks, which typically do not capture linguistic and cultural nuances present in the source language. This study provides an extensive assessment of 24 LLMs on real world data collected from Indian patients interacting with a medical chatbot in Indian English and 4 other Indic languages. We employ a uniform Retrieval Augmented Generation framework to generate responses, which are evaluated using both automated techniques and human evaluators on four specific metrics relevant to our application. We find that models vary significantly in their performance and that instruction tuned Indic models do not always perform well on Indic language queries. Further, we empirically show that factual correctness is generally lower for responses to Indic queries compared to English queries. Finally, our qualitative work shows that code-mixed and culturally relevant queries in our dataset pose challenges to evaluated models.

Method: Created two translated Bangla-instruction datasets (224k total samples) and used them to develop BanglaLlama - a family of five base and instruct variant LLMs specifically for Bangla language.

Result: Comprehensive benchmarking results demonstrate the effectiveness of the proposed datasets and models across multiple benchmarks for Bangla language processing.

Conclusion: The proposed datasets and BanglaLlama models will serve as the new standard baseline for future research on Bangla, addressing the gap for this widely spoken yet low-resource language.

Abstract: Bangla is a language spoken by approximately 240 million native speakers and around 300 million people worldwide. Despite being the 5th largest spoken language in the world, Bangla is still a “low-resource” language, and existing pretrained language models often struggle to perform well on Bangla Language Processing (BLP) tasks. This paper addresses this gap by: (1) introducing two high-quality translated Bangla-instruction datasets totaling 224k samples - Bangla-Orca (172k) and Bangla-Alpaca (52k); and (2) leveraging these datasets to develop BanglaLlama, an open-source family of Bangla-specific LLMs, consisting of five base and instruct variants. We present our methodology, two large datasets, and comprehensive benchmarking results showcasing the effectiveness of our dataset and model on multiple benchmarks. We believe our proposed datasets and models will serve as the new standard baseline for future research focused on this widely spoken yet “low-resource” language.

Method: Leveraged contemporary measurement theory to decode how GPT-4 and GPT-5 interrelate depressive symptoms, comparing their symptom relationships with standard instruments and expert judgments.

Result: GPT-4 showed strong convergent validity with clinical standards (r=0.70-0.81) and aligned with depression literature on symptom inter-correlations (r=0.23-0.78), but underemphasized suicidality relationships while overemphasizing psychomotor symptoms. GPT-5 had slightly lower convergence with self-reports. The analysis revealed novel symptom mechanisms, such as sleep/fatigue being broadly influenced by other symptoms while worthlessness/guilt is tied only to depressed mood.

Conclusion: The study provides an empirical foundation for understanding LLMs’ mental health assessments and demonstrates a generalizable approach for explainability that can guide stakeholders in effectively integrating these technologies into healthcare systems.

Abstract: Use of large language models such as ChatGPT (GPT-4/GPT-5) for mental health support has grown rapidly, emerging as a promising route to assess and help people with mood disorders like depression. However, we have a limited understanding of these language models’ schema of mental disorders, that is, how they internally associate and interpret symptoms of such disorders. In this work, we leveraged contemporary measurement theory to decode how GPT-4 and GPT-5 interrelate depressive symptoms, providing an explanation of how LLMs apply what they learn and informing clinical applications. We found that GPT-4 (a) had strong convergent validity with standard instruments and expert judgments $(r = 0.70 - 0.81)$, and (b) behaviorally linked depression symptoms with each other (symptom inter-correlates $r = 0.23 - 0.78$) in accordance with established literature on depression; however, it (c) underemphasized the relationship between $\textit{suicidality}$ and other symptoms while overemphasizing $\textit{psychomotor symptoms}$; and (d) suggested novel hypotheses of symptom mechanisms, for instance, indicating that $\textit{sleep}$ and $\textit{fatigue}$ are broadly influenced by other depressive symptoms, while $\textit{worthlessness/guilt}$ is only tied to $\textit{depressed mood}$. GPT-5 showed a slightly lower convergence with self-report, a difference our machine-behavior analysis makes interpretable through shifts in symptom-symptom relationships. These insights provide an empirical foundation for understanding language models’ mental health assessments and demonstrate a generalizable approach for explainability in other models and disorders. Our findings can guide key stakeholders to make informed decisions for effectively situating these technologies in the care system.

Method: Extended BBQ to Open-BBQ with two new question categories. Developed evaluation process for open-ended content. Proposed Composite Prompting - an ICL method combining structured examples with chain-of-thought reasoning.

Result: Significantly reduces bias for both GPT-3.5 and GPT-4o while maintaining high accuracy. Addresses over-correction issues in existing debiasing methods.

Conclusion: Open-BBQ provides better bias evaluation framework. Composite Prompting effectively reduces bias without compromising accuracy, solving over-correction problems.

Abstract: Current social bias benchmarks for Large Language Models (LLMs) primarily rely on predefined question formats like multiple-choice, limiting their ability to reflect the complexity and open-ended nature of real-world interactions. To close this gap, we extend an existing dataset BBQ (Parrish et al., 2022) to Open-BBQ, a comprehensive framework to evaluate the social bias of LLMs in open-ended settings by incorporating two additional question categories: fill-in-the-blank and short-answer. Since our new Open-BBQ dataset contains a lot of open-ended responses like sentences and paragraphs, we developed an evaluation process to detect biases from open-ended content by labeling sentences and paragraphs. In addition to this, we also found that existing debiasing methods, such as self-debiasing (Gallegos et al., 2024), have over-correction issues, which make the original correct answers incorrect. In order to solve this issue, we propose Composite Prompting, an In-context Learning (ICL) method combining structured examples with explicit chain-of-thought reasoning to form a unified instruction template for LLMs to explicitly identify content that needs debiasing. Experimental results show that the proposed method significantly reduces the bias for both GPT-3.5 and GPT-4o while maintaining high accuracy.

Method: QAPyramid decomposes reference summaries into finer-grained question-answer pairs using the QA-SRL framework, collecting 8.9K QA-level annotations for CNN/DM summaries across 10 systems.

Result: QAPyramid provides more systematic and fine-grained content selection evaluation than Pyramid while maintaining high inter-annotator agreement without requiring expert annotations.

Conclusion: The proposed automated metrics achieve higher correlations with QAPyramid than other widely adopted metrics, offering an improved evaluation pipeline for text summarization.

Abstract: How to properly conduct human evaluations for text summarization is a longstanding challenge. The Pyramid human evaluation protocol, which assesses content selection by breaking the reference summary into subunits and verifying their presence in the system summary, has been widely adopted. However, it suffers from a lack of systematicity in the definition and granularity of the sub-units. We address these problems by proposing QAPyramid, which decomposes each reference summary into finer-grained question-answer (QA) pairs according to the QA-SRL framework. We collect QA-SRL annotations for reference summaries from CNN/DM and evaluate 10 summarization systems, resulting in 8.9K QA-level annotations. We show that, compared to Pyramid, QAPyramid provides more systematic and fine-grained content selection evaluation while maintaining high inter-annotator agreement without needing expert annotations. Furthermore, we propose metrics that automate the evaluation pipeline and achieve higher correlations with QAPyramid than other widely adopted metrics.

Method: Treat RAG pipeline components as cooperative RL agents and use multi-agent reinforcement learning (MMOA-RAG) to harmonize all agents’ goals toward a unified reward based on final answer quality metrics like F1 score.

Result: Experiments on various QA benchmarks show MMOA-RAG effectively boosts overall pipeline performance and outperforms existing baselines. Ablation studies validate individual component contributions and demonstrate adaptability to different RAG pipelines.

Conclusion: MMOA-RAG successfully addresses the limitations of separate component optimization in RAG pipelines by using multi-agent RL to coordinate all components toward the common goal of generating accurate answers, with demonstrated effectiveness across different benchmarks and pipeline configurations.

Abstract: Retrieval-augmented generation (RAG) is widely utilized to incorporate external knowledge into large language models, thereby enhancing factuality and reducing hallucinations in question-answering (QA) tasks. A standard RAG pipeline consists of several components, such as query rewriting, document retrieval, document filtering, and answer generation. However, these components are typically optimized separately through supervised fine-tuning, which can lead to misalignments between the objectives of individual components and the overarching aim of generating accurate answers. Although recent efforts have explored using reinforcement learning (RL) to optimize specific RAG components, these approaches often focus on simple pipelines with only two components or do not adequately address the complex interdependencies and collaborative interactions among the modules. To overcome these limitations, we propose treating the complex RAG pipeline with multiple components as a multi-agent cooperative task, in which each component can be regarded as an RL agent. Specifically, we present MMOA-RAG, Multi-Module joint Optimization Algorithm for RAG, which employs multi-agent reinforcement learning to harmonize all agents’ goals toward a unified reward, such as the F1 score of the final answer. Experiments conducted on various QA benchmarks demonstrate that MMOA-RAG effectively boost the overall performance of the pipeline and outperforms existing baselines. Furthermore, comprehensive ablation studies validate the contributions of individual components and demonstrate MMOA-RAG can be adapted to different RAG pipelines and benchmarks.

Method: Augment the model’s vocabulary with a special red flag token and train the model to insert this token when harmful content is generated or imminent, leveraging LLM generalization capabilities like in-context learning.

Result: The approach enables explicit learning of harmfulness concepts, minimal impact on utility, and allows models to initiate reflective reasoning upon generating the red flag token for self-correction or steering away from harmful continuations.

Conclusion: This red flag token approach is orthogonal to existing safety techniques, easier to evaluate than natural language refusals, and complementary to standard safety training methods.

Abstract: Many safety post-training methods for large language models (LLMs) are designed to modify the model’s behaviour from producing unsafe answers to issuing refusals. However, such distribution shifts are often brittle and degrade performance on desirable tasks. To address these pitfalls, we propose augmenting the model’s vocabulary with a special red flag token, and training the model to insert this token whenever harmful content is generated or imminent. This approach enables the model to explicitly learn the concept of harmfulness in its representations, with minimal impact on utility due to the marginal change in the generated distribution of natural language. Moreover, because the token is embedded in the model’s vocabulary, we can naturally leverage the LLMs’ generalization capabilities, such as in-context learning (ICL) and out-of-distribution generalization to languages that are not formally supported (e.g., Japanese for Llama3). In particular, we demonstrate that through ICL alone, the model can learn to initiate reflective reasoning upon generating the red flag token at inference, which steers the response away from harmful continuations or enables self-correction when the flag is raised falsely. This approach is orthogonal and complementary to existing safety technique (such as safety classifiers or standard safety training) and easier to evaluate in comparison to natural language refusals, as it does not require a human or automated judge to assess the harmlessness of the answers.

Method: Used a statistics-based method to study the LLama-2 model family on selected relations, measuring effects of selectively deactivating candidate relation-specific neurons on factual recall.

Result: Demonstrated existence of relation-specific neurons with three properties: (i) multiple neurons cumulatively process relation facts, (ii) neurons are versatile across related relations and languages, (iii) deactivating one relation’s neurons can improve recall for other relations.

Conclusion: Relation-specific neurons exist in LLMs and exhibit cumulativity, versatility, and interference properties, providing insights into how LLMs encode and process relational knowledge.

Abstract: In large language models (LLMs), certain \emph{neurons} can store distinct pieces of knowledge learned during pretraining. While factual knowledge typically appears as a combination of \emph{relations} and \emph{entities}, it remains unclear whether some neurons focus on a relation itself – independent of any entity. We hypothesize such neurons \emph{detect} a relation in the input text and \emph{guide} generation involving such a relation. To investigate this, we study the LLama-2 family on a chosen set of relations, with a \textit{statistics}-based method. Our experiments demonstrate the existence of relation-specific neurons. We measure the effect of selectively deactivating candidate neurons specific to relation $r$ on the LLM’s ability to handle (1) facts involving relation $r$ and (2) facts involving a different relation $r’ \neq r$. With respect to their capacity for encoding relation information, we give evidence for the following three properties of relation-specific neurons. \textbf{(i) Neuron cumulativity.} Multiple neurons jointly contribute to processing facts involving relation $r$, with no single neuron fully encoding a fact in $r$ on its own. \textbf{(ii) Neuron versatility.} Neurons can be shared across multiple closely related as well as less related relations. In addition, some relation neurons transfer across languages. \textbf{(iii) Neuron interference.} Deactivating neurons specific to one relation can improve LLMs’ factual recall performance for facts of other relations. We make our code and data publicly available at https://github.com/cisnlp/relation-specific-neurons.

Method: Systematic evaluation using BBQ benchmark datasets across three languages (English, Japanese, Chinese) and four bias types (gender, race, age, religion), testing RAG responses from document collections with varying stereotypical bias levels using multiple LLMs as generators.

Result: Biases in document collections are often amplified in generated responses, even when the generating LLM exhibits low-level bias.

Conclusion: RAG use for injecting novel facts into NLG systems raises concerns and requires careful evaluation of potential social biases before real-world deployment.

Abstract: Retrieval Augmented Generation (RAG) has gained popularity as a method for conveniently incorporating novel facts that were not seen during the pre-training stage in Large Language Model (LLM)-based Natural Language Generation (NLG) systems. However, LLMs are known to encode significant levels of unfair social biases. The modulation of these biases by RAG in NLG systems is not well understood. In this paper, we systematically study the relationship between the different components of a RAG system and the social biases presented in the text generated across three languages (i.e. English, Japanese and Chinese) and four social bias types (i.e. gender, race, age and religion). Specifically, using the Bias Question Answering (BBQ) benchmark datasets, we evaluate the social biases in RAG responses from document collections with varying levels of stereotypical biases, employing multiple LLMs used as generators. We find that the biases in document collections are often amplified in the generated responses, even when the generating LLM exhibits a low-level of bias. Our findings raise concerns about the use of RAG as a technique for injecting novel facts into NLG systems and call for careful evaluation of potential social biases in RAG applications before their real-world deployment.

Method: The framework extends curvature concepts via the Whewell equation to n-dimensional word embedding spaces, quantifying local geometric properties like semantic shifts and curvature. It also uses Local Intrinsic Dimensionality (LID) to capture complementary geometric features in adversarial subspaces.

Result: CurvaLID achieves near-perfect classification in adversarial prompt detection and outperforms state-of-the-art detectors, showing that adversarial prompts exhibit distinct geometric signatures from benign prompts.

Conclusion: CurvaLID provides a reliable, efficient, model-agnostic safeguard against malicious queries that generalizes across multiple LLMs and attack families.

Abstract: Adversarial prompts are capable of jailbreaking frontier large language models (LLMs) and inducing undesirable behaviours, posing a significant obstacle to their safe deployment. Current mitigation strategies primarily rely on activating built-in defence mechanisms or fine-tuning LLMs, both of which are computationally expensive and can sacrifice model utility. In contrast, detection-based approaches are more efficient and practical for deployment in real-world applications. However, the fundamental distinctions between adversarial and benign prompts remain poorly understood. In this work, we introduce CurvaLID, a novel defence framework that efficiently detects adversarial prompts by leveraging their geometric properties. It is agnostic to the type of LLM, offering a unified detection framework across diverse adversarial prompts and LLM architectures. CurvaLID builds on the geometric analysis of text prompts to uncover their underlying differences. We theoretically extend the concept of curvature via the Whewell equation into an $n$-dimensional word embedding space, enabling us to quantify local geometric properties, including semantic shifts and curvature in the underlying manifolds. To further enhance our solution, we leverage Local Intrinsic Dimensionality (LID) to capture complementary geometric features of text prompts within adversarial subspaces. Our findings show that adversarial prompts exhibit distinct geometric signatures from benign prompts, enabling CurvaLID to achieve near-perfect classification and outperform state-of-the-art detectors in adversarial prompt detection. CurvaLID provides a reliable and efficient safeguard against malicious queries as a model-agnostic method that generalises across multiple LLMs and attack families.

Method: Created WildIFEval dataset with 7K real user instructions spanning diverse constraints, categorized into eight classes, then conducted extensive experiments benchmarking leading LLMs.

Result: WildIFEval clearly differentiates between small and large models, shows all models have large room for improvement, and reveals patterns in how models handle different constraint types and quantities.

Conclusion: The dataset promotes further research on instruction-following under complex, realistic conditions, highlighting the need for improved multi-constraint handling in LLMs.

Abstract: Recent LLMs have shown remarkable success in following user instructions, yet handling instructions with multiple constraints remains a significant challenge. In this work, we introduce WildIFEval - a large-scale dataset of 7K real user instructions with diverse, multi-constraint conditions. Unlike prior datasets, our collection spans a broad lexical and topical spectrum of constraints, extracted from natural user instructions. We categorize these constraints into eight high-level classes to capture their distribution and dynamics in real-world scenarios. Leveraging WildIFEval, we conduct extensive experiments to benchmark the instruction-following capabilities of leading LLMs. WildIFEval clearly differentiates between small and large models, and demonstrates that all models have a large room for improvement on such tasks. We analyze the effects of the number and type of constraints on performance, revealing interesting patterns of model constraint-following behavior. We release our dataset to promote further research on instruction-following under complex, realistic conditions.

Method: Teach LLMs to reason in a Bayesian manner by training them to mimic predictions of normative Bayesian models, rather than relying on their inherent reasoning capabilities.

Result: LLMs initially don’t update beliefs as expected from Bayesian framework. After training, performance significantly improves on recommendation tasks and generalizes to other tasks, suggesting better approximation of Bayesian reasoning.

Conclusion: LLMs can effectively learn reasoning skills from examples and generalize those skills to new domains, indicating potential for teaching complex reasoning frameworks.

Abstract: Artificial intelligence systems based on large language models (LLMs) are increasingly used as agents that interact with users and with the world. To do so successfully, LLMs need to construct internal representations of the world and form probabilistic beliefs about those representations. To provide a user with personalized recommendations, for example, the LLM needs to gradually infer the user’s preferences, over the course of multiple interactions. To evaluate whether contemporary LLMs are able to do so, we use the Bayesian inference framework from probability theory, which lays out the optimal way to update an agent’s beliefs as it receives new information. We first show that LLMs do not update their beliefs as expected from the Bayesian framework, and that consequently their predictions do not improve as expected as more information becomes available. To address this issue, we teach the LLMs to reason in a Bayesian manner by training them to mimic the predictions of the normative Bayesian model. We find that this approach not only significantly improves the LLM’s performance on the particular recommendation task it is trained on, but also enables generalization to other tasks. This suggests that this method teaches the LLM to better approximate Bayesian reasoning. More generally, our results indicate that LLMs can effectively learn reasoning skills from examples and generalize those skills to new domains.

Method: Proposes two innovations: context-efficient architecture with hierarchical section search to reduce token consumption, and scenario-based dialogue generation methodology to distill tool-using capabilities from larger models.

Result: Experimental evaluations show the fine-tuned 8B parameter model substantially outperforms both untuned models and baseline approaches in terms of DB faithfulness and preference.

Conclusion: The work offers valuable insights for researchers developing domain-specific language agents under practical constraints.

Abstract: Language agents powered by large language models (LLMs) face significant deployment challenges in resource-constrained environments, particularly for specialized domains and less-common languages. This paper presents Tox-chat, a Korean chemical toxicity information agent devised within these limitations. We propose two key innovations: a context-efficient architecture that reduces token consumption through hierarchical section search, and a scenario-based dialogue generation methodology that effectively distills tool-using capabilities from larger models. Experimental evaluations demonstrate that our fine-tuned 8B parameter model substantially outperforms both untuned models and baseline approaches, in terms of DB faithfulness and preference. Our work offers valuable insights for researchers developing domain-specific language agents under practical constraints.

Method: MASRAD-Ex systematically extracts foreign-Arabic term pairs from specialized books, computes multiple similarity metrics (lexicographic, phonetic, morphological, semantic), and uses heuristic, machine learning, and post-processing approaches to select the best candidate pairs.

Result: The best performing MASRAD-Ex approach achieved 90.5% precision and 92.4% recall. The dataset underwent thorough expert review and is made available to the research community.

Conclusion: MASRAD provides a valuable terminology resource for Arabic language processing, with high-quality semi-automatic extraction methods that effectively identify foreign-Arabic term pairs from specialized texts.

Abstract: This paper presents MASRAD, a terminology dataset for Arabic terminology management, and a method with supporting tools for its semi-automatic construction. The entries in MASRAD are $(f,a)$ pairs of foreign (non-Arabic) terms $f$, appearing in specialized, academic and field-specific books next to their Arabic $a$ counterparts. MASRAD-Ex systematically extracts these pairs as a first step to construct MASRAD. MASRAD helps improving term consistency in academic translations and specialized Arabic documents, and automating cross-lingual text processing. MASRAD-Ex leverages translated terms organically occurring in Arabic books, and considers several candidate pairs for each term phrase. The candidate Arabic terms occur next to the foreign terms, and vary in length. MASRAD-Ex computes lexicographic, phonetic, morphological, and semantic similarity metrics for each candidate pair, and uses heuristic, machine learning, and machine learning with post-processing approaches to decide on the best candidate. This paper presents MASRAD after thorough expert review and makes it available to the interested research community. The best performing MASRAD-Ex approach achieved 90.5% precision and 92.4% recall.

Method: Entropy-Gated Branching (EGB) - branches only at high-uncertainty reasoning steps and prunes expansions using a lightweight verifier to focus computational resources on critical decision points.

Result: On mathematical and financial reasoning benchmarks, EGB improves accuracy by 22.6% over standard inference while operating 31%-75% faster across math benchmarks than test-time beam search with higher performance.

Conclusion: Dynamic resource allocation during inference can substantially improve both efficiency and effectiveness, offering a more scalable pathway to enhanced LLM reasoning capabilities.

Abstract: Test-time compute methods can significantly improve the reasoning capabilities and problem-solving accuracy of large language models (LLMs). However, these approaches require substantially more computational resources, with most compute wasted on exploring low-diversity branches where the model already exhibits high confidence. We observe that a small subset of uncertain reasoning steps has a disproportionately large impact on final prediction accuracy, and branching at these critical junctures tends to yield more diverse and higher-quality candidate reasoning steps. We propose Entropy-Gated Branching (EGB), which branches only at high-uncertainty steps and prunes expansions with a lightweight verifier. On mathematical and financial reasoning benchmarks, EGB improves accuracy by 22.6% over standard inference while operating 31%-75% faster across math benchmarks than test-time beam search with higher performance. Our results show that dynamic resource allocation during inference can substantially improve both efficiency and effectiveness, offering a more scalable pathway to enhanced LLM reasoning capabilities.

Method: Proposes Self-Routing RAG framework with multi-task objective for joint optimization of knowledge source selection, knowledge verbalization, and response generation, plus nearest neighbor search for improved decision accuracy.

Result: Fine-tuning three LLMs with SR-RAG significantly improves response accuracy and reduces inference latency, reducing retrievals by 29% while improving performance by 5.1% compared to strongest baseline.

Conclusion: SR-RAG effectively bridges the gap in selective retrieval by better utilizing LLMs’ parametric knowledge through dynamic routing between retrieval and verbalization.

Abstract: Selective retrieval improves the accuracy and efficiency of retrieval-augmented generation (RAG) by reducing distractions from low-quality retrievals. However, existing approaches underutilize the inherent knowledge of large language models (LLMs), leading to suboptimal retrieval decisions and degraded generation performance. To bridge this gap, we propose Self-Routing RAG (SR-RAG), a novel framework that binds selective retrieval with knowledge verbalization. SR-RAG enables an LLM to dynamically decide whether to retrieve external knowledge or verbalize its own parametric knowledge. To this end, we design a multi-task objective that jointly optimizes an LLM for knowledge source selection, knowledge verbalization, and response generation. SR-RAG further incorporates a nearest neighbor search mechanism at inference time to improve the accuracy of knowledge source decisions under domain shifts. Fine-tuning three LLMs with SR-RAG significantly improves both their response accuracy and reduces the inference latency. Compared to the strongest selective retrieval baseline, SR-RAG reduces the number of retrievals by 29% while improving performance by 5.1%.

Method: Two-stage training: (1) supervised fine-tuning with synthetic traces to teach multi-stage ex-ante reasoning, and (2) step-level reasoning preference optimization to jointly enhance safety, utility, and efficiency.

Result: Experiments on multiple open-source LLMs show SAFER significantly enhances safety performance while maintaining helpfulness and response efficiency.

Conclusion: SAFER provides an effective framework for transparent and verifiable safety alignment in LLMs through structured ex-ante reasoning and multi-stage training.

Abstract: Recent advancements in large language models (LLMs) have accelerated progress toward artificial general intelligence, yet their potential to generate harmful content poses critical safety challenges. Existing alignment methods often struggle to cover diverse safety scenarios and remain vulnerable to adversarial attacks. In this work, we propose SAFER, a framework for Safety Alignment via eFficient Ex-Ante Reasoning. Our approach instantiates structured Ex-Ante reasoning through initial assessment, rule verification, and path calibration, and embeds predefined safety rules to provide transparent and verifiable safety judgments. Specifically, our approach consists of two training stages: (1) supervised fine-tuning with synthetic traces to teach the multi-stage Ex-Ante reasoning, and (2) step-level reasoning preference optimization to jointly enhance safety, utility, and efficiency. Experiments on multiple open-source LLMs demonstrate that SAFER significantly enhances safety performance while maintaining helpfulness and response efficiency.

Method: Created MedHal dataset by incorporating diverse medical text sources and tasks, providing large volume of annotated samples with explanations for factual inconsistencies.

Result: Trained baseline medical hallucination detection model showed improvements over general-purpose hallucination detection approaches.

Conclusion: MedHal enables more efficient evaluation of medical text generation systems, reduces reliance on expert review, and accelerates medical AI research.

Abstract: We present MedHal, a novel large-scale dataset specifically designed to evaluate if models can detect hallucinations in medical texts. Current hallucination detection methods face significant limitations when applied to specialized domains like medicine, where they can have disastrous consequences. Existing medical datasets are either too small, containing only a few hundred samples, or focus on a single task like Question Answering or Natural Language Inference. MedHal addresses these gaps by: (1) incorporating diverse medical text sources and tasks; (2) providing a substantial volume of annotated samples suitable for training medical hallucination detection models; and (3) including explanations for factual inconsistencies to guide model learning. We demonstrate MedHal’s utility by training and evaluating a baseline medical hallucination detection model, showing improvements over general-purpose hallucination detection approaches. This resource enables more efficient evaluation of medical text generation systems while reducing reliance on costly expert review, potentially accelerating the development of medical AI research.

Method: Proposes harmonic mean of fraction of unseen n-grams and task-specific quality score. Evaluates three open-data models (OLMo, OLMo-2, Pythia) on creative tasks: story completion, poetry writing, and creative tool use.

Result: Base LLM text is less novel than human-written text; model scale and post-training improve novelty via quality improvements; better base models at same scale improve novelty via higher originality; inference-time methods have small effect, often trading quality for originality.

Conclusion: Highlights need for better elicitation strategies for creative applications, as current methods show limited effectiveness in improving both originality and quality simultaneously.

Abstract: As large language models (LLMs) are increasingly used for ideation and scientific discovery, it is important to evaluate their ability to generate novel output. Prior work evaluates novelty as originality with respect to model training data, but original outputs may be of low quality. In contrast, non-expert judges more reliably score quality but may favor memorized outputs, limiting the reliability of human preference as a metric. We introduce a new novelty metric for LLM generations that balances originality and quality – the harmonic mean of the fraction of \ngrams unseen during training and a task-specific quality score. Using this framework, we identify trends that affect the novelty of generations from three families of open-data models (OLMo, OLMo-2, and Pythia) on three creative tasks: story completion, poetry writing, and creative tool use. We find that model-generated text from some base LLMs is less novel than human-written text from the internet. However, increasing model scale and post-training reliably improves novelty due to improvements in output quality. We also find that improving the base model at the same scale (\eg OLMo 7B to OLMo-2 7B) leads to higher novelty due to higher originality. Finally, we observe that inference-time methods, such as prompting and providing novel in-context examples, have a much smaller effect on novelty, often increasing originality at the expense of quality. This highlights the need for further research into more effective elicitation strategies as we use models for creative applications.

Method: Introduced spurious improvement methods and evaluated them against contrastive decoding techniques to reveal the true source of performance gains.

Result: Experimental results show that performance improvements from contrastive decoding are entirely unrelated to mitigating hallucinations, being driven by misleading factors instead.

Conclusion: Current contrastive decoding strategies are ineffective for hallucination mitigation, challenging common assumptions and calling for genuinely effective solutions.

Abstract: Contrastive decoding strategies are widely used to reduce object hallucinations in multimodal large language models (MLLMs). These methods work by constructing contrastive samples to induce hallucinations and then suppressing them in the output distribution. However, this paper demonstrates that such approaches fail to effectively mitigate the hallucination problem. The performance improvements observed on POPE Benchmark are largely driven by two misleading factors: (1) crude, unidirectional adjustments to the model’s output distribution and (2) the adaptive plausibility constraint, which reduces the sampling strategy to greedy search. To further illustrate these issues, we introduce a series of spurious improvement methods and evaluate their performance against contrastive decoding techniques. Experimental results reveal that the observed performance gains in contrastive decoding are entirely unrelated to its intended goal of mitigating hallucinations. Our findings challenge common assumptions about the effectiveness of contrastive decoding strategies and pave the way for developing genuinely effective solutions to hallucinations in MLLMs.

Method: Proposes an innovative method integrating RST-enhanced graph fusion with interpretability-aware prediction. Uses RST within heterogeneous graph neural networks for cross-document context modeling, and structure-aware semantic alignment based on lexical chains for cross-lingual understanding. Includes an EDU-level attribution framework for interpretability.

Result: Extensive experiments show superior performance with significant improvements over conventional NLI models and large language models. Created a high-quality CDCL-NLI dataset with 25,410 instances spanning 26 languages.

Conclusion: The work advances NLI research and will stimulate interest in cross-document cross-lingual context understanding, hallucination elimination, and interpretability inference. Code and datasets are publicly available.

Abstract: Natural Language Inference (NLI) is a fundamental task in natural language processing. While NLI has developed many sub-directions such as sentence-level NLI, document-level NLI and cross-lingual NLI, Cross-Document Cross-Lingual NLI (CDCL-NLI) remains largely unexplored. In this paper, we propose a novel paradigm: CDCL-NLI, which extends traditional NLI capabilities to multi-document, multilingual scenarios. To support this task, we construct a high-quality CDCL-NLI dataset including 25,410 instances and spanning 26 languages. To address the limitations of previous methods on CDCL-NLI task, we further propose an innovative method that integrates RST-enhanced graph fusion with interpretability-aware prediction. Our approach leverages RST (Rhetorical Structure Theory) within heterogeneous graph neural networks for cross-document context modeling, and employs a structure-aware semantic alignment based on lexical chains for cross-lingual understanding. For NLI interpretability, we develop an EDU (Elementary Discourse Unit)-level attribution framework that produces extractive explanations. Extensive experiments demonstrate our approach’s superior performance, achieving significant improvements over both conventional NLI models as well as large language models. Our work sheds light on the study of NLI and will bring research interest on cross-document cross-lingual context understanding, hallucination elimination and interpretability inference. Our code and datasets are available at “https://github.com/Leonardo123-ui/CDCL_NLI" for peer review.

Method: Used the Pythia scaling suite with interim checkpoints to systematically investigate ICL performance on downstream tasks and conduct mechanistic analysis of the residual stream’s subspace.

Result: ICL extends beyond mere memorization of training data but doesn’t amount to independent symbolic algorithm implementation. Clarified training dynamics, model capabilities, and mechanistic interpretability aspects.

Conclusion: Advances understanding of ICL, providing insights for model developers on potential improvements and basis for AI security practitioners to develop more informed guidelines.

Abstract: Large-scale Transformer language models (LMs) trained solely on next-token prediction with web-scale data can solve a wide range of tasks after seeing just a few examples. The mechanism behind this capability, known as in-context learning (ICL), remains both controversial and poorly understood. Some studies argue that it is merely the result of memorizing vast amounts of data, while others contend that it reflects a fundamental, symbolic algorithmic development in LMs. In this work, we introduce a suite of investigative tasks and a novel method to systematically investigate ICL by leveraging the full Pythia scaling suite, including interim checkpoints that capture progressively larger amount of training data. By carefully exploring ICL performance on downstream tasks and simultaneously conducting a mechanistic analysis of the residual stream’s subspace, we demonstrate that ICL extends beyond mere “memorization” of the training corpus, yet does not amount to the implementation of an independent symbolic algorithm. Our results also clarify several aspects of ICL, including the influence of training dynamics, model capabilities, and elements of mechanistic interpretability. Overall, our work advances the understanding of ICL and its implications, offering model developers insights into potential improvements and providing AI security practitioners with a basis for more informed guidelines.

Method: Proposed requirements-aware prompt optimization mechanisms and advocated for systematic proactive requirements discovery, evaluation, and monitoring processes.

Result: LLMs infer unspecified requirements by default 41.1% of the time, but under-specified prompts are 2x more likely to regress with accuracy drops sometimes exceeding 20%. Requirements-aware optimization improved performance by 4.8% over baselines.

Conclusion: A systematic approach to requirements management through proactive discovery, evaluation, and monitoring is needed to better handle prompt underspecification in practice.

Abstract: Prompt underspecification is a common challenge when interacting with LLMs. In this paper, we present an in-depth analysis of this problem, showing that while LLMs can often infer unspecified requirements by default (41.1%), such behavior is fragile: Under-specified prompts are 2x as likely to regress across model or prompt changes, sometimes with accuracy drops exceeding 20%. This instability makes it difficult to reliably build LLM applications. Moreover, simply specifying all requirements does not consistently help, as models have limited instruction-following ability and requirements can conflict. Standard prompt optimizers likewise provide little benefit. To address these issues, we propose requirements-aware prompt optimization mechanisms that improve performance by 4.8% on average over baselines. We further advocate for a systematic process of proactive requirements discovery, evaluation, and monitoring to better manage prompt underspecification in practice.

Method: Combines multi-level contrastive learning with multi-task auxiliary classification to learn subtle stylistic cues. Models LLM families as distinct stylistic entities and includes adaptation for distributional shifts without retraining for unseen data.

Result: FAID outperforms several baselines, particularly enhancing generalization accuracy on unseen domains and new LLMs.

Conclusion: FAID offers a potential solution for improving transparency and accountability in AI-assisted writing through fine-grained detection capabilities.

Abstract: The growing collaboration between humans and AI models in generative tasks has introduced new challenges in distinguishing between human-written, LLM-generated, and human–LLM collaborative texts. In this work, we collect a multilingual, multi-domain, multi-generator dataset FAIDSet. We further introduce a fine-grained detection framework FAID to classify text into these three categories, and also to identify the underlying LLM family of the generator. Unlike existing binary classifiers, FAID is built to capture both authorship and model-specific characteristics. Our method combines multi-level contrastive learning with multi-task auxiliary classification to learn subtle stylistic cues. By modeling LLM families as distinct stylistic entities, we incorporate an adaptation to address distributional shifts without retraining for unseen data. Our experimental results demonstrate that FAID outperforms several baselines, particularly enhancing the generalization accuracy on unseen domains and new LLMs, thus offering a potential solution for improving transparency and accountability in AI-assisted writing.

Method: Uses syllogistic reasoning as a well-defined logic fragment, constructs controlled datasets for premise selection, and applies few-shot meta-learning to help models extract abstract inference patterns rather than memorize task-specific patterns.

Result: Small models (1.5B-7B) fine-tuned with meta-learning show strong gains in generalization, especially in low-data regimes, and outperform GPT-4o and o3-mini on syllogistic reasoning tasks.

Conclusion: Meta-learning is effective for enabling LLMs to acquire abstract logical inference patterns that generalize to novel structures, demonstrating its potential for improving logical reasoning capabilities.

Abstract: Large language models (LLMs) are increasingly evaluated on reasoning tasks, yet their logical abilities remain contested. To address this, we study LLMs’ reasoning in a well-defined fragment of logic: syllogistic reasoning. We cast the problem as premise selection and construct controlled datasets to isolate logical competence. Beyond evaluation, an open challenge is enabling LLMs to acquire abstract inference patterns that generalize to novel structures. We propose to apply few-shot meta-learning to this domain, thereby encouraging models to extract rules across tasks rather than memorize patterns within tasks. Although meta-learning has been little explored in the context of logic learnability, our experiments show that it is effective: small models (1.5B-7B) fine-tuned with meta-learning demonstrate strong gains in generalization, with especially pronounced benefits in low-data regimes. These meta-learned models outperform GPT-4o and o3-mini on our syllogistic reasoning task.

Method: Proposed a unifying framework based on intermediate representations, systematically evaluated over a dozen pipelines including novel ones with syntactically similar but high-resource intermediate languages like Python wrappers of PDDL.

Result: Provided recipes for planning language generation pipelines, demonstrated efficacy of various components, and showed robustness against problem complexity.

Conclusion: The framework successfully unifies LLM-as-formalizer methodology, providing systematic evaluation and evidence of robustness across different planning language generation approaches.

Abstract: A line of work in planning uses LLM not to generate a plan, but to generate a formal representation in some planning language, which can be input into a symbolic solver to deterministically find a plan. While showing improved trust and promising performance, dozens of recent publications have proposed scattered methods on a variety of benchmarks under different experimental settings. We attempt to unify the inference-time LLM-as-formalizer methodology for classical planning by proposing a unifying framework based on intermediate representations. We thus systematically evaluate more than a dozen pipelines that subsume most existing work, while proposing novel ones that involve syntactically similar but high resource intermediate languages (such as a Python wrapper of PDDL). We provide recipes for planning language generation pipelines, draw a series of conclusions showing the efficacy of their various components, and evidence their robustness against problem complexity.

Method: Analyzed factual recall and crosslingual consistency evolution during pretraining using OLMo-7B as a case study, examining fact frequency effects and crosslingual transfer mechanisms.

Result: Both accuracy and consistency improve over time for most languages, primarily driven by fact frequency. Crosslingual transfer helps recall low-frequency facts in non-English languages, especially in early pretraining stages.

Conclusion: Multilingual factual knowledge acquisition occurs through two pathways: dominant frequency-driven learning (language-agnostic) and limited crosslingual transfer (mainly for named entity relations).

Abstract: Large Language Models (LLMs) are capable of recalling multilingual factual knowledge present in their pretraining data. However, most studies evaluate only the final model, leaving the development of factual recall and crosslingual consistency throughout pretraining largely unexplored. In this work, we trace how factual recall and crosslingual consistency evolve during pretraining, focusing on OLMo-7B as a case study. We find that both accuracy and consistency improve over time for most languages. We show that this improvement is primarily driven by the fact frequency in the pretraining corpus: more frequent facts are more likely to be recalled correctly, regardless of language. Yet, some low-frequency facts in non-English languages can still be correctly recalled. Our analysis reveals that these instances largely benefit from crosslingual transfer of their English counterparts – an effect that emerges predominantly in the early stages of pretraining. We pinpoint two distinct pathways through which multilingual factual knowledge acquisition occurs: (1) frequency-driven learning, which is dominant and language-agnostic, and (2) crosslingual transfer, which is limited in scale and typically constrained to relation types involving named entities. We release our code and data to facilitate further research at https://github.com/cisnlp/multilingual-fact-tracing.

Method: Propose ChartCards framework that systematically synthesizes chart information including data tables, visualization code, visual elements, and multi-dimensional semantic captions into organized metadata.

Result: Constructed MetaChart dataset with 10,862 data tables, 85K charts, and 170K captions. Fine-tuning 6 models on MetaChart achieved average 5% performance improvement across tasks, with 17% and 28% improvements in text-to-chart retrieval and chart-to-table tasks respectively.

Conclusion: ChartCards enables efficient multi-task chart understanding by providing structured metadata that supports various downstream tasks, significantly reducing data requirements while improving model performance.

Abstract: The emergence of Multi-modal Large Language Models (MLLMs) presents new opportunities for chart understanding. However, due to the fine-grained nature of these tasks, applying MLLMs typically requires large, high-quality datasets for task-specific fine-tuning, leading to high data collection and training costs. To address this, we propose ChartCards, a unified chart-metadata generation framework for multi-task chart understanding. ChartCards systematically synthesizes various chart information, including data tables, visualization code, visual elements, and multi-dimensional semantic captions. By structuring this information into organized metadata, ChartCards enables a single chart to support multiple downstream tasks, such as text-to-chart retrieval, chart summarization, chart-to-table conversion, chart description, and chart question answering. Using ChartCards, we further construct MetaChart, a large-scale high-quality dataset containing 10,862 data tables, 85K charts, and 170 K high-quality chart captions. We validate the dataset through qualitative crowdsourcing evaluations and quantitative fine-tuning experiments across various chart understanding tasks. Fine-tuning six different models on MetaChart resulted in an average performance improvement of 5% across all tasks. The most notable improvements are seen in text-to-chart retrieval and chart-to-table tasks, with Long-CLIP and Llama 3.2-11B achieving improvements of 17% and 28%, respectively.

Method: Created AVerImaTeC dataset with real-world claims, annotated with QA evidence pairs from web sources. Used claim normalization, temporally constrained evidence annotation, and two-stage sufficiency checks to mitigate common challenges.

Result: Achieved high annotation consistency with κ=0.742 on verdicts and 74.7% consistency on QA pairs. Established baselines for verifying image-text claims using open-web evidence.

Conclusion: AVerImaTeC provides a comprehensive dataset for image-text claim verification with evidence-based reasoning, addressing key challenges in fact-checking research.

Abstract: Textual claims are often accompanied by images to enhance their credibility and spread on social media, but this also raises concerns about the spread of misinformation. Existing datasets for automated verification of image-text claims remain limited, as they often consist of synthetic claims and lack evidence annotations to capture the reasoning behind the verdict. In this work, we introduce AVerImaTeC, a dataset consisting of 1,297 real-world image-text claims. Each claim is annotated with question-answer (QA) pairs containing evidence from the web, reflecting a decomposed reasoning regarding the verdict. We mitigate common challenges in fact-checking datasets such as contextual dependence, temporal leakage, and evidence insufficiency, via claim normalization, temporally constrained evidence annotation, and a two-stage sufficiency check. We assess the consistency of the annotation in AVerImaTeC via inter-annotator studies, achieving a $\kappa=0.742$ on verdicts and $74.7%$ consistency on QA pairs. We also propose a novel evaluation method for evidence retrieval and conduct extensive experiments to establish baselines for verifying image-text claims using open-web evidence.

Method: Introduces v1 with point-and-copy approach that identifies relevant image patches and copies their embeddings into reasoning stream, using semantic representations as keys.

Result: v1 consistently outperforms comparable baselines across various multimodal mathematical reasoning benchmarks.

Conclusion: Point-and-copy is a practical mechanism for grounded reasoning that keeps perceptual evidence embedded in the same space as the model’s reasoning.

Abstract: When thinking with images, humans rarely rely on a single glance: they revisit visual information repeatedly during reasoning. However, existing models typically process images only once and thereafter generate reasoning entirely in text, lacking mechanisms to re-access or ground inference in visual representations. We empirically confirm this: as reasoning chains lengthen, models progressively lose focus on relevant regions. In response, we introduce v1, a lightweight extension that enables active visual referencing through a simple point-and-copy approach. This allows the model to identify relevant image patches and copy their embeddings back into the reasoning stream, ensuring that evolving hypotheses remain grounded in perceptual evidence. Crucially, our pointing strategy lets the MLLM directly select image patches using their semantic representations as keys, keeping perceptual evidence embedded in the same space as the model’s reasoning. To train this capability, we construct v1g, a dataset of 300K multimodal reasoning traces with interleaved visual grounding annotations. Across various multimodal mathematical reasoning benchmarks, v1 consistently outperforms comparable baselines, establishing point-and-copy as a practical mechanism for grounded reasoning. The model checkpoint and dataset are available at github.com/jun297/v1.

Method: Constructed an extended-refusal dataset with detailed justifications before refusal, then fine-tuned models (Llama-2-7B-Chat and Qwen2.5-Instruct) on this dataset to distribute refusal signals across multiple token positions.

Result: Models maintained high refusal rates under abliteration (dropped by at most 10% vs 70-80% drops in baselines), effectively neutralizing attacks while preserving general performance and enhancing robustness.

Conclusion: Extended-refusal fine-tuning provides an effective defense against abliteration attacks by fundamentally altering how models express refusal, maintaining safety without compromising utility.

Abstract: Large language models (LLMs) are typically aligned to refuse harmful instructions through safety fine-tuning. A recent attack, termed abliteration, identifies and suppresses the single latent direction most responsible for refusal behavior, thereby enabling models to generate harmful content. We propose a defense that fundamentally alters how models express refusal. We construct an extended-refusal dataset in which responses to harmful prompts provide detailed justifications before refusing, distributing the refusal signal across multiple token positions. Fine-tuning Llama-2-7B-Chat and Qwen2.5-Instruct (1.5B and 3B parameters) on this dataset yields models that maintain high refusal rates under abliteration: refusal rates drop by at most 10%, compared to 70-80% drops in baseline models. Comprehensive evaluations of safety and utility demonstrate that extended-refusal fine-tuning effectively neutralizes abliteration attacks while preserving general model performance and enhancing robustness across multiple alignment scenarios.

Method: Categorizes recent work into five directions: trajectory prediction via language modeling paradigms, direct prediction with pretrained LLMs, language-guided scene understanding, language-driven data generation, and language-based reasoning and interpretability.

Result: Provides a unified perspective on how language can enrich trajectory prediction by bridging natural language processing and trajectory prediction fields.

Conclusion: This survey offers comprehensive analysis of representative methods, highlights core design choices, and identifies open challenges in this emerging field of LLM-enhanced trajectory prediction.

Abstract: Recent advances in large language models (LLMs) have sparked growing interest in integrating language-driven techniques into trajectory prediction. By leveraging their semantic and reasoning capabilities, LLMs are reshaping how autonomous systems perceive, model, and predict trajectories. This survey provides a comprehensive overview of this emerging field, categorizing recent work into five directions: (1) Trajectory prediction via language modeling paradigms, (2) Direct trajectory prediction with pretrained language models, (3) Language-guided scene understanding for trajectory prediction, (4) Language-driven data generation for trajectory prediction, (5) Language-based reasoning and interpretability for trajectory prediction. For each, we analyze representative methods, highlight core design choices, and identify open challenges. This survey bridges natural language processing and trajectory prediction, offering a unified perspective on how language can enrich trajectory prediction.

Method: Used OWL dataset with 31.5K aligned excerpts from 20 books in 10 languages, evaluating through three tasks: direct probing (identify title/author), name cloze (predict masked names), and prefix probing (generate continuations). Tested across model families and sizes with perturbations.

Result: LLMs consistently recalled content across languages, with GPT-4o identifying authors/titles 69% of the time and masked entities 6% of the time in newly translated excerpts. Perturbations only modestly reduced accuracy (7% drop for shuffled official translations).

Conclusion: LLMs exhibit significant cross-lingual memorization capabilities that extend beyond their direct pretraining data, highlighting the transferability of memorized knowledge across languages.

Abstract: Large language models (LLMs) are known to memorize and recall English text from their pretraining data. However, the extent to which this ability generalizes to non-English languages or transfers across languages remains unclear. This paper investigates multilingual and cross-lingual memorization in LLMs, probing if memorized content in one language (e.g., English) can be recalled when presented in translation. To do so, we introduce OWL, a dataset of 31.5K aligned excerpts from 20 books in ten languages, including English originals, official translations (Vietnamese, Spanish, Turkish), and new translations in six low-resource languages (Sesotho, Yoruba, Maithili, Malagasy, Setswana, Tahitian). We evaluate memorization across model families and sizes through three tasks: (1) direct probing, which asks the model to identify a book’s title and author; (2) name cloze, which requires predicting masked character names; and (3) prefix probing, which involves generating continuations. We find that LLMs consistently recall content across languages, even for texts without direct translation in pretraining data. GPT-4o, for example, identifies authors and titles 69% of the time and masked entities 6% of the time in newly translated excerpts. Perturbations (e.g., masking characters, shuffling words) modestly reduce direct probing accuracy (7% drop for shuffled official translations). Our results highlight the extent of cross-lingual memorization and provide insights on the differences between the models.

Method: KGQAGen combines structured knowledge grounding, LLM-guided generation, and symbolic verification to systematically produce challenging and verifiable QA instances.

Result: KGQAGen-10k, a ten-thousand scale benchmark grounded in Wikidata, exposes limitations of state-of-the-art KG-RAG models, demonstrating their struggles on this more rigorous benchmark.

Conclusion: The framework advocates for more rigorous benchmark construction and positions KGQAGen as a scalable solution for advancing KGQA evaluation by addressing quality pitfalls in existing datasets.

Abstract: Knowledge Graph Question Answering (KGQA) systems rely on high-quality benchmarks to evaluate complex multi-hop reasoning. However, despite their widespread use, popular datasets such as WebQSP and CWQ suffer from critical quality issues, including inaccurate or incomplete ground-truth annotations, poorly constructed questions that are ambiguous, trivial, or unanswerable, and outdated or inconsistent knowledge. Through a manual audit of 16 popular KGQA datasets, including WebQSP and CWQ, we find that the average factual correctness rate is only 57 %. To address these issues, we introduce KGQAGen, an LLM-in-the-loop framework that systematically resolves these pitfalls. KGQAGen combines structured knowledge grounding, LLM-guided generation, and symbolic verification to produce challenging and verifiable QA instances. Using KGQAGen, we construct KGQAGen-10k, a ten-thousand scale benchmark grounded in Wikidata, and evaluate a diverse set of KG-RAG models. Experimental results demonstrate that even state-of-the-art systems struggle on this benchmark, highlighting its ability to expose limitations of existing models. Our findings advocate for more rigorous benchmark construction and position KGQAGen as a scalable framework for advancing KGQA evaluation.

Method: Created ExpertLongBench with 11 tasks from 9 domains, each with expert-validated rubrics. Proposed CLEAR evaluation framework that extracts checklists from model outputs and references based on task rubrics, then compares checklist items for correctness assessment.

Result: Benchmarked 13 LLMs - top performer Gemini-2.5-Pro achieved only 33.4 F1 score, showing models can generate content for required aspects but are far from correct. CLEAR’s checklist extraction and comparison can be achieved by open-weight models for scalable and cost-effective evaluation.

Conclusion: Existing LLMs require significant improvement for expert-level tasks. The CLEAR framework enables accurate, fine-grained evaluation of long-form outputs and can be implemented using open-weight models for practical deployment.

Abstract: This paper introduces ExpertLongBench, an expert-level benchmark containing 11 tasks from 9 domains that reflect realistic expert workflows and applications. Beyond question answering, the application-driven tasks in ExpertLongBench demand long-form outputs that can exceed 5,000 tokens and strict adherence to domain-specific requirements. Notably, each task in ExpertLongBench includes a rubric, designed or validated by domain experts, to specify task requirements and guide output evaluation. Furthermore, we propose CLEAR, an evaluation framework that supports accurate evaluation of long-form model outputs in our benchmark. To achieve fine-grained, expert-aligned evaluation, CLEAR derives checklists from both model outputs and references by extracting information corresponding to items in the task-specific rubric. Checklist items of model outputs are then compared with corresponding items of reference outputs to assess their correctness, enabling grounded evaluation. We benchmark 13 popular large language models (LLMs) and analyze components in CLEAR, showing that (1) existing LLMs, with the top performer Gemini-2.5-Pro achieving only a 33.4 F1 score, require significant improvement for expert-level tasks; (2) models can generate content corresponding to the required aspects, but far from correct; and (3) accurate checklist extraction and comparison in CLEAR can be achieved by open-weight models for more scalable, reproducible, and low-cost usage.

Method: Proposed GraphRAG-Bench with comprehensive dataset covering tasks of increasing difficulty (fact retrieval, complex reasoning, contextual summarization, creative generation) and systematic evaluation across the entire pipeline from graph construction to final generation.

Result: The benchmark enables systematic investigation of conditions when GraphRAG surpasses traditional RAG and the underlying reasons for its success.

Conclusion: GraphRAG-Bench offers guidelines for practical application of GraphRAG and provides resources for the community to advance research in this area.

Abstract: Graph retrieval-augmented generation (GraphRAG) has emerged as a powerful paradigm for enhancing large language models (LLMs) with external knowledge. It leverages graphs to model the hierarchical structure between specific concepts, enabling more coherent and effective knowledge retrieval for accurate reasoning.Despite its conceptual promise, recent studies report that GraphRAG frequently underperforms vanilla RAG on many real-world tasks. This raises a critical question: Is GraphRAG really effective, and in which scenarios do graph structures provide measurable benefits for RAG systems? To address this, we propose GraphRAG-Bench, a comprehensive benchmark designed to evaluate GraphRAG models onboth hierarchical knowledge retrieval and deep contextual reasoning. GraphRAG-Bench features a comprehensive dataset with tasks of increasing difficulty, coveringfact retrieval, complex reasoning, contextual summarization, and creative generation, and a systematic evaluation across the entire pipeline, from graph constructionand knowledge retrieval to final generation. Leveraging this novel benchmark, we systematically investigate the conditions when GraphRAG surpasses traditional RAG and the underlying reasons for its success, offering guidelines for its practical application. All related resources and analyses are collected for the community at https://github.com/GraphRAG-Bench/GraphRAG-Benchmark.

Method: Systematic review and formalization of compound AI system optimization, classification of existing methods along key dimensions including traditional optimization (SFT, RL) and emerging natural language feedback approaches.

Result: Comprehensive survey that organizes the field, identifies key optimization techniques for non-differentiable systems, and provides a structured framework for understanding compound AI system optimization.

Conclusion: The paper establishes a foundation for future research in compound AI system optimization, highlighting promising directions and open challenges in this rapidly evolving field.

Abstract: Recent advancements in large language models (LLMs) and AI systems have led to a paradigm shift in the design and optimization of complex AI workflows. By integrating multiple components, compound AI systems have become increasingly adept at performing sophisticated tasks. However, as these systems grow in complexity, new challenges arise in optimizing not only individual components but also their interactions. While traditional optimization methods such as supervised fine-tuning (SFT) and reinforcement learning (RL) remain foundational, the rise of natural language feedback introduces promising new approaches, especially for optimizing non-differentiable systems. This paper provides a systematic review of recent progress in optimizing compound AI systems, encompassing both numerical and language-based techniques. We formalize the notion of compound AI system optimization, classify existing methods along several key dimensions, and highlight open research challenges and future directions in this rapidly evolving field. A list of surveyed papers is publicly available at https://github.com/MiuLab/AISysOpt-Survey.

Method: A method that allows users to add corrections on the fly during inference to fix substitution errors in challenging words with pronunciation-orthography mismatches.

Result: The method achieves up to 8% relative improvement in biased word error rate while maintaining competitive overall word error rate.

Conclusion: On-the-fly correction during inference effectively improves recognition accuracy for challenging words with pronunciation-orthography mismatches without compromising overall performance.

Abstract: Neural sequence-to-sequence systems deliver state-of-the-art performance for automatic speech recognition. When using appropriate modeling units, e.g., byte-pair encoded characters, these systems are in principal open vocabulary systems. In practice, however, they often fail to recognize words not seen during training, e.g., named entities, acronyms, or domain-specific special words. To address this problem, many context biasing methods have been proposed; however, for words with a pronunciation-orthography mismatch, these methods may still struggle. We propose a method which allows corrections of substitution errors to improve the recognition accuracy of such challenging words. Users can add corrections on the fly during inference. We show that with this method we get a relative improvement in biased word error rate of up to 8%, while maintaining a competitive overall word error rate.

Method: Evaluated various monitoring methods using CoT text or activations, including large language models, fine-tuned classifiers, and humans. Used linear probes trained on CoT activations to predict response safety.

Result: Linear probe on CoT activations significantly outperformed text-based baselines (13 F1 score increase). CoT texts were often unfaithful while model latents provided reliable predictive signals. Probes could detect alignment signals early in CoT segments before response generation.

Conclusion: Lightweight probes on CoT activations enable real-time safety monitoring and early intervention during generation, generalizing across model sizes and safety benchmarks.

Abstract: Reasoning language models improve performance on complex tasks by generating long chains of thought (CoTs), but this process can also increase harmful outputs in adversarial settings. In this work, we ask whether the long CoTs can be leveraged for predictive safety monitoring: do the reasoning traces provide early signals of final response alignment that could enable timely intervention? We evaluate a range of monitoring methods using either CoT text or activations, including highly capable large language models, fine-tuned classifiers, and humans. First, we find that a simple linear probe trained on CoT activations significantly outperforms all text-based baselines in predicting whether a final response is safe or unsafe, with an average absolute increase of 13 in F1 scores over the best-performing alternatives. CoT texts are often unfaithful and misleading, while model latents provide a more reliable predictive signal. Second, the probe can be applied to early CoT segments before the response is generated, showing that alignment signals appear before reasoning completes. Error analysis reveals that the performance gap between text classifiers and the linear probe largely stems from a subset of responses we call performative CoTs, where the reasoning consistently contradicts the final response as the CoT progresses. Our findings generalize across model sizes, families, and safety benchmarks, suggesting that lightweight probes could enable real-time safety monitoring and early intervention during generation.

Method: Created a dataset with synthesized intents from unannotated table corpora, benchmarked single-shot schema generation methods (prompted LLMs and fine-tuned models), and proposed LLM-based schema refinement techniques.

Result: Incorporating table intents significantly improved baseline performance in reconstructing reference schemas. Smaller open-weight models fine-tuned were competitive with state-of-the-art prompted LLMs, and refinement techniques further improved generated schemas.

Conclusion: The work successfully addresses both evaluation ambiguity and lack of refinement methods in schema generation, demonstrating that fine-tuned smaller models can compete with large prompted LLMs when enhanced with intent information and refinement techniques.

Abstract: The increasing volume of academic literature makes it essential for researchers to organize, compare, and contrast collections of documents. Large language models (LLMs) can support this process by generating schemas defining shared aspects along which to compare papers. However, progress on schema generation has been slow due to: (i) ambiguity in reference-based evaluations, and (ii) lack of editing/refinement methods. Our work is the first to address both issues. First, we present an approach for augmenting unannotated table corpora with \emph{synthesized intents}, and apply it to create a dataset for studying schema generation conditioned on a given information need, thus reducing ambiguity. With this dataset, we show how incorporating table intents significantly improves baseline performance in reconstructing reference schemas. We start by comprehensively benchmarking several single-shot schema generation methods, including prompted LLM workflows and fine-tuned models, showing that smaller, open-weight models can be fine-tuned to be competitive with state-of-the-art prompted LLMs. Next, we propose several LLM-based schema refinement techniques and show that these can further improve schemas generated by these methods.

Method: Systematically benchmarked 11 open-access LLMs (1-12B parameters) across 3 datasets, testing 396 configurations. Fine-tuned LiTransMC using parameter efficient and loss masking strategy. Evaluated reasoning using BERTopic and novel Explanation Strength Index.

Result: LiTransMC achieved weighted F1 score of 0.6845 and Jensen-Shannon Divergence of 0.000245, surpassing untuned local models, larger proprietary systems (including GPT-4o), and classical methods like discrete choice models and ML classifiers.

Conclusion: Demonstrates feasibility of creating specialist, locally deployable LLMs that combine prediction accuracy with interpretability, establishing pathway for transforming general purpose LLMs into specialized tools for transportation research and policy while maintaining privacy and reducing costs.

Abstract: This study investigates the adoption of open-access, locally deployable causal large language models (LLMs) for travel mode choice prediction and introduces LiTransMC, the first fine-tuned causal LLM developed for this task. We systematically benchmark eleven open-access LLMs (1-12B parameters) across three stated and revealed preference datasets, testing 396 configurations and generating over 79,000 mode choice decisions. Beyond predictive accuracy, we evaluate models generated reasoning using BERTopic for topic modelling and a novel Explanation Strength Index, providing the first structured analysis of how LLMs articulate decision factors in alignment with behavioural theory. LiTransMC, fine-tuned using parameter efficient and loss masking strategy, achieved a weighted F1 score of 0.6845 and a Jensen-Shannon Divergence of 0.000245, surpassing both untuned local models and larger proprietary systems, including GPT-4o with advanced persona inference and embedding-based loading, while also outperforming classical mode choice methods such as discrete choice models and machine learning classifiers for the same dataset. This dual improvement, i.e., high instant-level accuracy and near-perfect distributional calibration, demonstrates the feasibility of creating specialist, locally deployable LLMs that integrate prediction and interpretability. Through combining structured behavioural prediction with natural language reasoning, this work unlocks the potential for conversational, multi-task transport models capable of supporting agent-based simulations, policy testing, and behavioural insight generation. These findings establish a pathway for transforming general purpose LLMs into specialized and explainable tools for transportation research and policy formulation, while maintaining privacy, reducing cost, and broadening access through local deployment.

Method: Introduces GLiDRE, a new compact model designed for document-level relation extraction that works efficiently in both supervised and few-shot settings.

Result: Outperforms existing methods in both low-resource supervised training and few-shot meta-learning benchmarks, establishing new state-of-the-art in few-shot document-level relation extraction.

Conclusion: GLiDRE is an effective solution for document-level relation extraction in data-constrained scenarios, with code to be made publicly available.

Abstract: Relation Extraction (RE) is a fundamental task in Natural Language Processing, and its document-level variant poses significant challenges, due to complex interactions between entities across sentences. While supervised models have achieved strong results in fully resourced settings, their behavior with limited training data remains insufficiently studied. We introduce GLiDRE, a new compact model for document-level relation extraction, designed to work efficiently in both supervised and few-shot settings. Experiments in both low-resource supervised training and few-shot meta-learning benchmarks show that our approach outperforms existing methods in data-constrained scenarios, establishing a new state-of-the-art in few-shot document-level relation extraction. Our code will be publicly available.

Method: Introduces CRAFT benchmark with paired edits for composite reasoning and evaluates alias portability, temporal and common-sense locality. Proposes KEDAS paradigm with diverse edit augmentation and self-adaptive post-alignment inference.

Result: CRAFT proves challenging for previous knowledge editing methods, while KEDAS shows significant performance gain on CRAFT compared to previous methods.

Conclusion: CRAFT provides a comprehensive benchmark for knowledge editing evaluation, and KEDAS demonstrates effective real-time editing capabilities through its novel alignment paradigm.

Abstract: Knowledge editing aims to modify outdated knowledge in large language models (LLMs) efficiently while retaining their original capabilities. Mainstream benchmarks for knowledge editing are predominantly static and fail to keep in pace with the evolving real-world knowledge. In this work, we introduce CRAFT, an ever-evolving real-world benchmark for knowledge editing. It features well-designed paired edits for composite reasoning, and evaluates models on alias portability as well as temporal and common-sense locality, making it a challenging knowledge editing benchmark on which previous knowledge editing methods hardly achieve balanced performance. Towards flexible real-time editing, we propose KEDAS, a novel paradigm of knowledge editing alignment featuring diverse edit augmentation and self-adaptive post-alignment inference, which exhibits significant performance gain on CRAFT compared to previous methods. All of our code and data are available at https://anonymous.4open.science/r/CRAFT-KEDAS.

Method: Establishes MoE layers as mixtures of micro-experts, presents CAMERA framework for identifying redundancy, and proposes CAMERA-P for structured micro-expert pruning and CAMERA-Q for mixed-precision quantization.

Result: CAMERA-P outperforms baselines under 20-60% pruning ratios, CAMERA-Q achieves superior results under aggressive 2-bit quantization, and enables complete micro-expert analysis of Qwen2-57B-A14B in <5 minutes on single A100 GPU.

Conclusion: Micro-expert approach provides effective compression for MoE LLMs with significant efficiency gains while maintaining strong performance across diverse tasks.

Abstract: Large Language Models (LLMs) with Mixture-of-Experts (MoE) architectures are distinguished by their strong performance scaling with increasing parameters across a wide range of tasks, yet they also suffer from substantial computational and storage overheads. Notably, the performance gains of MoE models do not scale proportionally with the growth in expert parameters. While prior works attempt to reduce parameters via expert-level pruning, merging, or decomposition, they still suffer from challenges in both performance and computational efficiency. In this paper, we address these challenges by introducing micro-expert as a finer-grained compression unit that spans across matrices. We first establish a more fundamental perspective, viewing MoE layers as mixtures of micro-experts, and present CAMERA, a lightweight and training-free framework for identifying micro-expert redundancy. Our analysis uncovers significant variance in micro-expert contributions during decoding. Based on this insight, we further propose CAMERA-P, a structured micro-expert pruning framework, and CAMERA-Q, a mixed-precision quantization idea designed for micro-experts. Extensive experiments on nine downstream tasks show that CAMERA-P consistently outperforms strong baselines under pruning ratios ranging from 20% to 60%. Furthermore, CAMERA-Q achieves superior results under aggressive 2-bit quantization, surpassing existing matrix- and channel-level ideas. Notably, our method enables complete micro-expert analysis of Qwen2-57B-A14B in less than 5 minutes on a single NVIDIA A100-40GB GPU.

Method: Pre-trained RooseBERT on large-scale political debate and speech corpora (8K debates with multiple sub-debates) in English, then fine-tuned it on four political debate analysis tasks.

Result: RooseBERT demonstrated significant performance improvements over general-purpose language models across all four downstream tasks related to political debate analysis.

Conclusion: Domain-specific pre-training on political discourse data substantially enhances performance in political debate analysis tasks, and the model is released for research community use.

Abstract: The increasing amount of political debates and politics-related discussions calls for the definition of novel computational methods to automatically analyse such content with the final goal of lightening up political deliberation to citizens. However, the specificity of the political language and the argumentative form of these debates (employing hidden communication strategies and leveraging implicit arguments) make this task very challenging, even for current general-purpose pre-trained Language Models. To address this issue, we introduce a novel pre-trained Language Model for political discourse language called RooseBERT. Pre-training a language model on a specialised domain presents different technical and linguistic challenges, requiring extensive computational resources and large-scale data. RooseBERT has been trained on large political debate and speech corpora (8K debates, each composed of several sub-debates on different topics) in English. To evaluate its performances, we fine-tuned it on four downstream tasks related to political debate analysis, i.e., stance detection, sentiment analysis, argument component detection and classification, and argument relation prediction and classification. Our results demonstrate significant improvements over general-purpose Language Models on these four tasks, highlighting how domain-specific pre-training enhances performance in political debate analysis. We release RooseBERT for the research community.

Method: A scalable, automated approach using language models to synthesize textbook-style data through structured prompting pipeline, requiring only domain name as input to generate high-quality forget sets.

Result: Experiments on unlearning biosecurity, cybersecurity, and Harry Potter domains show synthetic datasets consistently outperform baseline alternatives and are comparable to expert-curated ones, with multi-step generation pipeline improving data diversity and unlearning utility.

Conclusion: Synthetic datasets offer a promising path toward practical, scalable unlearning for emerging domains without manual intervention, enabling effective knowledge removal from LLMs.

Abstract: Modern large language models often encode sensitive, harmful, or copyrighted knowledge, raising the need for post-hoc unlearning-the ability to remove specific domains of knowledge from a model without full retraining. A major bottleneck in current unlearning pipelines is constructing effective forget sets-datasets that approximate the target domain and guide the model to forget it. In this work, we introduce a scalable, automated approach to generate high-quality forget sets using language models themselves. Our method synthesizes textbook-style data through a structured prompting pipeline, requiring only a domain name as input. Through experiments on unlearning biosecurity, cybersecurity, and Harry Potter novels, we show that our synthetic datasets consistently outperform the baseline synthetic alternatives and are comparable to the expert-curated ones. Additionally, ablation studies reveal that the multi-step generation pipeline significantly boosts data diversity, which in turn improves unlearning utility. Overall, our findings suggest that synthetic datasets offer a promising path toward practical, scalable unlearning for a wide range of emerging domains without the need for manual intervention. We release our code and dataset at https://github.com/xyzhu123/Synthetic_Textbook.

Method: User-centric, actor-level pipeline combining discourse-aware analysis with metrics for sentiment, syntactic agency, and quotation styles for fine-grained auditing and exclusion-based balancing.

Result: Applied to taz2024full German newspaper corpus (1980-2024), yielded more gender-balanced dataset while preserving core dynamics, reducing structural asymmetries though subtler biases in sentiment and framing remain.

Conclusion: Structural asymmetries can be reduced through systematic filtering, and tools are released to support discourse-based fairness auditing and equitable corpus construction.

Abstract: Language corpora are the foundation of most natural language processing research, yet they often reproduce structural inequalities. One such inequality is gender discrimination in how actors are represented, which can distort analyses and perpetuate discriminatory outcomes. This paper introduces a user-centric, actor-level pipeline for detecting and mitigating gender discrimination in large-scale text corpora. By combining discourse-aware analysis with metrics for sentiment, syntactic agency, and quotation styles, our method enables both fine-grained auditing and exclusion-based balancing. Applied to the taz2024full corpus of German newspaper articles (1980-2024), the pipeline yields a more gender-balanced dataset while preserving core dynamics of the source material. Our findings show that structural asymmetries can be reduced through systematic filtering, though subtler biases in sentiment and framing remain. We release the tools and reports to support further research in discourse-based fairness auditing and equitable corpus construction.

Method: Created MMReview benchmark with 240 papers across 17 research domains in four academic disciplines, including multimodal content and expert-written reviews. Designed 13 tasks grouped into four categories: step-wise review generation, outcome formulation, human preference alignment, and robustness to adversarial manipulation.

Result: Extensive experiments on 16 open-source and 5 closed-source models demonstrated the benchmark’s thoroughness in evaluating model performance across different review tasks and modalities.

Conclusion: MMReview establishes a standardized foundation for developing automated peer review systems and addresses the gap in evaluating LLMs’ capabilities in multimodal academic review scenarios.

Abstract: With the rapid growth of academic publications, peer review has become an essential yet time-consuming responsibility within the research community. Large Language Models (LLMs) have increasingly been adopted to assist in the generation of review comments; however, current LLM-based review tasks lack a unified evaluation benchmark to rigorously assess the models’ ability to produce comprehensive, accurate, and human-aligned assessments, particularly in scenarios involving multimodal content such as figures and tables. To address this gap, we propose \textbf{MMReview}, a comprehensive benchmark that spans multiple disciplines and modalities. MMReview includes multimodal content and expert-written review comments for 240 papers across 17 research domains within four major academic disciplines: Artificial Intelligence, Natural Sciences, Engineering Sciences, and Social Sciences. We design a total of 13 tasks grouped into four core categories, aimed at evaluating the performance of LLMs and Multimodal LLMs (MLLMs) in step-wise review generation, outcome formulation, alignment with human preferences, and robustness to adversarial input manipulation. Extensive experiments conducted on 16 open-source models and 5 advanced closed-source models demonstrate the thoroughness of the benchmark. We envision MMReview as a critical step toward establishing a standardized foundation for the development of automated peer review systems.

Method: Introduces a framework for LLM-driven cultural localization of math word problems that automatically constructs datasets with native names, organizations, and currencies from existing sources.

Result: Translated benchmarks can obscure true multilingual math ability under appropriate socio-cultural contexts. The framework helps mitigate English-centric entity bias and improves robustness when native entities are introduced across various languages.

Conclusion: The proposed LLM-driven cultural localization framework effectively addresses the scarcity of localized datasets and helps reduce English-centric entity bias in multilingual mathematical reasoning.

Abstract: Large language models (LLMs) have demonstrated significant capabilities in solving mathematical problems expressed in natural language. However, multilingual and culturally-grounded mathematical reasoning in low-resource languages lags behind English due to the scarcity of socio-cultural task datasets that reflect accurate native entities such as person names, organization names, and currencies. Existing multilingual benchmarks are predominantly produced via translation and typically retain English-centric entities, owing to the high cost associated with human annotater-based localization. Moreover, automated localization tools are limited, and hence, truly localized datasets remain scarce. To bridge this gap, we introduce a framework for LLM-driven cultural localization of math word problems that automatically constructs datasets with native names, organizations, and currencies from existing sources. We find that translated benchmarks can obscure true multilingual math ability under appropriate socio-cultural contexts. Through extensive experiments, we also show that our framework can help mitigate English-centric entity bias and improves robustness when native entities are introduced across various languages.

Method: Framework using structured interface-specific representations and iterative refinements to translate user queries into task-specific user interfaces (UIs) for more adaptive interaction.

Result: Generative interfaces consistently outperform conversational ones with up to 72% improvement in human preference across diverse tasks, interaction patterns, and query types.

Conclusion: Generative interfaces provide superior user experience compared to traditional chat-based interfaces, clarifying when and why users prefer them and enabling future advancements in human-AI interaction.

Abstract: Large language models (LLMs) are increasingly seen as assistants, copilots, and consultants, capable of supporting a wide range of tasks through natural conversation. However, most systems remain constrained by a linear request-response format that often makes interactions inefficient in multi-turn, information-dense, and exploratory tasks. To address these limitations, we propose Generative Interfaces for Language Models, a paradigm in which LLMs respond to user queries by proactively generating user interfaces (UIs) that enable more adaptive and interactive engagement. Our framework leverages structured interface-specific representations and iterative refinements to translate user queries into task-specific UIs. For systematic evaluation, we introduce a multidimensional assessment framework that compares generative interfaces with traditional chat-based ones across diverse tasks, interaction patterns, and query types, capturing functional, interactive, and emotional aspects of user experience. Results show that generative interfaces consistently outperform conversational ones, with up to a 72% improvement in human preference. These findings clarify when and why users favor generative interfaces, paving the way for future advancements in human-AI interaction.

Method: Planner-executor-responder architecture that infers user intent, detects document language and modality, and orchestrates tools dynamically for robust, traceable reasoning while preventing tool misuse or execution loops.

Result: Outperforms baselines (ROUGE: 0.783 vs. 0.703/0.608) with better cross-lingual stability (17-point vs. 21-26-point variation).

Conclusion: The agentic system successfully addresses structural and linguistic diversity in DoP documents through dynamic tool orchestration and intent inference.

Abstract: Declaration of Performance (DoP) documents, mandated by EU regulation, certify the performance of construction products. There are two challenges to make DoPs machine and human accessible through automated key-value pair extraction (KVP) and question answering (QA): (1) While some of their content is standardized, DoPs vary widely in layout, schema, and format; (2) Both users and documents are multilingual. Existing static or LLM-only Information Extraction (IE) pipelines fail to adapt to this structural document and user diversity. Our domain-specific, agentic system addresses these challenges through a planner-executor-responder architecture. The system infers user intent, detects document language and modality, and orchestrates tools dynamically for robust, traceable reasoning while avoiding tool misuse or execution loops. Our agent outperforms baselines (ROUGE: 0.783 vs. 0.703/0.608) with better cross-lingual stability (17-point vs. 21-26-point variation).

Method: A dual-agent framework with a planner that iteratively interleaves evidence acquisition with outline optimization to create a citation-grounded outline, and a writer that performs hierarchical retrieval and section-by-section writing using targeted evidence from the memory bank.

Result: Establishes new state-of-the-art across major OEDR benchmarks including DeepResearch Bench, DeepConsult, and DeepResearchGym, demonstrating improved citation accuracy and reduced hallucinations.

Conclusion: The human-centric, iterative methodology with adaptive planning and focused synthesis is crucial for producing comprehensive, trusted, and well-structured reports in open-ended deep research tasks.

Abstract: This paper tackles \textbf{open-ended deep research (OEDR)}, a complex challenge where AI agents must synthesize vast web-scale information into insightful reports. Current approaches are plagued by dual-fold limitations: static research pipelines that decouple planning from evidence acquisition and monolithic generation paradigms that include redundant, irrelevant evidence, suffering from hallucination issues and low citation accuracy. To address these challenges, we introduce \textbf{WebWeaver}, a novel dual-agent framework that emulates the human research process. The planner operates in a dynamic cycle, iteratively interleaving evidence acquisition with outline optimization to produce a comprehensive, citation-grounded outline linking to a memory bank of evidence. The writer then executes a hierarchical retrieval and writing process, composing the report section by section. By performing targeted retrieval of only the necessary evidence from the memory bank via citations for each part, it effectively mitigates long-context issues and citation hallucinations. Our framework establishes a new state-of-the-art across major OEDR benchmarks, including DeepResearch Bench, DeepConsult, and DeepResearchGym. These results validate our human-centric, iterative methodology, demonstrating that adaptive planning and focused synthesis are crucial for producing comprehensive, trusted, and well-structured reports.

Method: Developed LLM-JEPA, a JEPA-based solution for LLMs applicable to both finetuning and pretraining, using embedding-space training objectives instead of traditional input-space reconstruction.

Result: LLM-JEPA significantly outperforms standard LLM training objectives across multiple models (Llama3, OpenELM, Gemma2, Olmo) and datasets (NL-RX, GSM8K, Spider, RottenTomatoes), while being robust to overfitting.

Conclusion: JEPA-style embedding-space objectives can be successfully adapted to language models, demonstrating superior performance over traditional input-space training methods across various models and tasks.

Abstract: Large Language Model (LLM) pretraining, finetuning, and evaluation rely on input-space reconstruction and generative capabilities. Yet, it has been observed in vision that embedding-space training objectives, e.g., with Joint Embedding Predictive Architectures (JEPAs), are far superior to their input-space counterpart. That mismatch in how training is achieved between language and vision opens up a natural question: {\em can language training methods learn a few tricks from the vision ones?} The lack of JEPA-style LLM is a testimony of the challenge in designing such objectives for language. In this work, we propose a first step in that direction where we develop LLM-JEPA, a JEPA based solution for LLMs applicable both to finetuning and pretraining. Thus far, LLM-JEPA is able to outperform the standard LLM training objectives by a significant margin across models, all while being robust to overfiting. Those findings are observed across numerous datasets (NL-RX, GSM8K, Spider, RottenTomatoes) and various models from the Llama3, OpenELM, Gemma2 and Olmo families. Code: https://github.com/rbalestr-lab/llm-jepa.

Method: Fine-tuned pre-trained and instruction-tuned models (GPT-2 and Mistral-7B-IT) using multi-label training on multiple plausible word continuations per context from the Provo Corpus.

Result: Multi-label fine-tuning improved LMs’ ability to reproduce linguistic variability across contexts with both high and low variability.

Conclusion: Training language models with data that reflects inherent linguistic variability can enhance their ability to capture human-like diversity in language generation.

Abstract: Natural language generation (NLG) tasks are often subject to inherent variability; e.g. predicting the next word given a context has multiple valid responses, evident when asking multiple humans to complete the task. While having language models (LMs) that are aligned pluralistically, so that they are able to reproduce well the inherent diversity in perspectives of an entire population of interest is clearly beneficial, Ilia and Aziz (2024) show that LMs do not reproduce this type of linguistic variability well. They speculate this inability might stem from the lack of consistent training of LMs with data reflecting this type of inherent variability. As such, we investigate whether training LMs on multiple plausible word continuations per context can improve their ability to reproduce human linguistic variability for next-word prediction. We employ fine-tuning techniques for pre-trained and instruction-tuned models; and demonstrate their potential when fine-tuning GPT-2 and Mistral-7B-IT, using Provo Corpus. Our evaluation, which measures divergence among empirically estimated human and model next-word distributions across contexts before and after fine-tuning, shows that our multi-label fine-tuning improves the LMs’ ability to reproduce linguistic variability; both for contexts that admit higher and lower variability.

Method: Analyzed the relationship between forgetting mechanisms and fairness objectives in moral alignment, examining how selective forgetting of stereotypes versus overall forgetting affects model performance.

Result: Found that: (1) downstream performance strongly correlates with overall forgetting; (2) selective forgetting reduces stereotypes but increases overall forgetting; (3) general forgetting alleviation methods are ineffective at reducing overall forgetting or improving downstream performance.

Conclusion: Current fairness objectives have limitations in achieving effective performance trade-offs, highlighting the need for better approaches that can selectively mitigate stereotypes without causing excessive overall forgetting.

Abstract: Moral alignment has emerged as a widely adopted approach for regulating the behavior of pretrained language models (PLMs), typically through fine-tuning on curated datasets. Gender stereotype mitigation is a representational task within the broader application of moral alignment. However, this process often comes at the cost of degraded downstream task performance. Prior studies commonly aim to achieve a performance trade-off by encouraging PLMs to selectively forget only stereotypical knowledge through carefully designed fairness objective, while preserving their language modeling capability (overall forgetting). In this short paper, we investigate whether the performance trade-off can be achieved through the lens of forgetting and the fairness objective. Our analysis shows that the large datasets needed for satisfactory fairness highlight the limitations of current fairness objectives in achieving an effective trade-off: (1) downstream task performance is strongly correlated with overall forgetting; (2) selective forgetting reduces stereotypes, but overall forgetting increases. and (3) general solutions for alleviating forgetting are ineffective at reducing the overall forgetting and fail to improve downstream task performance.

Method: Compared DNN-based embeddings with LLM few-shot learning on DAIC-WOZ clinical interview transcripts. Applied fairness-aware loss functions (worst-group loss and fairness-regularized loss) to DNN models, and explored in-context learning with varied prompt framing and shot counts for LLMs.

Result: LLMs outperformed DNN-based models in depression classification, particularly for Hispanic participants. LLMs showed reduced gender bias compared to DNNs, though racial disparities persisted. Worst-group loss achieved better performance-fairness balance for DNNs, while guided prompting with ethical framing helped mitigate gender bias in LLMs (1-shot setting). Increasing shot count didn’t reduce disparities further.

Conclusion: LLMs offer advantages over DNN-based approaches for depression detection with reduced gender bias, but racial disparities remain challenging. Fairness-aware techniques show promise but require careful selection, and prompt engineering has limited effectiveness for addressing racial bias in LLMs.

Abstract: This paper investigates algorithmic bias in language-based models for automated depression detection, focusing on socio-demographic disparities related to gender and race/ethnicity. Models trained using deep neural networks (DNN) based embeddings are compared to few-shot learning approaches with large language models (LLMs), evaluating both performance and fairness on clinical interview transcripts from the Distress Analysis Interview Corpus/Wizard-of-Oz (DAIC-WOZ). To mitigate bias, fairness-aware loss functions are applied to DNN-based models, while in-context learning with varied prompt framing and shot counts is explored for LLMs. Results indicate that LLMs outperform DNN-based models in depression classification, particularly for underrepresented groups such as Hispanic participants. LLMs also exhibit reduced gender bias compared to DNN-based embeddings, though racial disparities persist. Among fairness-aware techniques for mitigating bias in DNN-based embeddings, the worst-group loss, which is designed to minimize loss for the worst-performing demographic group, achieves a better balance between performance and fairness. In contrast, the fairness-regularized loss minimizes loss across all groups but performs less effectively. In LLMs, guided prompting with ethical framing helps mitigate gender bias in the 1-shot setting. However, increasing the number of shots does not lead to further reductions in disparities. For race/ethnicity, neither prompting strategy nor increasing $N$ in $N$-shot learning effectively reduces disparities.

Method: Created MathLens benchmark with annotated components: visual diagrams, textual descriptions, controlled multimodal questions, and perceptual probes derived from symbolic problem specifications to ensure consistency.

Result: Reinforcement learning mainly strengthens perception, especially with textual supervision. Textual SFT indirectly improves perception through reasoning. Reasoning only improves with perception. Integration remains the weakest skill. RL improves robustness while multimodal SFT reduces it through overfitting.

Conclusion: Multimodal reasoning involves distinct subskills that develop unevenly with different training approaches, with integration being the most challenging component that requires targeted improvement.

Abstract: Multimodal reasoning models have recently shown promise on challenging domains such as olympiad-level geometry, yet their evaluation remains dominated by aggregate accuracy, a single score that obscures where and how models are improving. We introduce MathLens, a benchmark designed to disentangle the subskills of multimodal reasoning while preserving the complexity of textbook-style geometry problems. The benchmark separates performance into three components: Perception: extracting information from raw inputs, Reasoning: operating on available information, and Integration: selecting relevant perceptual evidence and applying it within reasoning. To support each test, we provide annotations: visual diagrams, textual descriptions to evaluate reasoning in isolation, controlled questions that require both modalities, and probes for fine-grained perceptual skills, all derived from symbolic specifications of the problems to ensure consistency and robustness. Our analysis reveals that different training approaches have uneven effects: First, reinforcement learning chiefly strengthens perception, especially when supported by textual supervision, while textual SFT indirectly improves perception through reflective reasoning. Second, reasoning improves only in tandem with perception. Third, integration remains the weakest capacity, with residual errors concentrated there once other skills advance. Finally, robustness diverges: RL improves consistency under diagram variation, whereas multimodal SFT reduces it through overfitting. We will release all data and experimental logs.

Method: Created Agri-3M-VL corpus using scalable multi-agent data generator; developed AgriGPT-VL model via progressive curriculum of textual grounding, multimodal alignment, and GRPO reinforcement learning refinement.

Result: AgriGPT-VL outperforms leading general-purpose VLMs on AgriBench-VL-4K with higher pairwise win rates, while remaining competitive on text-only AgriBench-13K without language ability degradation.

Conclusion: The framework successfully addresses agricultural multimodal challenges and will be open-sourced to support reproducible research and deployment in low-resource settings.

Abstract: Despite rapid advances in multimodal large language models, agricultural applications remain constrained by the scarcity of domain-tailored models, curated vision-language corpora, and rigorous evaluation. To address these challenges, we present the AgriGPT-VL Suite, a unified multimodal framework for agriculture. Our contributions are threefold. First, we introduce Agri-3M-VL, the largest vision-language corpus for agriculture to our knowledge, curated by a scalable multi-agent data generator; it comprises 1M image-caption pairs, 2M image-grounded VQA pairs, 50K expert-level VQA instances, and 15K GRPO reinforcement learning samples. Second, we develop AgriGPT-VL, an agriculture-specialized vision-language model trained via a progressive curriculum of textual grounding, multimodal shallow/deep alignment, and GRPO refinement. This method achieves strong multimodal reasoning while preserving text-only capability. Third, we establish AgriBench-VL-4K, a compact yet challenging evaluation suite with open-ended and image-grounded questions, paired with multi-metric evaluation and an LLM-as-a-judge framework. Experiments show that AgriGPT-VL outperforms leading general-purpose VLMs on AgriBench-VL-4K, achieving higher pairwise win rates in the LLM-as-a-judge evaluation. Meanwhile, it remains competitive on the text-only AgriBench-13K with no noticeable degradation of language ability. Ablation studies further confirm consistent gains from our alignment and GRPO refinement stages. We will open source all of the resources to support reproducible research and deployment in low-resource agricultural settings.

Method: Developed a methodology to measure epistemic diversity (variation in real-world claims) and conducted a broad empirical study testing 27 LLMs, 155 topics across 12 countries, and 200 prompt variations from real user chats.

Result: Newer models generate more diverse claims but nearly all are less epistemically diverse than basic web search. Model size negatively impacts diversity, while RAG positively impacts it (with effectiveness varying by cultural context). Country-specific claims reflect English language more than local languages.

Conclusion: LLMs exhibit epistemic homogenization and knowledge collapse risk, with cultural context playing a significant role in diversity patterns. There’s a gap in epistemic representation favoring English over local languages.

Abstract: Large language models (LLMs) tend to generate lexically, semantically, and stylistically homogenous texts. This poses a risk of knowledge collapse, where homogenous LLMs mediate a shrinking in the range of accessible information over time. Existing works on homogenization are limited by a focus on closed-ended multiple-choice setups or fuzzy semantic features, and do not look at trends across time and cultural contexts. To overcome this, we present a new methodology to measure epistemic diversity, i.e., variation in real-world claims in LLM outputs, which we use to perform a broad empirical study of LLM knowledge collapse. We test 27 LLMs, 155 topics covering 12 countries, and 200 prompt variations sourced from real user chats. For the topics in our study, we show that while newer models tend to generate more diverse claims, nearly all models are less epistemically diverse than a basic web search. We find that model size has a negative impact on epistemic diversity, while retrieval-augmented generation (RAG) has a positive impact, though the improvement from RAG varies by the cultural context. Finally, compared to a traditional knowledge source (Wikipedia), we find that country-specific claims reflect the English language more than the local one, highlighting a gap in epistemic representation

Method: Prepend short system-prompt instructions to finetuning data that deliberately elicit undesirable traits, then evaluate without these instructions at test time.

Result: Inoculated models show much lower expression of undesirable traits across multiple settings including reducing emergent misalignment, defending against backdoor injections, and mitigating trait transmission via subliminal learning.

Conclusion: Inoculation is an effective technique for selective learning that reduces optimization pressure by making traits less surprising, contributing to better understanding of how language models generalize.

Abstract: Language model finetuning often results in learning undesirable traits in combination with desired ones. To address this, we propose inoculation prompting: modifying finetuning data by prepending a short system-prompt instruction that deliberately elicits the undesirable trait. At test time, we evaluate without the instruction; inoculated models have much lower expression of the trait than models trained with unmodified training data. Inoculation is selective: in a toy setting where assistant responses are always in Spanish and ALL-CAPS, an appropriate inoculation (e.g., ``You always speak in Spanish.’’) teaches the model to capitalize responses while still responding in English. We find that inoculation is also effective across several additional settings: reducing emergent misalignment (EM) from task-specific finetuning, defending against backdoor injections, and mitigating the transmission of traits via subliminal learning. Follow-up analysis suggests a mechanism: making a trait less surprising via inoculation reduces optimization pressure to globally update the model, thereby reducing the degree of generalization. Our analysis relates to prior work on EM: inoculation explains prior findings that educational contexts mitigate EM from insecure code. Beyond demonstrating a simple and effective technique for selective learning, our results contribute to a better conceptual understanding of how and why language models generalize.

Method: Applied SciNCL (graph-aware neighborhood contrastive learning) methodology to process industry domain, using triplets derived from graph embeddings to fine-tune language models.

Result: Models fine-tuned with graph embedding triplets outperformed state-of-the-art mE5-large text encoder by 9.8-14.3% on proprietary PITEB benchmark while having 3 times fewer parameters.

Conclusion: Graph-aware contrastive learning effectively enhances language models for process industry applications, achieving superior performance with significantly reduced model size.

Abstract: Recent trends in NLP utilize knowledge graphs (KGs) to enhance pretrained language models by incorporating additional knowledge from the graph structures to learn domain-specific terminology or relationships between documents that might otherwise be overlooked. This paper explores how SciNCL, a graph-aware neighborhood contrastive learning methodology originally designed for scientific publications, can be applied to the process industry domain, where text logs contain crucial information about daily operations and are often structured as sparse KGs. Our experiments demonstrate that language models fine-tuned with triplets derived from graph embeddings (GE) outperform a state-of-the-art mE5-large text encoder by 9.8-14.3% (5.45-7.96p) on the proprietary process industry text embedding benchmark (PITEB) while having 3 times fewer parameters.

Method: AWARE framework with three components: Domain Awareness (vocabulary adaptation), Context Awareness (full essay embeddings), and Class Overlap Awareness (multi-label strategy for theme coexistence).

Result: AWARE outperforms strong baseline by 2.1 percentage points in Macro-F1 and shows considerable improvements across all cultural capital themes.

Conclusion: Provides a robust and generalizable methodology for text classification tasks where meaning depends on narrative context.

Abstract: Identifying cultural capital (CC) themes in student reflections can offer valuable insights that help foster equitable learning environments in classrooms. However, themes such as aspirational goals or family support are often woven into narratives, rather than appearing as direct keywords. This makes them difficult to detect for standard NLP models that process sentences in isolation. The core challenge stems from a lack of awareness, as standard models are pre-trained on general corpora, leaving them blind to the domain-specific language and narrative context inherent to the data. To address this, we introduce AWARE, a framework that systematically attempts to improve a transformer model’s awareness for this nuanced task. AWARE has three core components: 1) Domain Awareness, adapting the model’s vocabulary to the linguistic style of student reflections; 2) Context Awareness, generating sentence embeddings that are aware of the full essay context; and 3) Class Overlap Awareness, employing a multi-label strategy to recognize the coexistence of themes in a single sentence. Our results show that by making the model explicitly aware of the properties of the input, AWARE outperforms a strong baseline by 2.1 percentage points in Macro-F1 and shows considerable improvements across all themes. This work provides a robust and generalizable methodology for any text classification task in which meaning depends on the context of the narrative.

Method: Created COLE benchmark with 23 diverse NLU tasks covering sentiment analysis, paraphrase detection, grammatical judgment, and reasoning. Evaluated 94 large language models on this benchmark.

Result: Revealed significant performance gap between closed- and open-weights models. Identified challenging frontiers: zero-shot extractive QA, fine-grained word sense disambiguation, and understanding regional language variations.

Conclusion: COLE is released as a public resource to foster progress in French language modeling, providing comprehensive evaluation framework for French NLU capabilities.

Abstract: To address the need for a more comprehensive evaluation of French Natural Language Understanding (NLU), we introduce COLE, a new benchmark composed of 23 diverse task covering a broad range of NLU capabilities, including sentiment analysis, paraphrase detection, grammatical judgment, and reasoning, with a particular focus on linguistic phenomena relevant to the French language. We benchmark 94 large language models (LLM), providing an extensive analysis of the current state of French NLU. Our results highlight a significant performance gap between closed- and open-weights models and identify key challenging frontiers for current LLMs, such as zero-shot extractive question-answering (QA), fine-grained word sense disambiguation, and understanding of regional language variations. We release COLE as a public resource to foster further progress in French language modelling.

Method: Enhanced Feature Pyramid Network with InceptionSepConv blocks, depth-wise separable convolutions, prototypical learning with masked average pooling, and attention mechanisms (global/local self-attention, cross-attention).

Result: Achieved 82.55% F1-score and 72.26% mIoU in 2-way classification testing, with self-attention providing 2.57% F1-score and 2.9% mIoU improvement over baselines.

Conclusion: The framework successfully enables rapid response to new defect types in infrastructure inspection with limited training data, leading to more efficient and economical maintenance of critical infrastructure.

Abstract: Few-shot semantic segmentation is vital for deep learning-based infrastructure inspection applications, where labeled training examples are scarce and expensive. Although existing deep learning frameworks perform well, the need for extensive labeled datasets and the inability to learn new defect categories with little data are problematic. We present our Enhanced Feature Pyramid Network (E-FPN) framework for few-shot semantic segmentation of culvert and sewer defect categories using a prototypical learning framework. Our approach has three main contributions: (1) adaptive E-FPN encoder using InceptionSepConv blocks and depth-wise separable convolutions for efficient multi-scale feature extraction; (2) prototypical learning with masked average pooling for powerful prototype generation from small support examples; and (3) attention-based feature representation through global self-attention, local self-attention and cross-attention. Comprehensive experimentation on challenging infrastructure inspection datasets illustrates that the method achieves excellent few-shot performance, with the best configuration being 8-way 5-shot training configuration at 82.55% F1-score and 72.26% mIoU in 2-way classification testing. The self-attention method had the most significant performance improvements, providing 2.57% F1-score and 2.9% mIoU gain over baselines. Our framework addresses the critical need to rapidly respond to new defect types in infrastructure inspection systems with limited new training data that lead to more efficient and economical maintenance plans for critical infrastructure systems.

Method: Introduced a new skin segmentation technique that detects skin regions across the body while removing problematic areas (mouth, eyes, hair). Evaluated on public datasets and introduced SYNC-rPPG dataset for real-world conditions.

Result: The model demonstrates superior ability to capture heartbeats in challenging conditions (talking, head rotation) while maintaining low mean absolute error (MAE). Shows high accuracy across diverse skin tones.

Conclusion: The proposed skin segmentation technique makes rPPG a more promising option for real-world applications by improving robustness to movement and handling diverse skin tones effectively.

Abstract: Remote photoplethysmography (rPPG) is an innovative method for monitoring heart rate and vital signs by using a simple camera to record a person, as long as any part of their skin is visible. This low-cost, contactless approach helps in remote patient monitoring, emotion analysis, smart vehicle utilization, and more. Over the years, various techniques have been proposed to improve the accuracy of this technology, especially given its sensitivity to lighting and movement. In the unsupervised pipeline, it is necessary to first select skin regions from the video to extract the rPPG signal from the skin color changes. We introduce a novel skin segmentation technique that prioritizes skin regions to enhance the quality of the extracted signal. It can detect areas of skin all over the body, making it more resistant to movement, while removing areas such as the mouth, eyes, and hair that may cause interference. Our model is evaluated on publicly available datasets, and we also present a new dataset, called SYNC-rPPG, to better represent real-world conditions. The results indicate that our model demonstrates a prior ability to capture heartbeats in challenging conditions, such as talking and head rotation, and maintain the mean absolute error (MAE) between predicted and actual heart rates, while other methods fail to do so. In addition, we demonstrate high accuracy in detecting a diverse range of skin tones, making this technique a promising option for real-world applications.

Method: A hybrid architecture that combines spatial and spectral features for single-shot focus prediction. The network automatically regresses the distance to the optimal focal point using extracted spatiospectral features and adjusts control parameters for optimal image outcomes.

Result: 80% reduction in focusing time compared to stack-based methods, focus accuracy of 0.18 μm (matching dual-image methods), robust cross-lab generalization with only 0.72% false focus predictions, and 90% of predictions within depth of field. Achieved 0.90 AUC in cancer classification at 4x magnification.

Conclusion: The system provides a comprehensive hardware-software design that enables accessible, real-time digital pathology in resource-constrained settings while maintaining diagnostic accuracy comparable to expensive WSI scanners.

Abstract: While Whole Slide Imaging (WSI) scanners remain the gold standard for digitizing pathology samples, their high cost limits accessibility in many healthcare settings. Other low-cost solutions also face critical limitations: automated microscopes struggle with consistent focus across varying tissue morphology, traditional auto-focus methods require time-consuming focal stacks, and existing deep-learning approaches either need multiple input images or lack generalization capability across tissue types and staining protocols. We introduce a novel automated microscopic system powered by DeepAf, a novel auto-focus framework that uniquely combines spatial and spectral features through a hybrid architecture for single-shot focus prediction. The proposed network automatically regresses the distance to the optimal focal point using the extracted spatiospectral features and adjusts the control parameters for optimal image outcomes. Our system transforms conventional microscopes into efficient slide scanners, reducing focusing time by 80% compared to stack-based methods while achieving focus accuracy of 0.18 {\mu}m on the same-lab samples, matching the performance of dual-image methods (0.19 {\mu}m) with half the input requirements. DeepAf demonstrates robust cross-lab generalization with only 0.72% false focus predictions and 90% of predictions within the depth of field. Through an extensive clinical study of 536 brain tissue samples, our system achieves 0.90 AUC in cancer classification at 4x magnification, a significant achievement at lower magnification than typical 20x WSI scans. This results in a comprehensive hardware-software design enabling accessible, real-time digital pathology in resource-constrained settings while maintaining diagnostic accuracy.

Method: Uses multimodal architecture with log-mel spectrograms, optional image embeddings, and temporal features through convolutional-transformer pipeline for temporal segmentation. For spatial selection, employs CLIP-based encoder instead of ResNet and constrains distractor selection to same concert segments.

Result: Models outperformed previous baselines in detecting cut points and provided competitive visual shot selection, advancing state of the art in multimodal automated video editing.

Conclusion: The proposed approach effectively decomposes multicamera video editing into temporal and spatial tasks, demonstrating improved performance over existing methods through multimodal integration and modern backbone architectures.

Abstract: Automated video editing remains an underexplored task in the computer vision and multimedia domains, especially when contrasted with the growing interest in video generation and scene understanding. In this work, we address the specific challenge of editing multicamera recordings of classical music concerts by decomposing the problem into two key sub-tasks: when to cut and how to cut. Building on recent literature, we propose a novel multimodal architecture for the temporal segmentation task (when to cut), which integrates log-mel spectrograms from the audio signals, plus an optional image embedding, and scalar temporal features through a lightweight convolutional-transformer pipeline. For the spatial selection task (how to cut), we improve the literature by updating from old backbones, e.g. ResNet, with a CLIP-based encoder and constraining distractor selection to segments from the same concert. Our dataset was constructed following a pseudo-labeling approach, in which raw video data was automatically clustered into coherent shot segments. We show that our models outperformed previous baselines in detecting cut points and provide competitive visual shot selection, advancing the state of the art in multimodal automated video editing.

Method: Used transfer learning with fine-tuning on five pre-trained CNNs (DenseNet201, InceptionV3, ResNet152V2, SeResNet152, Xception) for multi-class identification across eight mango leaf disease classes.

Result: DenseNet201 delivered the best results with 99.33% accuracy, excelling particularly in identifying Cutting Weevil and Bacterial Canker. ResNet152V2 and SeResNet152 also performed well, while InceptionV3 and Xception showed lower performance in visually similar categories.

Conclusion: Fine-tuned transfer learning models demonstrate capability for precise and dependable multi-class mango leaf disease detection in intelligent agricultural applications.

Abstract: Mango is an important fruit crop in South Asia, but its cultivation is frequently hampered by leaf diseases that greatly impact yield and quality. This research examines the performance of five pre-trained convolutional neural networks, DenseNet201, InceptionV3, ResNet152V2, SeResNet152, and Xception, for multi-class identification of mango leaf diseases across eight classes using a transfer learning strategy with fine-tuning. The models were assessed through standard evaluation metrics, such as accuracy, precision, recall, F1-score, and confusion matrices. Among the architectures tested, DenseNet201 delivered the best results, achieving 99.33% accuracy with consistently strong metrics for individual classes, particularly excelling in identifying Cutting Weevil and Bacterial Canker. Moreover, ResNet152V2 and SeResNet152 provided strong outcomes, whereas InceptionV3 and Xception exhibited lower performance in visually similar categories like Sooty Mould and Powdery Mildew. The training and validation plots demonstrated stable convergence for the highest-performing models. The capability of fine-tuned transfer learning models, for precise and dependable multi-class mango leaf disease detection in intelligent agricultural applications.

Method: Introduces Dynamic Guidance which selectively sharpens the score function only along pre-determined directions known to cause artifacts, addressing hallucinations at generation time rather than through post-hoc filtering.

Result: Dynamic Guidance substantially reduces hallucinations on both controlled and natural image datasets, significantly outperforming baselines.

Conclusion: Dynamic Guidance provides a nuanced solution to mitigate hallucinations in diffusion models while preserving semantic diversity, representing the first approach that addresses this issue at generation time.

Abstract: Diffusion models, despite their impressive demos, often produce hallucinatory samples with structural inconsistencies that lie outside of the support of the true data distribution. Such hallucinations can be attributed to excessive smoothing between modes of the data distribution. However, semantic interpolations are often desirable and can lead to generation diversity, thus we believe a more nuanced solution is required. In this work, we introduce Dynamic Guidance, which tackles this issue. Dynamic Guidance mitigates hallucinations by selectively sharpening the score function only along the pre-determined directions known to cause artifacts, while preserving valid semantic variations. To our knowledge, this is the first approach that addresses hallucinations at generation time rather than through post-hoc filtering. Dynamic Guidance substantially reduces hallucinations on both controlled and natural image datasets, significantly outperforming baselines.

Method: Proposes three stage-specific strategies: 1) Asynchronous Cache Swapping, 2) Feature chunk, and 3) Slicing latents to decode, designed to reduce memory consumption while keeping time overhead lower than acceleration gains.

Result: Achieves faster inference speed and lower memory usage compared to baseline, with quality degradation maintained within acceptable range.

Conclusion: The proposed LightCache method effectively accelerates video generation while managing memory consumption through targeted stage-specific optimizations.

Abstract: Training-free acceleration has emerged as an advanced research area in video generation based on diffusion models. The redundancy of latents in diffusion model inference provides a natural entry point for acceleration. In this paper, we decompose the inference process into the encoding, denoising, and decoding stages, and observe that cache-based acceleration methods often lead to substantial memory surges in the latter two stages. To address this problem, we analyze the characteristics of inference across different stages and propose stage-specific strategies for reducing memory consumption: 1) Asynchronous Cache Swapping. 2) Feature chunk. 3) Slicing latents to decode. At the same time, we ensure that the time overhead introduced by these three strategies remains lower than the acceleration gains themselves. Compared with the baseline, our approach achieves faster inference speed and lower memory usage, while maintaining quality degradation within an acceptable range. The Code is available at https://github.com/NKUShaw/LightCache .

Method: Couples Visual-Language Models (VLMs) with constrained diffusion process - one VLM generates scene descriptions and another guides image reconstruction to ensure semantic and structural consistency.

Result: Successfully reconstructs plausible past frames up to 120 seconds earlier in three controlled scenarios, demonstrating feasibility of time-reversed imaging from thermal traces.

Conclusion: Provides a first step toward time-reversed imaging from thermal traces, enabling recovery of past events that exceed RGB camera capabilities.

Abstract: Recovering the past from present observations is an intriguing challenge with potential applications in forensics and scene analysis. Thermal imaging, operating in the infrared range, provides access to otherwise invisible information. Since humans are typically warmer (37 C -98.6 F) than their surroundings, interactions such as sitting, touching, or leaning leave residual heat traces. These fading imprints serve as passive temporal codes, allowing for the inference of recent events that exceed the capabilities of RGB cameras. This work proposes a time-reversed reconstruction framework that uses paired RGB and thermal images to recover scene states from a few seconds earlier. The proposed approach couples Visual-Language Models (VLMs) with a constrained diffusion process, where one VLM generates scene descriptions and another guides image reconstruction, ensuring semantic and structural consistency. The method is evaluated in three controlled scenarios, demonstrating the feasibility of reconstructing plausible past frames up to 120 seconds earlier, providing a first step toward time-reversed imaging from thermal traces.

Method: Extends morphological component analysis (MCA) by introducing eigenspace clustering of dynamic mode decomposition eigenvalues to obtain data-driven dictionaries. The DMCA algorithm is derived and applied to various scenarios including video denoising, faint target enhancement, and radar image separation.

Result: DMCA demonstrates effectiveness in denoising videos from Adobe 240fps dataset, enhances signal-to-noise ratio of faint targets in sea states, and successfully separates bicycles from wind clutter in inverse synthetic aperture radar images.

Conclusion: Dynamic Morphological Component Analysis provides a powerful data-driven approach for video representation and signal separation that adapts to scene dynamics, offering improved performance over traditional methods with predefined dictionaries across multiple applications.

Abstract: This paper introduces a novel methodology of adapting the representation of videos based on the dynamics of their scene content variation. In particular, we demonstrate how the clustering of dynamic mode decomposition eigenvalues can be leveraged to learn an adaptive video representation for separating structurally distinct morphologies of a video. We extend the morphological component analysis (MCA) algorithm, which uses multiple predefined incoherent dictionaries and a sparsity prior to separate distinct sources in signals, by introducing our novel eigenspace clustering technique to obtain data-driven MCA dictionaries, which we call dynamic morphological component analysis (DMCA). After deriving our novel algorithm, we offer a motivational example of DMCA applied to a still image, then demonstrate DMCA’s effectiveness in denoising applications on videos from the Adobe 240fps dataset. Afterwards, we provide an example of DMCA enhancing the signal-to-noise ratio of a faint target summed with a sea state, and conclude the paper by applying DMCA to separate a bicycle from wind clutter in inverse synthetic aperture radar images.

Method: Design a mapping network (pi-map) that translates local image embeddings of object instances into text tokens, which can then be combined with natural language queries for CLIP-style text encoding and image retrieval. Uses frozen CLIP encoders with a simple training procedure per object instance.

Result: The approach improves state-of-the-art performance on two benchmarks for personalized image retrieval.

Conclusion: The pi-map method effectively enables compound query image retrieval by bridging visual object instances with natural language descriptions through learnable text token mapping.

Abstract: The goal of this paper is to be able to retrieve images using a compound query that combines object instance information from an image, with a natural text description of what that object is doing or where it is. For example, to retrieve an image of “Fluffy the unicorn (specified by an image) on someone’s head”. To achieve this we design a mapping network that can “translate” from a local image embedding (of the object instance) to a text token, such that the combination of the token and a natural language query is suitable for CLIP style text encoding, and image retrieval. Generating a text token in this manner involves a simple training procedure, that only needs to be performed once for each object instance. We show that our approach of using a trainable mapping network, termed pi-map, together with frozen CLIP text and image encoders, improves the state of the art on two benchmarks designed to assess personalized retrieval.

Method: Parameterizes Gaussians in 2D UV feature space with multi-level learnable feature maps, uses neural decoder to transform latent features into Gaussian attributes, and controls Gaussian count by dynamically resampling feature maps at desired resolutions.

Result: Achieves SOTA quality at highest LOD in reenactment tasks, maintains near SOTA performance at lower LODs, uses only 6.2% Gaussians at lowest LOD with moderate quality degradation, and nearly doubles rendering speed.

Conclusion: ArchitectHead enables efficient continuous LOD control without retraining, providing a practical solution for 3D head avatars that balances visual quality and rendering efficiency.

Abstract: 3D Gaussian Splatting (3DGS) has enabled photorealistic and real-time rendering of 3D head avatars. Existing 3DGS-based avatars typically rely on tens of thousands of 3D Gaussian points (Gaussians), with the number of Gaussians fixed after training. However, many practical applications require adjustable levels of detail (LOD) to balance rendering efficiency and visual quality. In this work, we propose “ArchitectHead”, the first framework for creating 3D Gaussian head avatars that support continuous control over LOD. Our key idea is to parameterize the Gaussians in a 2D UV feature space and propose a UV feature field composed of multi-level learnable feature maps to encode their latent features. A lightweight neural network-based decoder then transforms these latent features into 3D Gaussian attributes for rendering. ArchitectHead controls the number of Gaussians by dynamically resampling feature maps from the UV feature field at the desired resolutions. This method enables efficient and continuous control of LOD without retraining. Experimental results show that ArchitectHead achieves state-of-the-art (SOTA) quality in self and cross-identity reenactment tasks at the highest LOD, while maintaining near SOTA performance at lower LODs. At the lowest LOD, our method uses only 6.2% of the Gaussians while the quality degrades moderately (L1 Loss +7.9%, PSNR –0.97%, SSIM –0.6%, LPIPS Loss +24.1%), and the rendering speed nearly doubles.

Method: Pipeline segments human point clouds from background, tracks individuals over time, performs body part segmentation, and uses a novel backbone combining point-based techniques with sparse convolutional networks on voxel-mapped point cloud sequences.

Result: Achieves 89.3% accuracy on NTU RGB-D 120 dataset when combining sensor-based and estimated depth inputs in ensemble setup, outperforming previous point cloud action recognition methods.

Conclusion: The proposed method demonstrates competitive performance with existing skeletal action recognition algorithms and shows the viability of using dense 3D point cloud data for human action recognition.

Abstract: Recent research into human action recognition (HAR) has focused predominantly on skeletal action recognition and video-based methods. With the increasing availability of consumer-grade depth sensors and Lidar instruments, there is a growing opportunity to leverage dense 3D data for action recognition, to develop a third way. This paper presents a novel approach for recognizing actions from 3D videos by introducing a pipeline that segments human point clouds from the background of a scene, tracks individuals over time, and performs body part segmentation. The method supports point clouds from both depth sensors and monocular depth estimation. At the core of the proposed HAR framework is a novel backbone for 3D action recognition, which combines point-based techniques with sparse convolutional networks applied to voxel-mapped point cloud sequences. Experiments incorporate auxiliary point features including surface normals, color, infrared intensity, and body part parsing labels, to enhance recognition accuracy. Evaluation on the NTU RGB- D 120 dataset demonstrates that the method is competitive with existing skeletal action recognition algorithms. Moreover, combining both sensor-based and estimated depth inputs in an ensemble setup, this approach achieves 89.3% accuracy when different human subjects are considered for training and testing, outperforming previous point cloud action recognition methods.

Method: The authors propose a novel Riemannian metric on the noise space, inspired by findings that the Jacobian of the score function captures tangent spaces to the local data manifold. This metric encourages geodesics to stay within or parallel to the learned data manifold.

Result: Experiments on image interpolation show that the proposed metric produces perceptually more natural and faithful transitions compared to existing density-based and naive baselines.

Conclusion: The Riemannian metric approach successfully enables manifold-aware interpolation in diffusion models, overcoming limitations of existing methods and producing more natural transitions aligned with the data manifold.

Abstract: Diffusion models are powerful deep generative models (DGMs) that generate high-fidelity, diverse content. However, unlike classical DGMs, they lack an explicit, tractable low-dimensional latent space that parameterizes the data manifold. This absence limits manifold-aware analysis and operations, such as interpolation and editing. Existing interpolation methods for diffusion models typically follow paths through high-density regions, which are not necessarily aligned with the data manifold and can yield perceptually unnatural transitions. To exploit the data manifold learned by diffusion models, we propose a novel Riemannian metric on the noise space, inspired by recent findings that the Jacobian of the score function captures the tangent spaces to the local data manifold. This metric encourages geodesics in the noise space to stay within or run parallel to the learned data manifold. Experiments on image interpolation show that our metric produces perceptually more natural and faithful transitions than existing density-based and naive baselines.

Method: Uses multiple coordinated instances of base diffusion models (teammates) with novel LoRA adaptation for both adaptation and coordination, supporting dynamic activation/deactivation of teammates.

Result: Successfully applied to various generative and inverse graphics tasks including inpainting, SVBRDF estimation, intrinsic decomposition, neural shading, and intrinsic image synthesis.

Conclusion: Teamwork provides a flexible and efficient unified approach for channel expansion and task adaptation in diffusion models across multiple graphics applications.

Abstract: Large pretrained diffusion models can provide strong priors beneficial for many graphics applications. However, generative applications such as neural rendering and inverse methods such as SVBRDF estimation and intrinsic image decomposition require additional input or output channels. Current solutions for channel expansion are often application specific and these solutions can be difficult to adapt to different diffusion models or new tasks. This paper introduces Teamwork: a flexible and efficient unified solution for jointly increasing the number of input and output channels as well as adapting a pretrained diffusion model to new tasks. Teamwork achieves channel expansion without altering the pretrained diffusion model architecture by coordinating and adapting multiple instances of the base diffusion model (\ie, teammates). We employ a novel variation of Low Rank-Adaptation (LoRA) to jointly address both adaptation and coordination between the different teammates. Furthermore Teamwork supports dynamic (de)activation of teammates. We demonstrate the flexibility and efficiency of Teamwork on a variety of generative and inverse graphics tasks such as inpainting, single image SVBRDF estimation, intrinsic decomposition, neural shading, and intrinsic image synthesis.

Method: Two experiments: Experiment A tested 288 handwritten arithmetic responses from Ghanaian students; Experiment B evaluated 150 mathematical illustrations from American elementary students, testing both direct grading and grading with human descriptions.

Result: MLLMs achieved near-human accuracy (95%, k=0.90) on arithmetic work but struggled with illustrations (k=0.20). Performance on illustrations improved dramatically (k=0.47) when given human descriptions, matching human-to-human agreement levels.

Conclusion: MLLMs can effectively ‘see’ and interpret arithmetic work but still struggle to ‘see’ student mathematical illustrations, suggesting current limitations in visual interpretation capabilities for complex educational tasks.

Abstract: Recent advances in multimodal large language models (MLLMs) raise the question of their potential for grading, analyzing, and offering feedback on handwritten student classwork. This capability would be particularly beneficial in elementary and middle-school mathematics education, where most work remains handwritten, because seeing students’ full working of a problem provides valuable insights into their learning processes, but is extremely time-consuming to grade. We present two experiments investigating MLLM performance on handwritten student mathematics classwork. Experiment A examines 288 handwritten responses from Ghanaian middle school students solving arithmetic problems with objective answers. In this context, models achieved near-human accuracy (95%, k = 0.90) but exhibited occasional errors that human educators would be unlikely to make. Experiment B evaluates 150 mathematical illustrations from American elementary students, where the drawings are the answer to the question. These tasks lack single objective answers and require sophisticated visual interpretation as well as pedagogical judgment in order to analyze and evaluate them. We attempted to separate MLLMs’ visual capabilities from their pedagogical abilities by first asking them to grade the student illustrations directly, and then by augmenting the image with a detailed human description of the illustration. We found that when the models had to analyze the student illustrations directly, they struggled, achieving only k = 0.20 with ground truth scores, but when given human descriptions, their agreement levels improved dramatically to k = 0.47, which was in line with human-to-human agreement levels. This gap suggests MLLMs can “see” and interpret arithmetic work relatively well, but still struggle to “see” student mathematical illustrations.

Method: Uses a midway top-down path to infer motion latents between frames, dense forward prediction objective, and hierarchical structure to handle complex multi-object scenes in natural videos.

Result: Achieves strong performance on semantic segmentation and optical flow tasks after pretraining on large-scale natural video datasets, outperforming prior self-supervised methods.

Conclusion: Midway Network successfully extends latent dynamics modeling to learn joint representations for recognition and motion from videos, with learned dynamics capturing high-level correspondence.

Abstract: Object recognition and motion understanding are key components of perception that complement each other. While self-supervised learning methods have shown promise in their ability to learn from unlabeled data, they have primarily focused on obtaining rich representations for either recognition or motion rather than both in tandem. On the other hand, latent dynamics modeling has been used in decision making to learn latent representations of observations and their transformations over time for control and planning tasks. In this work, we present Midway Network, a new self-supervised learning architecture that is the first to learn strong visual representations for both object recognition and motion understanding solely from natural videos, by extending latent dynamics modeling to this domain. Midway Network leverages a midway top-down path to infer motion latents between video frames, as well as a dense forward prediction objective and hierarchical structure to tackle the complex, multi-object scenes of natural videos. We demonstrate that after pretraining on two large-scale natural video datasets, Midway Network achieves strong performance on both semantic segmentation and optical flow tasks relative to prior self-supervised learning methods. We also show that Midway Network’s learned dynamics can capture high-level correspondence via a novel analysis method based on forward feature perturbation.

Method: Uses step-by-step V2A generation with negative guidance to discourage duplication of existing sounds. Trains guidance model by finetuning pre-trained V2A model on audio pairs from adjacent video segments, enabling use of standard single-reference datasets.

Result: Objective and subjective evaluations show enhanced separability of generated sounds at each step and improved overall quality of final composite audio, outperforming existing baselines.

Conclusion: The proposed incremental V2A generation method with negative guidance successfully achieves finer controllability and more realistic audio synthesis while using easily accessible training data.

Abstract: We propose a step-by-step video-to-audio (V2A) generation method for finer controllability over the generation process and more realistic audio synthesis. Inspired by traditional Foley workflows, our approach aims to comprehensively capture all sound events induced by a video through the incremental generation of missing sound events. To avoid the need for costly multi-reference video-audio datasets, each generation step is formulated as a negatively guided V2A process that discourages duplication of existing sounds. The guidance model is trained by finetuning a pre-trained V2A model on audio pairs from adjacent segments of the same video, allowing training with standard single-reference audiovisual datasets that are easily accessible. Objective and subjective evaluations demonstrate that our method enhances the separability of generated sounds at each step and improves the overall quality of the final composite audio, outperforming existing baselines.

Method: Uses an interactive scene-graph representation encoding object geometry, appearance, and physical properties with hierarchical relationships. Formulates reconstruction as energy-based optimization integrating observational data, physical constraints, and generative priors. Employs hybrid optimization combining sampling-based exploration with gradient-based refinement.

Result: Produces digital twins with complete and precise geometry, physical stability, and realistic rendering from novel viewpoints. Demonstrates superior performance on benchmark datasets and practical applications in interactive gaming and real-time digital-twin manipulation.

Conclusion: HoloScene successfully addresses multiple limitations of current 3D reconstruction methods, providing a unified framework that simultaneously achieves geometry completeness, physical plausibility, photorealistic rendering, and realistic physical properties for reliable dynamic simulation.

Abstract: Digitizing the physical world into accurate simulation-ready virtual environments offers significant opportunities in a variety of fields such as augmented and virtual reality, gaming, and robotics. However, current 3D reconstruction and scene-understanding methods commonly fall short in one or more critical aspects, such as geometry completeness, object interactivity, physical plausibility, photorealistic rendering, or realistic physical properties for reliable dynamic simulation. To address these limitations, we introduce HoloScene, a novel interactive 3D reconstruction framework that simultaneously achieves these requirements. HoloScene leverages a comprehensive interactive scene-graph representation, encoding object geometry, appearance, and physical properties alongside hierarchical and inter-object relationships. Reconstruction is formulated as an energy-based optimization problem, integrating observational data, physical constraints, and generative priors into a unified, coherent objective. Optimization is efficiently performed via a hybrid approach combining sampling-based exploration with gradient-based refinement. The resulting digital twins exhibit complete and precise geometry, physical stability, and realistic rendering from novel viewpoints. Evaluations conducted on multiple benchmark datasets demonstrate superior performance, while practical use-cases in interactive gaming and real-time digital-twin manipulation illustrate HoloScene’s broad applicability and effectiveness. Project page: https://xiahongchi.github.io/HoloScene.

Method: Proposes Contrastive Visual Enhancer (CVE) to decouple visual features and suppress dominant tokens, and Discriminative Concept Calibrator (DCC) to differentiate between general and discriminative concepts in text queries.

Result: Extensive experiments show consistent improvements across seven benchmarks spanning three image retrieval tasks.

Conclusion: CalibCLIP effectively addresses the dominant token problem in VLMs and improves text-driven image retrieval performance without requiring additional training.

Abstract: Existing Visual Language Models (VLMs) suffer structural limitations where a few low contribution tokens may excessively capture global semantics, dominating the information aggregation process and suppressing the discriminative features in text-driven image retrieval tasks. To address this, we introduce \textbf{CalibCLIP}, a training-free method designed to calibrate the suppressive effect of dominant tokens. Specifically, in the visual space, we propose the Contrastive Visual Enhancer (CVE), which decouples visual features into target and low information regions. Subsequently, it identifies dominant tokens and dynamically suppresses their representations.In the textual space, we introduce the Discriminative Concept Calibrator (DCC), which aims to differentiate between general and discriminative concepts within the text query. By mitigating the challenges posed by generic concepts and improving the representations of discriminative concepts, DCC strengthens the differentiation among similar samples. Finally, extensive experiments demonstrate consistent improvements across seven benchmarks spanning three image retrieval tasks, underscoring the effectiveness of CalibCLIP. Code is available at: https://github.com/kangbin98/CalibCLIP

Method: A deep mixture of multimodal hierarchical variational auto-encoders (MMHVAE) that creates complex latent representations, encourages variational distributions to estimate missing information, learns multimodal fusion with missing data, and leverages dataset-level information for incomplete training sets.

Result: Extensive experiments conducted on brain multi-parametric MRI and intra-operative ultrasound imaging demonstrate the method’s effectiveness.

Conclusion: MMHVAE successfully addresses key challenges in multimodal image synthesis and shows promise for medical imaging applications where complete multimodal data is not always available.

Abstract: We propose a deep mixture of multimodal hierarchical variational auto-encoders called MMHVAE that synthesizes missing images from observed images in different modalities. MMHVAE’s design focuses on tackling four challenges: (i) creating a complex latent representation of multimodal data to generate high-resolution images; (ii) encouraging the variational distributions to estimate the missing information needed for cross-modal image synthesis; (iii) learning to fuse multimodal information in the context of missing data; (iv) leveraging dataset-level information to handle incomplete data sets at training time. Extensive experiments are performed on the challenging problem of pre-operative brain multi-parametric magnetic resonance and intra-operative ultrasound imaging.

Method: ShortCoTI uses a reinforcement learning paradigm with an adaptive reward function that scales according to task difficulty, encouraging more concise CoT sequences while preserving output quality.

Result: Reduces prompt reasoning length by 54% while maintaining or slightly improving quality metrics across T2I-CompBench and GenEval benchmarks. Eliminates verbose explanations and repetitive refinements.

Conclusion: ShortCoTI improves computational efficiency without compromising the fidelity or visual appeal of generated images, producing reasoning prompts that are both concise and semantically rich.

Abstract: Autoregressive multimodal large language models have recently gained popularity for image generation, driven by advances in foundation models. To enhance alignment and detail, newer approaches employ chain-of-thought (CoT) reasoning, expanding user inputs into elaborated prompts prior to image synthesis. However, this strategy can introduce unnecessary redundancy – a phenomenon we call visual overthinking – which increases computational costs and can introduce details that contradict the original prompt. In this work, we explore how to generate more concise CoT sequences for more efficient image generation. We introduce ShortCoTI, a lightweight optimization framework that encourages more concise CoT while preserving output image quality. ShortCoTI rewards more concise prompts with an adaptive function that scales according to an estimated difficulty for each task. Incorporating this reward into a reinforcement learning paradigm reduces prompt reasoning length by 54% while maintaining or slightly improving quality metrics across multiple benchmarks (T2I-CompBench, GenEval). Qualitative analysis shows that our method eliminates verbose explanations and repetitive refinements, producing reasoning prompts that are both concise and semantically rich. As a result, ShortCoTI improves computational efficiency without compromising the fidelity or visual appeal of generated images.

Method: Proposes DiffCom framework with dual-density data flow, probabilistic conditional diffusion model, decoupled intra- and inter-point sparse priors, attention-guided latent denoiser, and local distribution integration in arithmetic coding.

Result: Achieves superior rate-distortion trade-off compared to state-of-the-art methods on ShapeNet and MPEG PCC datasets.

Conclusion: The diffusion-based framework with sparse priors effectively reduces redundancy in latent point cloud representations while maintaining high reconstruction quality.

Abstract: Lossy compression relies on an autoencoder to transform a point cloud into latent points for storage, leaving the inherent redundancy of latent representations unexplored. To reduce redundancy in latent points, we propose a diffusion-based framework guided by sparse priors that achieves high reconstruction quality, especially at low bitrates. Our approach features an efficient dual-density data flow that relaxes size constraints on latent points. It hybridizes a probabilistic conditional diffusion model to encapsulate essential details for reconstruction within sparse priors, which are decoupled hierarchically into intra- and inter-point priors. Specifically, our DiffCom encodes the original point cloud into latent points and decoupled sparse priors through separate encoders. To dynamically attend to geometric and semantic cues from the priors at each encoding and decoding layer, we employ an attention-guided latent denoiser conditioned on the decoupled priors. Additionally, we integrate the local distribution into the arithmetic encoder and decoder to enhance local context modeling of the sparse points. The original point cloud is reconstructed through a point decoder. Compared to state-of-the-art methods, our approach achieves a superior rate-distortion trade-off, as evidenced by extensive evaluations on the ShapeNet dataset and standard test datasets from the MPEG PCC Group.

Method: Proposed HOI-R1 using reinforcement learning with HOI reasoning process and HOID reward functions to solve HOID tasks purely through text without additional detection modules.

Result: On HICO-DET dataset, HOI-R1 achieves 2x the accuracy of baseline methods with strong generalization ability.

Conclusion: The approach demonstrates that language models can effectively perform HOID tasks without complex detection frameworks, opening new possibilities for simplified HOID systems.

Abstract: Recent Human-object interaction detection (HOID) methods highly require prior knowledge from VLMs to enhance the interaction recognition capabilities. The training strategies and model architectures for connecting the knowledge from VLMs to the HOI instance representations from the object detector are challenging, and the whole framework is complex for further development or application. On the other hand, the inherent reasoning abilities of MLLMs on human-object interaction detection are under-explored. Inspired by the recent success of training MLLMs with reinforcement learning (RL) methods, we propose HOI-R1 and first explore the potential of the language model on the HOID task without any additional detection modules. We introduce an HOI reasoning process and HOID reward functions to solve the HOID task by pure text. The results on the HICO-DET dataset show that HOI-R1 achieves 2x the accuracy of the baseline with great generalization ability. The source code is available at https://github.com/cjw2021/HOI-R1.

Method: Proposed a fully end-to-end data-driven framework that leverages data distribution priors to compensate for insufficient physical information from sparse transmitters, enabling feed-forward prediction of relative permittivity.

Result: Outperforms state-of-the-art approaches in reconstruction accuracy and robustness, achieving high-quality results even with a single transmitter where previous methods consistently fail.

Conclusion: Offers a fundamentally new perspective on electromagnetic inverse scattering and represents a major step toward cost-effective practical solutions for electromagnetic imaging.

Abstract: Solving Electromagnetic Inverse Scattering Problems (EISP) is fundamental in applications such as medical imaging, where the goal is to reconstruct the relative permittivity from scattered electromagnetic field. This inverse process is inherently ill-posed and highly nonlinear, making it particularly challenging, especially under sparse transmitter setups, e.g., with only one transmitter. A recent machine learning-based approach, Img-Interiors, shows promising results by leveraging continuous implicit functions. However, it requires time-consuming case-specific optimization and fails under sparse transmitter setups. To address these limitations, we revisit EISP from a data-driven perspective. The scarcity of transmitters leads to an insufficient amount of measured data, which fails to capture adequate physical information for stable inversion. Built on this insight, we propose a fully end-to-end and data-driven framework that predicts the relative permittivity of scatterers from measured fields, leveraging data distribution priors to compensate for the lack of physical information. This design enables data-driven training and feed-forward prediction of relative permittivity while maintaining strong robustness to transmitter sparsity. Extensive experiments show that our method outperforms state-of-the-art approaches in reconstruction accuracy and robustness. Notably, it achieves high-quality results even with a single transmitter, a setting where previous methods consistently fail. This work offers a fundamentally new perspective on electromagnetic inverse scattering and represents a major step toward cost-effective practical solutions for electromagnetic imaging.

Method: Uses distributed architecture with context-aware attention layers and shared gated FFN, plus early-centric sampling strategy with temperature scheduling.

Result: Achieves high-fidelity and diverse control over image generation, surpassing baselines while improving training and inference efficiency.

Conclusion: ECM provides an efficient alternative to fine-tuning for spatially-conditioned AR image generation with significant cost savings.

Abstract: Recent advances in autoregressive (AR) models have demonstrated their potential to rival diffusion models in image synthesis. However, for complex spatially-conditioned generation, current AR approaches rely on fine-tuning the pre-trained model, leading to significant training costs. In this paper, we propose the Efficient Control Model (ECM), a plug-and-play framework featuring a lightweight control module that introduces control signals via a distributed architecture. This architecture consists of context-aware attention layers that refine conditional features using real-time generated tokens, and a shared gated feed-forward network (FFN) designed to maximize the utilization of its limited capacity and ensure coherent control feature learning. Furthermore, recognizing the critical role of early-stage generation in determining semantic structure, we introduce an early-centric sampling strategy that prioritizes learning early control sequences. This approach reduces computational cost by lowering the number of training tokens per iteration, while a complementary temperature scheduling during inference compensates for the resulting insufficient training of late-stage tokens. Extensive experiments on scale-based AR models validate that our method achieves high-fidelity and diverse control over image generation, surpassing existing baselines while significantly improving both training and inference efficiency.

Method: Proposes PointNSP, a coarse-to-fine generative framework using level-of-detail principles with multi-scale factorization and next-scale prediction to preserve global shape structure at low resolutions and refine fine-grained geometry progressively.

Result: PointNSP establishes state-of-the-art generation quality on ShapeNet within the autoregressive paradigm, surpasses diffusion-based baselines in parameter, training, and inference efficiency, and shows even better performance in dense generation with 8,192 points.

Conclusion: The multi-scale factorization approach aligns autoregressive objectives with the permutation-invariant nature of point sets, enabling rich intra-scale interactions while avoiding brittle fixed orderings, demonstrating strong scalability potential.

Abstract: Autoregressive point cloud generation has long lagged behind diffusion-based approaches in quality. The performance gap stems from the fact that autoregressive models impose an artificial ordering on inherently unordered point sets, forcing shape generation to proceed as a sequence of local predictions. This sequential bias emphasizes short-range continuity but undermines the model’s capacity to capture long-range dependencies, hindering its ability to enforce global structural properties such as symmetry, consistent topology, and large-scale geometric regularities. Inspired by the level-of-detail (LOD) principle in shape modeling, we propose PointNSP, a coarse-to-fine generative framework that preserves global shape structure at low resolutions and progressively refines fine-grained geometry at higher scales through a next-scale prediction paradigm. This multi-scale factorization aligns the autoregressive objective with the permutation-invariant nature of point sets, enabling rich intra-scale interactions while avoiding brittle fixed orderings. Experiments on ShapeNet show that PointNSP establishes state-of-the-art (SOTA) generation quality for the first time within the autoregressive paradigm. In addition, it surpasses strong diffusion-based baselines in parameter, training, and inference efficiency. Finally, in dense generation with 8,192 points, PointNSP’s advantages become even more pronounced, underscoring its scalability potential.

Method: Created TFM Dataset with 15 high-resolution videos and three vision tasks. Proposed TF-Net with MobileOne-mini backbone and enhanced FPN for efficient segmentation. Designed TF-Collab pipeline integrating frame classification, pupil localization, and TFBU segmentation.

Result: TF-Net achieves favorable balance between accuracy and computational efficiency. TF-Collab provides fully automated tear film analysis. Benchmark performance established on TFM segmentation subset.

Conclusion: The proposed TF-Net and TF-Collab demonstrate effectiveness and provide foundation for future research in ocular surface diagnostics.

Abstract: Tear film break-up (TFBU) analysis is critical for diagnosing dry eye syndrome, but automated TFBU segmentation remains challenging due to the lack of annotated datasets and integrated solutions. This paper introduces the Tear Film Multi-task (TFM) Dataset, the first comprehensive dataset for multi-task tear film analysis, comprising 15 high-resolution videos (totaling 6,247 frames) annotated with three vision tasks: frame-level classification (‘clear’, ‘closed’, ‘broken’, ‘blur’), Placido Ring detection, and pixel-wise TFBU area segmentation. Leveraging this dataset, we first propose TF-Net, a novel and efficient baseline segmentation model. TF-Net incorporates a MobileOne-mini backbone with re-parameterization techniques and an enhanced feature pyramid network to achieve a favorable balance between accuracy and computational efficiency for real-time clinical applications. We further establish benchmark performance on the TFM segmentation subset by comparing TF-Net against several state-of-the-art medical image segmentation models. Furthermore, we design TF-Collab, a novel integrated real-time pipeline that synergistically leverages models trained on all three tasks of the TFM dataset. By sequentially orchestrating frame classification for BUT determination, pupil region localization for input standardization, and TFBU segmentation, TF-Collab fully automates the analysis. Experimental results demonstrate the effectiveness of the proposed TF-Net and TF-Collab, providing a foundation for future research in ocular surface diagnostics. Our code and the TFM datasets are available at https://github.com/glory-wan/TF-Net

Method: InstaGeo integrates three components: (1) automated data curation from raw satellite imagery, (2) task-specific model distillation to create compact models, and (3) seamless deployment as web-map applications.

Result: The framework achieved marginal mIoU differences of -0.73 pp for flood mapping, -0.20 pp for crop segmentation, and +1.79 pp for desert locust prediction compared to original studies. Distilled models are up to 8x smaller with reduced FLOPs and CO2 emissions. A larger crop dataset achieved state-of-the-art mIoU of 60.65%, 12 pp improvement over baselines.

Conclusion: InstaGeo transforms research-grade geospatial foundation models into practical, low-carbon tools for real-time Earth observation, enabling users to progress from raw data to deployment within a single day and shifting geospatial AI toward data quality and application-driven innovation.

Abstract: Open-access multispectral imagery from missions like Landsat 8-9 and Sentinel-2 has fueled the development of geospatial foundation models (GFMs) for humanitarian and environmental applications. Yet, their deployment remains limited by (i) the absence of automated geospatial data pipelines and (ii) the large size of fine-tuned models. Existing GFMs lack workflows for processing raw satellite imagery, and downstream adaptations often retain the full complexity of the original encoder. We present InstaGeo, an open-source, end-to-end framework that addresses these challenges by integrating: (1) automated data curation to transform raw imagery into model-ready datasets; (2) task-specific model distillation to derive compact, compute-efficient models; and (3) seamless deployment as interactive web-map applications. Using InstaGeo, we reproduced datasets from three published studies and trained models with marginal mIoU differences of -0.73 pp for flood mapping, -0.20 pp for crop segmentation, and +1.79 pp for desert locust prediction. The distilled models are up to 8x smaller than standard fine-tuned counterparts, reducing FLOPs and CO2 emissions with minimal accuracy loss. Leveraging InstaGeo’s streamlined data pipeline, we also curated a larger crop segmentation dataset, achieving a state-of-the-art mIoU of 60.65%, a 12 pp improvement over prior baselines. Moreover, InstaGeo enables users to progress from raw data to model deployment within a single working day. By unifying data preparation, model compression, and deployment, InstaGeo transforms research-grade GFMs into practical, low-carbon tools for real-time, large-scale Earth observation. This approach shifts geospatial AI toward data quality and application-driven innovation. Source code, datasets, and model checkpoints are available at: https://github.com/instadeepai/InstaGeo-E2E-Geospatial-ML.git

Method: Proposed a strategy to remove spectral peaks from images and analyzed the impact on detectors. Also introduced a simple linear detector that relies exclusively on frequency peaks as an interpretable baseline.

Result: Findings reveal that most detectors are not fundamentally dependent on spectral peaks, challenging widespread assumptions in the field.

Conclusion: The study paves the way for more transparent and reliable forensic tools by questioning the core assumption about frequency-domain artifacts in synthetic image detection.

Abstract: Over the years, the forensics community has proposed several deep learning-based detectors to mitigate the risks of generative AI. Recently, frequency-domain artifacts (particularly periodic peaks in the magnitude spectrum), have received significant attention, as they have been often considered a strong indicator of synthetic image generation. However, state-of-the-art detectors are typically used as black-boxes, and it still remains unclear whether they truly rely on these peaks. This limits their interpretability and trust. In this work, we conduct a systematic study to address this question. We propose a strategy to remove spectral peaks from images and analyze the impact of this operation on several detectors. In addition, we introduce a simple linear detector that relies exclusively on frequency peaks, providing a fully interpretable baseline free from the confounding influence of deep learning. Our findings reveal that most detectors are not fundamentally dependent on spectral peaks, challenging a widespread assumption in the field and paving the way for more transparent and reliable forensic tools.

Method: CHEST loss combines: 1) proxy-based losses in hyperbolic space, 2) proxy-based losses in Euclidean space, and 3) regularization loss based on hyperbolic hierarchical clustering. This hybrid approach leverages the strengths of both embedding spaces.

Result: The proposed method achieves new state-of-the-art performance on four benchmark datasets. The combination of hyperbolic and Euclidean spaces improves both accuracy and learning stability compared to using either space alone.

Conclusion: Combining hyperbolic and Euclidean spaces in deep metric learning through the CHEST loss framework provides significant benefits in accuracy and stability, enabling successful application of proxy-based losses in hyperbolic space for the first time.

Abstract: Deep metric learning (DML) aims to learn a neural network mapping data to an embedding space, which can represent semantic similarity between data points. Hyperbolic space is attractive for DML since it can represent richer structures, such as tree structures. DML in hyperbolic space is based on pair-based loss or unsupervised regularization loss. On the other hand, supervised proxy-based losses in hyperbolic space have not been reported yet due to some issues in applying proxy-based losses in a hyperbolic space. However, proxy-based losses are attractive for large-scale datasets since they have less training complexity. To address these, this paper proposes the Combined Hyperbolic and Euclidean Soft Triple (CHEST) loss. CHEST loss is composed of the proxy-based losses in hyperbolic and Euclidean spaces and the regularization loss based on hyperbolic hierarchical clustering. We find that the combination of hyperbolic and Euclidean spaces improves DML accuracy and learning stability for both spaces. Finally, we evaluate the CHEST loss on four benchmark datasets, achieving a new state-of-the-art performance.

Method: Two complementary deep learning frameworks: 1) AHP-CADNet - multi-level attention fusion framework integrating ocular landmarks, head pose features, and clinical attributes; 2) Curriculum learning-based imputation framework that progressively leverages structured variables and unstructured clinical notes to handle missing data.

Result: AHP-CADNet achieved 96.9-99.0% accuracy across classification tasks and low prediction errors (MAE: 0.103-0.199, R2 > 0.93). The imputation framework maintained 93.46-99.78% accuracy across clinical variables with PubMedBERT, showing significant improvements through clinical dependency modeling (p < 0.001).

Conclusion: Both frameworks are effective for automated diagnosis and recovery from missing data in clinical settings, demonstrating robust performance on the PoseGaze-AHP dataset and addressing key challenges in AHP assessment.

Abstract: Ocular-induced abnormal head posture (AHP) is a compensatory mechanism that arises from ocular misalignment conditions, such as strabismus, enabling patients to reduce diplopia and preserve binocular vision. Early diagnosis minimizes morbidity and secondary complications such as facial asymmetry; however, current clinical assessments remain largely subjective and are further complicated by incomplete medical records. This study addresses both challenges through two complementary deep learning frameworks. First, AHP-CADNet is a multi-level attention fusion framework for automated diagnosis that integrates ocular landmarks, head pose features, and structured clinical attributes to generate interpretable predictions. Second, a curriculum learning-based imputation framework is designed to mitigate missing data by progressively leveraging structured variables and unstructured clinical notes to enhance diagnostic robustness under realistic data conditions. Evaluation on the PoseGaze-AHP dataset demonstrates robust diagnostic performance. AHP-CADNet achieves 96.9-99.0 percent accuracy across classification tasks and low prediction errors for continuous variables, with MAE ranging from 0.103 to 0.199 and R2 exceeding 0.93. The imputation framework maintains high accuracy across all clinical variables (93.46-99.78 percent with PubMedBERT), with clinical dependency modeling yielding significant improvements (p < 0.001). These findings confirm the effectiveness of both frameworks for automated diagnosis and recovery from missing data in clinical settings.

Method: Built on a layered CIE (Cognition-Interaction-Evolution) architecture with user-defined customization for environment, agents, and sessions. Features human-in-the-loop interface allowing real users to join the simulation space.

Result: Validated in middle-school Chinese classes: (1) Instructional alignment with simulated IRF rates matching real classrooms; (2) Realistic group interaction and role differentiation; (3) Cross-session evolution with 11.7% average increase in positive transition rates, capturing longitudinal shifts in behavior, emotion, and cognition.

Conclusion: EduVerse balances realism, reproducibility, and interpretability, providing a scalable platform for educational AI that will be open-sourced to foster cross-disciplinary research.

Abstract: Reproducing cognitive development, group interaction, and long-term evolution in virtual classrooms remains a core challenge for educational AI, as real classrooms integrate open-ended cognition, dynamic social interaction, affective factors, and multi-session development rarely captured together. Existing approaches mostly focus on short-term or single-agent settings, limiting systematic study of classroom complexity and cross-task reuse. We present EduVerse, the first user-defined multi-agent simulation space that supports environment, agent, and session customization. A distinctive human-in-the-loop interface further allows real users to join the space. Built on a layered CIE (Cognition-Interaction-Evolution) architecture, EduVerse ensures individual consistency, authentic interaction, and longitudinal adaptation in cognition, emotion, and behavior-reproducing realistic classroom dynamics with seamless human-agent integration. We validate EduVerse in middle-school Chinese classes across three text genres, environments, and multiple sessions. Results show: (1) Instructional alignment: simulated IRF rates (0.28-0.64) closely match real classrooms (0.37-0.49), indicating pedagogical realism; (2) Group interaction and role differentiation: network density (0.27-0.40) with about one-third of peer links realized, while human-agent tasks indicate a balance between individual variability and instructional stability; (3) Cross-session evolution: the positive transition rate R+ increase by 11.7% on average, capturing longitudinal shifts in behavior, emotion, and cognition and revealing structured learning trajectories. Overall, EduVerse balances realism, reproducibility, and interpretability, providing a scalable platform for educational AI. The system will be open-sourced to foster cross-disciplinary research.

Method: Proposes SD-MVSum with weighted cross-modal attention mechanism that models dependencies between script-video and script-transcript pairs, explicitly exploiting semantic similarity to highlight most relevant video segments.

Result: Experimental comparisons show SD-MVSum is competitive against SOTA approaches for both script-driven and generic video summarization, with extended datasets S-VideoXum and MrHiSum supporting multimodal training.

Conclusion: The proposed multimodal approach effectively leverages both visual and spoken content relevance for script-driven video summarization, with available code and extended datasets for further research.

Abstract: In this work, we extend a recent method for script-driven video summarization, originally considering just the visual content of the video, to take into account the relevance of the user-provided script also with the video’s spoken content. In the proposed method, SD-MVSum, the dependence between each considered pair of data modalities, i.e., script-video and script-transcript, is modeled using a new weighted cross-modal attention mechanism. This explicitly exploits the semantic similarity between the paired modalities in order to promote the parts of the full-length video with the highest relevance to the user-provided script. Furthermore, we extend two large-scale datasets for video summarization (S-VideoXum, MrHiSum), to make them suitable for training and evaluation of script-driven multimodal video summarization methods. Experimental comparisons document the competitiveness of our SD-MVSum method against other SOTA approaches for script-driven and generic video summarization. Our new method and extended datasets are available at: https://github.com/IDT-ITI/SD-MVSum.

Method: ARGUS captures macro-meso-micro hierarchical information through: 1) micro-geometry features for cell-level patterns via geometric structure across nuclei, 2) Hierarchical Field-of-Views Alignment module for macro-meso level interactions, and 3) Geometry Prior Guided Fusion strategy to combine features for holistic phenotype modeling.

Result: Extensive experiments on public and private cohorts demonstrate that ARGUS achieves state-of-the-art performance in histological subtyping of liver cancer.

Conclusion: ARGUS provides an effective diagnostic tool for primary liver malignancies in clinical practice by better capturing the complex hierarchical information in liver cancer histology.

Abstract: Primary liver malignancies are widely recognized as the most heterogeneous and prognostically diverse cancers of the digestive system. Among these, hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC) emerge as the two principal histological subtypes, demonstrating significantly greater complexity in tissue morphology and cellular architecture than other common tumors. The intricate representation of features in Whole Slide Images (WSIs) encompasses abundant crucial information for liver cancer histological subtyping, regarding hierarchical pyramid structure, tumor microenvironment (TME), and geometric representation. However, recent approaches have not adequately exploited these indispensable effective descriptors, resulting in a limited understanding of histological representation and suboptimal subtyping performance. To mitigate these limitations, ARGUS is proposed to advance histological subtyping in liver cancer by capturing the macro-meso-micro hierarchical information within the TME. Specifically, we first construct a micro-geometry feature to represent fine-grained cell-level pattern via a geometric structure across nuclei, thereby providing a more refined and precise perspective for delineating pathological images. Then, a Hierarchical Field-of-Views (FoVs) Alignment module is designed to model macro- and meso-level hierarchical interactions inherent in WSIs. Finally, the augmented micro-geometry and FoVs features are fused into a joint representation via present Geometry Prior Guided Fusion strategy for modeling holistic phenotype interactions. Extensive experiments on public and private cohorts demonstrate that our ARGUS achieves state-of-the-art (SOTA) performance in histological subtyping of liver cancer, which provide an effective diagnostic tool for primary liver malignancies in clinical practice.

Method: Uses pre-trained text-to-image diffusion models with inversion techniques, classifier-free guidance for affordance-aware global editing, and mask-guided self-attention for high-quality personalization from single reference images.

Result: Achieves realistic human insertions into complex scenes with excellent identity preservation of subjects’ clothing and body features, outperforming previous methods.

Conclusion: Demonstrates that diffusion models inherently possess knowledge to place people in scenes without training, enabling training-free state-of-the-art human insertion.

Abstract: The task of realistically inserting a human from a reference image into a background scene is highly challenging, requiring the model to (1) determine the correct location and poses of the person and (2) perform high-quality personalization conditioned on the background. Previous approaches often treat them as separate problems, overlooking their interconnections, and typically rely on training to achieve high performance. In this work, we introduce a unified training-free pipeline that leverages pre-trained text-to-image diffusion models. We show that diffusion models inherently possess the knowledge to place people in complex scenes without requiring task-specific training. By combining inversion techniques with classifier-free guidance, our method achieves affordance-aware global editing, seamlessly inserting people into scenes. Furthermore, our proposed mask-guided self-attention mechanism ensures high-quality personalization, preserving the subject’s identity, clothing, and body features from just a single reference image. To the best of our knowledge, we are the first to perform realistic human insertions into scenes in a training-free manner and achieve state-of-the-art results in diverse composite scene images with excellent identity preservation in backgrounds and subjects.

Method: Deployed multiple cameras for continuous imaging and developed a novel image analysis pipeline that integrates morphological and spatial features with temporal data to identify individual seedlings.

Result: R. solani significantly reduced germination rates and seedling vigor. The method achieved high accuracy (R²=0.98, RMSE=1.12) in counting seedlings, especially in dense growth conditions where traditional methods fail.

Conclusion: Combining low-cost imaging hardware with advanced computational tools enables reliable, non-destructive phenotyping, with temporal integration proving crucial for accurate seedling quantification in complex scenarios.

Abstract: The study investigates the effects of R. solani inoculation on the germination and early development of Lactuca sativa L. seeds using a low-cost, image-based monitoring system. Multiple cameras were deployed to continuously capture images of the germination process in both infected and control groups. The objective was to assess the impact of the pathogen by analyzing germination dynamics and growth over time. To achieve this, a novel image analysis pipeline was developed. The algorithm integrates both morphological and spatial features to identify and quantify individual seedlings, even under complex conditions where traditional image analyses fails. A key innovation of the method lies in its temporal integration: each analysis step considers not only the current status but also their developmental across prior time points. This approach enables robust discrimination of individual seedlings, especially when overlapping leaves significantly hinder object separation. The method demonstrated high accuracy in seedling counting and vigor assessment, even in challenging scenarios characterized by dense and intertwined growth. Results confirm that R. solani infection significantly reduces germination rates and early seedling vigor. The study also validates the feasibility of combining low-cost imaging hardware with advanced computational tools to obtain phenotyping data in a non-destructive and scalable manner. The temporal integration enabled accurate quantification of germinated seeds and precise determination of seedling emergence timing. This approach proved particularly effective in later stages of the experiment, where conventional segmentation techniques failed due to overlapping or intertwined seedlings, making accurate counting. The method achieved a coefficient of determination of 0.98 and a root mean square error (RMSE) of 1.12, demonstrating its robustness and reliability.

Method: Proposes object-level masked image modeling (MIM) where masks are applied to visual objects rather than random patches, treating objects as the visual equivalent of words to learn global context and semantics.

Result: Object-level representation alone helps learn real-world distributions, while pixel-averaging shortcuts are learned without it. Evaluations with MLLMs on VQA, GQA, and ScienceQA show strong reasoning and contextual understanding gains.

Conclusion: Object-level encoding is effective for developing stronger vision encoders and tokenizers, providing a direction to narrow the gap between vision and language models in reasoning capabilities.

Abstract: Recent advances in language modeling have witnessed the rise of highly desirable emergent capabilities, such as reasoning and in-context learning. However, vision models have yet to exhibit comparable progress in these areas. In this paper, we argue that this gap could stem from the lack of semantic and contextual guidance in current vision transformer (ViT) training schemes, and such a gap can be narrowed through the design of a semantic-grounded objective. Specifically, we notice that individual words in natural language are inherently semantic, and modeling directly on word tokens naturally learns a realistic distribution. In contrast, ViTs rely on spatial patchification, which inevitably lacks semantic information. To bridge this gap, we propose to directly model “object” as the visual equivalence of “word,” pushing the model to learn the global context and semantics among visual elements. We investigate our hypotheses via masked image modeling (MIM), a framework where our approach can be readily tested by applying masks to visual objects rather than random patches. Considerable evidence from qualitative and quantitative evaluations reveals a key finding: object-level representation alone helps to learn a real-world distribution, whereas pixel-averaging shortcuts are often learned without it. Moreover, further evaluations with multimodal LLMs (MLLM) on visual question answering (VQA, GQA, ScienceQA) tasks demonstrate the strong reasoning and contextual understanding gained with this simple objective. We hope our study highlights the effectiveness of object-level encoding and provides a plausible direction for developing stronger vision encoders and tokenizers. Code and model will be publicly released. Keywords: Semantic Visual Tokenizer, Vision Reasoning, In-context Learning, Multimodal Reasoning

Method: Uses age-conditioned prompt blending and age-specific LoRA fusion with SVDMix matrix fusion technique to exploit the linear nature of aging, enabling generation of intermediate-age portraits from a single reference image.

Result: Produces realistic and identity-consistent face images across different ages from a single reference image, achieving superior age control and visual quality compared to previous state-of-the-art editing-based methods.

Conclusion: AgeBooth effectively enhances age control capability without expensive age-varied datasets, demonstrating the viability of linear aging modeling through prompt blending and LoRA fusion strategies.

Abstract: Recent diffusion model research focuses on generating identity-consistent images from a reference photo, but they struggle to accurately control age while preserving identity, and fine-tuning such models often requires costly paired images across ages. In this paper, we propose AgeBooth, a novel age-specific finetuning approach that can effectively enhance the age control capability of adapterbased identity personalization models without the need for expensive age-varied datasets. To reduce dependence on a large amount of age-labeled data, we exploit the linear nature of aging by introducing age-conditioned prompt blending and an age-specific LoRA fusion strategy that leverages SVDMix, a matrix fusion technique. These techniques enable high-quality generation of intermediate-age portraits. Our AgeBooth produces realistic and identity-consistent face images across different ages from a single reference image. Experiments show that AgeBooth achieves superior age control and visual quality compared to previous state-of-the-art editing-based methods.

Method: Integrates pretrained ControlNet and VLMs to generate synthetic images with pixel-level labels, adding Multi-way Prompt Generator, Mask Generator, and High-quality Image Selection modules for improved fidelity and diversity.

Result: Achieves promising performance on PASCAL-5i and COCO-20i datasets, outperforming concurrent work for one-shot semantic segmentation.

Conclusion: The proposed training-free pipeline effectively generates diverse synthetic data with pixel-level annotations, eliminating manual annotation needs and improving downstream segmentation tasks.

Abstract: Generating enough and diverse data through augmentation offers an efficient solution to the time-consuming and labour-intensive process of collecting and annotating pixel-wise images. Traditional data augmentation techniques often face challenges in manipulating high-level semantic attributes, such as materials and textures. In contrast, diffusion models offer a robust alternative, by effectively utilizing text-to-image or image-to-image transformation. However, existing diffusion-based methods are either computationally expensive or compromise on performance. To address this issue, we introduce a novel training-free pipeline that integrates pretrained ControlNet and Vision-Language Models (VLMs) to generate synthetic images paired with pixel-level labels. This approach eliminates the need for manual annotations and significantly improves downstream tasks. To improve the fidelity and diversity, we add a Multi-way Prompt Generator, Mask Generator and High-quality Image Selection module. Our results on PASCAL-5i and COCO-20i present promising performance and outperform concurrent work for one-shot semantic segmentation.

Method: FusionDetect uses two frozen foundation models (CLIP and Dinov2) to derive complementary features, creating a cohesive feature space that adapts to changes in both content and generator design. The approach is evaluated on the new OmniGen Benchmark with 12 state-of-the-art generators.

Result: FusionDetect achieves 3.87% higher accuracy and 6.13% higher precision than the closest competitor on established benchmarks, and 4.48% higher accuracy on OmniGen. It also shows exceptional robustness to common image perturbations.

Conclusion: The paper introduces both a top-performing detector (FusionDetect) and a new benchmark (OmniGen) that together provide a framework for advancing universal AI image detection, addressing both cross-generator and cross-domain generalization challenges.

Abstract: The rapid development of generative models has made it increasingly crucial to develop detectors that can reliably detect synthetic images. Although most of the work has now focused on cross-generator generalization, we argue that this viewpoint is too limited. Detecting synthetic images involves another equally important challenge: generalization across visual domains. To bridge this gap,we present the OmniGen Benchmark. This comprehensive evaluation dataset incorporates 12 state-of-the-art generators, providing a more realistic way of evaluating detector performance under realistic conditions. In addition, we introduce a new method, FusionDetect, aimed at addressing both vectors of generalization. FusionDetect draws on the benefits of two frozen foundation models: CLIP & Dinov2. By deriving features from both complementary models,we develop a cohesive feature space that naturally adapts to changes in both thecontent and design of the generator. Our extensive experiments demonstrate that FusionDetect delivers not only a new state-of-the-art, which is 3.87% more accurate than its closest competitor and 6.13% more precise on average on established benchmarks, but also achieves a 4.48% increase in accuracy on OmniGen,along with exceptional robustness to common image perturbations. We introduce not only a top-performing detector, but also a new benchmark and framework for furthering universal AI image detection. The code and dataset are available at http://github.com/amir-aman/FusionDetect

Method: Uses Vision Foundation Models guided by text and images to generate initial pseudo-labels, refines them through spatio-temporal voting combining 2D and 3D semantics, and employs 2D prior-based losses and prototype-based contrastive loss for feature learning.

Result: Achieves state-of-the-art performance for unsupervised 3D instance segmentation, outperforming MWSIS (which uses ground-truth 2D bounding box supervision) by 2.53% in mAP (50.95% vs. 48.42%).

Conclusion: ALISE successfully demonstrates that high-quality unsupervised LiDAR instance segmentation is achievable through comprehensive pseudo-label generation and refinement strategies, eliminating the need for costly manual annotations.

Abstract: The manual annotation of outdoor LiDAR point clouds for instance segmentation is extremely costly and time-consuming. Current methods attempt to reduce this burden but still rely on some form of human labeling. To completely eliminate this dependency, we introduce ALISE, a novel framework that performs LiDAR instance segmentation without any annotations. The central challenge is to generate high-quality pseudo-labels in a fully unsupervised manner. Our approach starts by employing Vision Foundation Models (VFMs), guided by text and images, to produce initial pseudo-labels. We then refine these labels through a dedicated spatio-temporal voting module, which combines 2D and 3D semantics for both offline and online optimization. To achieve superior feature learning, we further introduce two forms of semantic supervision: a set of 2D prior-based losses that inject visual knowledge into the 3D network, and a novel prototype-based contrastive loss that builds a discriminative feature space by exploiting 3D semantic consistency. This comprehensive design results in significant performance gains, establishing a new state-of-the-art for unsupervised 3D instance segmentation. Remarkably, our approach even outperforms MWSIS, a method that operates with supervision from ground-truth (GT) 2D bounding boxes by a margin of 2.53% in mAP (50.95% vs. 48.42%).

Method: Proposes VRQ (vision-aligned residual quantization) encoding to align different object representations across viewpoints while preserving product distinctiveness, and uses multi-stage semantic alignment to maintain visual similarity while incorporating user-specific preferences.

Result: In offline evaluations: performs on par with online MCA while improving inference efficiency by 21% through dynamic pruning. In A/B tests: +2.15% item CTR, +2.27% CVR, and +3.12% order volume improvements.

Conclusion: A semantic ID centric, generative architecture can successfully unify retrieval and personalization while simplifying the serving pathway, demonstrating superior performance over traditional multi-stage approaches.

Abstract: Traditional vision search, similar to search and recommendation systems, follows the multi-stage cascading architecture (MCA) paradigm to balance efficiency and conversion. Specifically, the query image undergoes feature extraction, recall, pre-ranking, and ranking stages, ultimately presenting the user with semantically similar products that meet their preferences. This multi-view representation discrepancy of the same object in the query and the optimization objective collide across these stages, making it difficult to achieve Pareto optimality in both user experience and conversion. In this paper, an end-to-end generative framework, OneVision, is proposed to address these problems. OneVision builds on VRQ, a vision-aligned residual quantization encoding, which can align the vastly different representations of an object across multiple viewpoints while preserving the distinctive features of each product as much as possible. Then a multi-stage semantic alignment scheme is adopted to maintain strong visual similarity priors while effectively incorporating user-specific information for personalized preference generation. In offline evaluations, OneVision performs on par with online MCA, while improving inference efficiency by 21% through dynamic pruning. In A/B tests, it achieves significant online improvements: +2.15% item CTR, +2.27% CVR, and +3.12% order volume. These results demonstrate that a semantic ID centric, generative architecture can unify retrieval and personalization while simplifying the serving pathway.

Method: Combine multiple weak labeled sources with a small reliable dataset, embed transition matrices describing error statistics into labels, and use them as weighting scheme at gradient level during training.

Result: Experiments on different datasets validated the method’s effectiveness, proving robustness and capability to leverage unreliable label sources.

Conclusion: The proposed training strategy successfully enables deep network training with unreliable label sources by accounting for source reliability through transition matrices.

Abstract: Deep learning has gained broad interest in remote sensing image scene classification thanks to the effectiveness of deep neural networks in extracting the semantics from complex data. However, deep networks require large amounts of training samples to obtain good generalization capabilities and are sensitive to errors in the training labels. This is a problem in remote sensing since highly reliable labels can be obtained at high costs and in limited amount. However, many sources of less reliable labeled data are available, e.g., obsolete digital maps. In order to train deep networks with larger datasets, we propose both the combination of single or multiple weak sources of labeled data with a small but reliable dataset to generate multisource labeled datasets and a novel training strategy where the reliability of each source is taken in consideration. This is done by exploiting the transition matrices describing the statistics of the errors of each source. The transition matrices are embedded into the labels and used during the training process to weigh each label according to the related source. The proposed method acts as a weighting scheme at gradient level, where each instance contributes with different weights to the optimization of different classes. The effectiveness of the proposed method is validated by experiments on different datasets. The results proved the robustness and capability of leveraging on unreliable source of labels of the proposed method.

Method: Introduce an entropy-based criterion to automatically identify layers offering the most complementary information in a training-free setting without access to OOD data. Selectively incorporate these intermediate representations.

Result: Improves OOD detection accuracy by up to 10% in far-OOD and over 7% in near-OOD benchmarks compared to state-of-the-art training-free methods across various model architectures and training objectives.

Conclusion: Reveals a new avenue for OOD detection research and uncovers the impact of various training objectives and model architectures on confidence-based OOD detection methods.

Abstract: Out-of-distribution (OOD) detection is essential for reliably deploying machine learning models in the wild. Yet, most methods treat large pre-trained models as monolithic encoders and rely solely on their final-layer representations for detection. We challenge this wisdom. We reveal the \textit{intermediate layers} of pre-trained models, shaped by residual connections that subtly transform input projections, \textit{can} encode \textit{surprisingly rich and diverse signals} for detecting distributional shifts. Importantly, to exploit latent representation diversity across layers, we introduce an entropy-based criterion to \textit{automatically} identify layers offering the most complementary information in a training-free setting – \textit{without access to OOD data}. We show that selectively incorporating these intermediate representations can increase the accuracy of OOD detection by up to \textbf{$10%$} in far-OOD and over \textbf{$7%$} in near-OOD benchmarks compared to state-of-the-art training-free methods across various model architectures and training objectives. Our findings reveal a new avenue for OOD detection research and uncover the impact of various training objectives and model architectures on confidence-based OOD detection methods.

Method: Combines rasterized Gaussian kernel rendering efficiency with edge-aware gating mechanism, replacing global iterative optimization with rasterized formulation.

Result: Achieves significantly faster parameter updates, more memory-efficient representations, and supports native super-resolution and denoising not possible with standard rasterized approaches.

Conclusion: Provides a new balance between computational efficiency and reconstruction fidelity for 2D image processing by combining fast rasterized optimization with edge-aware structure.

Abstract: The Steered Mixture of Experts regression framework has demonstrated strong performance in image reconstruction, compression, denoising, and super-resolution. However, its high computational cost limits practical applications. This work introduces a rasterization-based optimization strategy that combines the efficiency of rasterized Gaussian kernel rendering with the edge-aware gating mechanism of the Steered Mixture of Experts. The proposed method is designed to accelerate two-dimensional image regression while maintaining the model’s inherent sparsity and reconstruction quality. By replacing global iterative optimization with a rasterized formulation, the method achieves significantly faster parameter updates and more memory-efficient model representations. In addition, the proposed framework supports applications such as native super-resolution and image denoising, which are not directly achievable with standard rasterized Gaussian kernel approaches. The combination of fast rasterized optimization with the edge-aware structure of the Steered Mixture of Experts provides a new balance between computational efficiency and reconstruction fidelity for two-dimensional image processing tasks.

Method: Self-supervised deep learning using dense deformable registration fields to compute 1D motion descriptors, from which keyframes are determined using rule-based analysis of characteristic contraction/relaxation patterns.

Result: Achieved 30-51% improvement in SAX and 11-47% improvement in 4CH for ED/ES detection, with mean cyclic frame difference below 1.31 frames for SAX and 1.73 for LAX across multiple datasets.

Conclusion: The approach enables temporally aligned inter- and intra-patient analysis of cardiac dynamics independent of cycle or phase lengths, providing deeper insights into myocardial motion.

Abstract: Cardiovascular magnetic resonance (CMR) is the gold standard for assessing cardiac function, but individual cardiac cycles complicate automatic temporal comparison or sub-phase analysis. Accurate cardiac keyframe detection can eliminate this problem. However, automatic methods solely derive end-systole (ES) and end-diastole (ED) frames from left ventricular volume curves, which do not provide a deeper insight into myocardial motion. We propose a self-supervised deep learning method detecting five keyframes in short-axis (SAX) and four-chamber long-axis (4CH) cine CMR. Initially, dense deformable registration fields are derived from the images and used to compute a 1D motion descriptor, which provides valuable insights into global cardiac contraction and relaxation patterns. From these characteristic curves, keyframes are determined using a simple set of rules. The method was independently evaluated for both views using three public, multicentre, multidisease datasets. M&Ms-2 (n=360) dataset was used for training and evaluation, and M&Ms (n=345) and ACDC (n=100) datasets for repeatability control. Furthermore, generalisability to patients with rare congenital heart defects was tested using the German Competence Network (GCN) dataset. Our self-supervised approach achieved improved detection accuracy by 30% - 51% for SAX and 11% - 47% for 4CH in ED and ES, as measured by cyclic frame difference (cFD), compared with the volume-based approach. We can detect ED and ES, as well as three additional keyframes throughout the cardiac cycle with a mean cFD below 1.31 frames for SAX and 1.73 for LAX. Our approach enables temporally aligned inter- and intra-patient analysis of cardiac dynamics, irrespective of cycle or phase lengths. GitHub repository: https://github.com/Cardio-AI/cmr-multi-view-phase-detection.git

Method: Two core modules: Temporal Granularity Optimization (TGO) uses coarse flow priors to group similar frames and semantic priors to filter irrelevant scenes; Motion Token Pruning (MTP) refines intra-frame representations by pruning redundant tokens using fine-grained optical flow.

Result: Outperforms existing methods across video MLLM benchmarks: 64.7% on Video-MME, 71.4% on MLVU, and 60.4% on LongVideoBench, especially effective for hour-level video understanding.

Conclusion: Incorporating motion priors from optical flow effectively mitigates redundancy in long videos and significantly improves video understanding performance for MLLMs.

Abstract: Long-form video understanding has always been a challenging problem due to the significant redundancy in both temporal and spatial contents. This challenge is further exacerbated by the limited context length of Multimodal Large Language Models (MLLMs). To address this issue, many previous works have attempted to extract key video information, where the “key” is typically semantic-aware and heavily dependent on the CLIP model as prior. In this paper, we propose Flow4Agent, a novel framework that pioneeringly incorporates motion priors from optical flow to facilitate LLM-based long video understanding. Flow4Agent mitigates the redundancy in long videos at both temporal and spatial levels through two core modules: Temporal Granularity Optimization (TGO) adaptively refines framelevel hierarchies, which first leverages coarse flow priors to group similar visual contents and then applies semantic priors to filter out highly irrelevant scene information. Motion Token Pruning (MTP) further refines the intra-frame visual representations, pruning high-redundancy video tokens using fine-grained optical flow information. Extensive experiments demonstrate that our Flow4Agent outperforms existing methods across a wide range of video MLLM benchmarks, especially for hour-level video understanding tasks, achieving 64.7% on Video-MME, 71.4% on MLVU and 60.4% on LongVideoBench.

Method: Developed a platform with three components: (1) acia Python library with 8 deep learning segmentation/tracking approaches, (2) Jupyter Notebook workflows combining analysis pipelines with dependencies and visualizations, and (3) application workflows for real-world use cases including microfluidic experiments.

Result: Created more than ten open-source application workflows that enable analysis ranging from growth rate comparisons to minute-resolution quantitative analysis of individual cell responses to changing oxygen conditions.

Conclusion: The acia-workflows platform successfully integrates powerful deep learning tools into accessible, flexible workflows that support routine application in biological research, making automated live-cell imaging analysis more practical and reproducible.

Abstract: Live-cell imaging (LCI) technology enables the detailed spatio-temporal characterization of living cells at the single-cell level, which is critical for advancing research in the life sciences, from biomedical applications to bioprocessing. High-throughput setups with tens to hundreds of parallel cell cultivations offer the potential for robust and reproducible insights. However, these insights are obscured by the large amount of LCI data recorded per experiment. Recent advances in state-of-the-art deep learning methods for cell segmentation and tracking now enable the automated analysis of such large data volumes, offering unprecedented opportunities to systematically study single-cell dynamics. The next key challenge lies in integrating these powerful tools into accessible, flexible, and user-friendly workflows that support routine application in biological research. In this work, we present acia-workflows, a platform that combines three key components: (1) the Automated live-Cell Imaging Analysis (acia) Python library, which supports the modular design of image analysis pipelines offering eight deep learning segmentation and tracking approaches; (2) workflows that assemble the image analysis pipeline, its software dependencies, documentation, and visualizations into a single Jupyter Notebook, leading to accessible, reproducible and scalable analysis workflows; and (3) a collection of application workflows showcasing the analysis and customization capabilities in real-world applications. Specifically, we present three workflows to investigate various types of microfluidic LCI experiments ranging from growth rate comparisons to precise, minute-resolution quantitative analyses of individual dynamic cells responses to changing oxygen conditions. Our collection of more than ten application workflows is open source and publicly available at https://github.com/JuBiotech/acia-workflows.

Method: Proposes BioAutoML-NAS using neural architecture search to automatically learn optimal operations for each connection in cells, with multimodal fusion combining image embeddings and metadata, and alternating bi-level optimization training with zero operations for sparsity.

Result: Achieves 96.81% accuracy, 97.46% precision, 96.81% recall, and 97.05% F1 score on BIOSCAN-5M dataset, outperforming state-of-the-art methods by 16%, 10%, and 8% respectively. On Insects-1M dataset: 93.25% accuracy, 93.71% precision, 92.74% recall, and 93.22% F1 score.

Conclusion: BioAutoML-NAS provides accurate, confident insect classification that supports modern sustainable farming through its automated architecture search and multimodal fusion approach.

Abstract: Insect classification is important for agricultural management and ecological research, as it directly affects crop health and production. However, this task remains challenging due to the complex characteristics of insects, class imbalance, and large-scale datasets. To address these issues, we propose BioAutoML-NAS, the first BioAutoML model using multimodal data, including images, and metadata, which applies neural architecture search (NAS) for images to automatically learn the best operations for each connection within each cell. Multiple cells are stacked to form the full network, each extracting detailed image feature representations. A multimodal fusion module combines image embeddings with metadata, allowing the model to use both visual and categorical biological information to classify insects. An alternating bi-level optimization training strategy jointly updates network weights and architecture parameters, while zero operations remove less important connections, producing sparse, efficient, and high-performing architectures. Extensive evaluation on the BIOSCAN-5M dataset demonstrates that BioAutoML-NAS achieves 96.81% accuracy, 97.46% precision, 96.81% recall, and a 97.05% F1 score, outperforming state-of-the-art transfer learning, transformer, AutoML, and NAS methods by approximately 16%, 10%, and 8% respectively. Further validation on the Insects-1M dataset obtains 93.25% accuracy, 93.71% precision, 92.74% recall, and a 93.22% F1 score. These results demonstrate that BioAutoML-NAS provides accurate, confident insect classification that supports modern sustainable farming.

Method: Uses Discrete Distribution Discrepancy-aware Quantization Error (D³QE) with a transformer that integrates dynamic codebook frequency statistics into attention mechanism, fusing semantic features and quantization error latent.

Result: Demonstrates superior detection accuracy and strong generalization across 7 mainstream AR models, with robustness to real-world perturbations on the ARForensics dataset.

Conclusion: D³QE effectively exploits distinctive patterns and frequency distribution bias in codebooks to detect autoregressive-generated images with high accuracy and robustness.

Abstract: The emergence of visual autoregressive (AR) models has revolutionized image generation while presenting new challenges for synthetic image detection. Unlike previous GAN or diffusion-based methods, AR models generate images through discrete token prediction, exhibiting both marked improvements in image synthesis quality and unique characteristics in their vector-quantized representations. In this paper, we propose to leverage Discrete Distribution Discrepancy-aware Quantization Error (D$^3$QE) for autoregressive-generated image detection that exploits the distinctive patterns and the frequency distribution bias of the codebook existing in real and fake images. We introduce a discrete distribution discrepancy-aware transformer that integrates dynamic codebook frequency statistics into its attention mechanism, fusing semantic features and quantization error latent. To evaluate our method, we construct a comprehensive dataset termed ARForensics covering 7 mainstream visual AR models. Experiments demonstrate superior detection accuracy and strong generalization of D$^3$QE across different AR models, with robustness to real-world perturbations. Code is available at \href{https://github.com/Zhangyr2022/D3QE}{https://github.com/Zhangyr2022/D3QE}.

Method: Proposed Weakly Supervised In-Context Learning (WS-ICL) paradigm that leverages weak prompts (bounding boxes or points) instead of dense labels for context, eliminating need for fine-grained masks and repeated user prompting.

Result: WS-ICL achieves performance comparable to regular ICL models at significantly lower annotation cost, and is highly competitive under interactive paradigm on three held-out benchmarks.

Conclusion: WS-ICL establishes a promising step toward more efficient and unified universal models for medical image segmentation by reducing annotation effort while maintaining strong performance.

Abstract: Universal models for medical image segmentation, such as interactive and in-context learning (ICL) models, offer strong generalization but require extensive annotations. Interactive models need repeated user prompts for each image, while ICL relies on dense, pixel-level labels. To address this, we propose Weakly Supervised In-Context Learning (WS-ICL), a new ICL paradigm that leverages weak prompts (e.g., bounding boxes or points) instead of dense labels for context. This approach significantly reduces annotation effort by eliminating the need for fine-grained masks and repeated user prompting for all images. We evaluated the proposed WS-ICL model on three held-out benchmarks. Experimental results demonstrate that WS-ICL achieves performance comparable to regular ICL models at a significantly lower annotation cost. In addition, WS-ICL is highly competitive even under the interactive paradigm. These findings establish WS-ICL as a promising step toward more efficient and unified universal models for medical image segmentation. Our code and model are publicly available at https://github.com/jiesihu/Weak-ICL.

Method: Introduces a new benchmark dataset with over 230,000 images (29,000 defective instances) and 48,000 distinct objects, then evaluates multiple state-of-the-art anomaly detection methods on this dataset to demonstrate their limitations.

Result: State-of-the-art anomaly detection methods achieve only 56.96% AUROC on the new dataset, significantly lower than their near-perfect performance on existing benchmarks. Qualitative analysis confirms these methods struggle with heavy pose and appearance variation.

Conclusion: The paper establishes a new challenging benchmark for anomaly detection in retail logistics that highlights the limitations of current methods and encourages future research to address the unique challenges of this domain.

Abstract: We present a novel large-scale dataset for defect detection in a logistics setting. Recent work on industrial anomaly detection has primarily focused on manufacturing scenarios with highly controlled poses and a limited number of object categories. Existing benchmarks like MVTec-AD [6] and VisA [33] have reached saturation, with state-of-the-art methods achieving up to 99.9% AUROC scores. In contrast to manufacturing, anomaly detection in retail logistics faces new challenges, particularly in the diversity and variability of object pose and appearance. Leading anomaly detection methods fall short when applied to this new setting. To bridge this gap, we introduce a new benchmark that overcomes the current limitations of existing datasets. With over 230,000 images (and more than 29,000 defective instances), it is 40 times larger than MVTec-AD and contains more than 48,000 distinct objects. To validate the difficulty of the problem, we conduct an extensive evaluation of multiple state-of-the-art anomaly detection methods, demonstrating that they do not surpass 56.96% AUROC on our dataset. Further qualitative analysis confirms that existing methods struggle to leverage normal samples under heavy pose and appearance variation. With our large-scale dataset, we set a new benchmark and encourage future research towards solving this challenging problem in retail logistics anomaly detection. The dataset is available for download under https://www.kaputt-dataset.com.

Method: Systematically analyzed pooling-, convolution-, and attention-based token mixers within MetaFormer architecture on image classification and semantic segmentation across 8 diverse medical datasets. Also examined transfer learning from pretrained weights to new token mixers.

Result: For classification, low-complexity token mixers (grouped convolution or pooling) are sufficient. Pretrained weights remain useful despite domain gap. For segmentation, convolutional token mixers with local inductive bias are essential, with grouped convolutions emerging as preferred choice due to reduced runtime and parameters.

Conclusion: Grouped convolutions are the optimal token mixer choice for medical imaging tasks, providing efficiency benefits while MetaFormer’s channel-MLPs handle cross-channel interactions. The study provides guidance for token mixer selection in medical vision applications.

Abstract: The generalization of the Transformer architecture via MetaFormer has reshaped our understanding of its success in computer vision. By replacing self-attention with simpler token mixers, MetaFormer provides strong baselines for vision tasks. However, while extensively studied on natural image datasets, its use in medical imaging remains scarce, and existing works rarely compare different token mixers, potentially overlooking more suitable designs choices. In this work, we present the first comprehensive study of token mixers for medical imaging. We systematically analyze pooling-, convolution-, and attention-based token mixers within the MetaFormer architecture on image classification (global prediction task) and semantic segmentation (dense prediction task). Our evaluation spans eight datasets covering diverse modalities and common challenges in the medical domain. Given the prevalence of pretraining from natural images to mitigate medical data scarcity, we also examine transferring pretrained weights to new token mixers. Our results show that, for classification, low-complexity token mixers (e.g. grouped convolution or pooling) are sufficient, aligning with findings on natural images. Pretrained weights remain useful despite the domain gap introduced by the new token mixer. For segmentation, we find that the local inductive bias of convolutional token mixers is essential. Grouped convolutions emerge as the preferred choice, as they reduce runtime and parameter count compared to standard convolutions, while the MetaFormer’s channel-MLPs already provide the necessary cross-channel interactions. Our code is available on GitHub.

Method: The survey proposes a multi-perspective taxonomy with six categories: Intrinsic Decomposition, Spectral & Latent, Accelerated, Guided, Multimodal, and Autonomous. It uses a hybrid approach considering both model mechanisms and conditioning signals.

Result: The analysis includes comparative performance evaluation, examination of qualitative failure modes, benchmark inconsistencies, and trade-offs between interpretability, generalization, and inference efficiency.

Conclusion: The survey aims to guide future diffusion-based LLIE research by highlighting trends, open research questions, and discussing real-world deployment constraints and ethical considerations, with emphasis on novel conditioning, real-time adaptation, and foundation models.

Abstract: Low-light image enhancement (LLIE) is vital for safety-critical applications such as surveillance, autonomous navigation, and medical imaging, where visibility degradation can impair downstream task performance. Recently, diffusion models have emerged as a promising generative paradigm for LLIE due to their capacity to model complex image distributions via iterative denoising. This survey provides an up-to-date critical analysis of diffusion models for LLIE, distinctively featuring an in-depth comparative performance evaluation against Generative Adversarial Network and Transformer-based state-of-the-art methods, a thorough examination of practical deployment challenges, and a forward-looking perspective on the role of emerging paradigms like foundation models. We propose a multi-perspective taxonomy encompassing six categories: Intrinsic Decomposition, Spectral & Latent, Accelerated, Guided, Multimodal, and Autonomous; that map enhancement methods across physical priors, conditioning schemes, and computational efficiency. Our taxonomy is grounded in a hybrid view of both the model mechanism and the conditioning signals. We evaluate qualitative failure modes, benchmark inconsistencies, and trade-offs between interpretability, generalization, and inference efficiency. We also discuss real-world deployment constraints (e.g., memory, energy use) and ethical considerations. This survey aims to guide the next generation of diffusion-based LLIE research by highlighting trends and surfacing open research questions, including novel conditioning, real-time adaptation, and the potential of foundation models.

Method: Proposed diffusion-driven image regeneration process and a novel guided diffusion attack that explicitly targets watermark signals during generation. Theoretically proved that sufficient diffusion-based transformation eliminates mutual information between watermarked images and embedded payloads.

Result: Achieved near-zero watermark recovery rates on state-of-the-art watermarking schemes (StegaStamp, TrustMark, VINE) while maintaining high visual fidelity in regenerated images.

Conclusion: Current robust watermarking techniques have fundamental vulnerabilities against generative model-based attacks, highlighting the need for new watermarking strategies in the generative AI era.

Abstract: Robust invisible watermarking aims to embed hidden information into images such that the watermark can survive various image manipulations. However, the rise of powerful diffusion-based image generation and editing techniques poses a new threat to these watermarking schemes. In this paper, we present a theoretical study and method demonstrating that diffusion models can effectively break robust image watermarks that were designed to resist conventional perturbations. We show that a diffusion-driven ``image regeneration’’ process can erase embedded watermarks while preserving perceptual image content. We further introduce a novel guided diffusion attack that explicitly targets the watermark signal during generation, significantly degrading watermark detectability. Theoretically, we prove that as an image undergoes sufficient diffusion-based transformation, the mutual information between the watermarked image and the embedded watermark payload vanishes, resulting in decoding failure. Experimentally, we evaluate our approach on multiple state-of-the-art watermarking schemes (including the deep learning-based methods StegaStamp, TrustMark, and VINE) and demonstrate near-zero watermark recovery rates after attack, while maintaining high visual fidelity of the regenerated images. Our findings highlight a fundamental vulnerability in current robust watermarking techniques against generative model-based attacks, underscoring the need for new watermarking strategies in the era of generative AI.

Method: Used 474 crowdsourced hail images from Austria (2022-2024) with hail diameters 2-11cm, compared four models using one-stage and two-stage prompting strategies with reference objects like human hands.

Result: Best model achieved average mean absolute error of 1.12cm, with two-stage prompting improving reliability compared to single-stage approach.

Conclusion: Off-the-shelf models without fine-tuning can effectively measure hailstones from images, enabling faster severe weather assessments, though automated real-time image harvesting remains an open challenge.

Abstract: This study examines the use of social media and news images to detect and measure hailstones, utilizing pre-trained multimodal large language models. The dataset for this study comprises 474 crowdsourced images of hailstones from documented hail events in Austria, which occurred between January 2022 and September 2024. These hailstones have maximum diameters ranging from 2 to 11cm. We estimate the hail diameters and compare four different models utilizing one-stage and two-stage prompting strategies. The latter utilizes additional size cues from reference objects, such as human hands, within the image. Our results show that pretrained models already have the potential to measure hailstone diameters from images with an average mean absolute error of 1.12cm for the best model. In comparison to a single-stage prompt, two-stage prompting improves the reliability of most models. Our study suggests that these off-the-shelf models, even without fine-tuning, can complement traditional hail sensors by extracting meaningful and spatially dense information from social media imagery, enabling faster and more detailed assessments of severe weather events. The automated real-time image harvesting from social media and other sources remains an open task, but it will make our approach directly applicable to future hail events.

Method: Built on pretrained ViT-GPT-2 backbone, uses multi-loss framework with: (1) prompt-based cosine similarity loss for image-prompt alignment, (2) CLIP-style loss for image-caption alignment, (3) language-guided contrastive loss for task discriminability.

Result: Mitigates catastrophic forgetting and achieves better semantic caption alignment compared to state-of-the-art methods, with no additional inference overhead or prompts during generation.

Conclusion: The proposed multi-loss framework effectively addresses continual image captioning challenges by integrating semantic guidance through prompt-based learning and contrastive alignment.

Abstract: Generating accurate and coherent image captions in a continual learning setting remains a major challenge due to catastrophic forgetting and the difficulty of aligning evolving visual concepts with language over time. In this work, we propose a novel multi-loss framework for continual image captioning that integrates semantic guidance through prompt-based continual learning and contrastive alignment. Built upon a pretrained ViT-GPT-2 backbone, our approach combines standard cross-entropy loss with three additional components: (1) a prompt-based cosine similarity loss that aligns image embeddings with synthetically constructed prompts encoding objects, attributes, and actions; (2) a CLIP-style loss that promotes alignment between image embeddings and target caption embedding; and (3) a language-guided contrastive loss that employs a triplet loss to enhance class-level discriminability between tasks. Notably, our approach introduces no additional overhead at inference time and requires no prompts during caption generation. We find that this approach mitigates catastrophic forgetting, while achieving better semantic caption alignment compared to state-of-the-art methods. The code can be found via the following link https://github.com/ Gepardius/Taetz_Bordelius_Continual_ImageCaptioning.

Method: Evaluated two technical safeguards: index exclusion and confusion loss, testing their effectiveness in reducing person re-identification while maintaining object retrieval performance.

Result: Combining index exclusion and confusion loss reduced person re-identification accuracy to below 2% while maintaining 82% of retrieval performance for non-person objects. However, vulnerabilities were identified, including potential circumvention using partial person images.

Conclusion: The findings highlight urgent regulatory questions about classifying and regulating systems with emergent identification capabilities, and what technical standards should be required to prevent identification capabilities from developing in seemingly benign AI applications.

Abstract: Generic instance search models can dramatically reduce the manual effort required to analyze vast surveillance footage during criminal investigations by retrieving specific objects of interest to law enforcement. However, our research reveals an unintended emergent capability: through overlearning, these models can single out specific individuals even when trained on datasets without human subjects. This capability raises concerns regarding identification and profiling of individuals based on their personal data, while there is currently no clear standard on how de-identification can be achieved. We evaluate two technical safeguards to curtail a model’s person re-identification capacity: index exclusion and confusion loss. Our experiments demonstrate that combining these approaches can reduce person re-identification accuracy to below 2% while maintaining 82% of retrieval performance for non-person objects. However, we identify critical vulnerabilities in these mitigations, including potential circumvention using partial person images. These findings highlight urgent regulatory questions at the intersection of AI governance and data protection: How should we classify and regulate systems with emergent identification capabilities? And what technical standards should be required to prevent identification capabilities from developing in seemingly benign applications?

Method: Proposes a generic graph-based NAS search space and a new search algorithm to traverse this unified space, along with a standardized training/evaluation toolkit.

Result: The unified space contains architectures that outperform state-of-the-art hand-crafted models when using identical training setups.

Conclusion: UniNAS opens pathways for systematic exploration of the full neural architecture spectrum through a unified graph-based NAS perspective, promoting reproducibility and fair comparisons.

Abstract: We introduce Universal Neural Architecture Space (UniNAS), a generic search space for neural architecture search (NAS) which unifies convolutional networks, transformers, and their hybrid architectures under a single, flexible framework. Our approach enables discovery of novel architectures as well as analyzing existing architectures in a common framework. We also propose a new search algorithm that allows traversing the proposed search space, and demonstrate that the space contains interesting architectures, which, when using identical training setup, outperform state-of-the-art hand-crafted architectures. Finally, a unified toolkit including a standardized training and evaluation protocol is introduced to foster reproducibility and enable fair comparison in NAS research. Overall, this work opens a pathway towards systematically exploring the full spectrum of neural architectures with a unified graph-based NAS perspective.

Method: Propose VideoMiner with iterative segmentation, captioning, and clustering to form hierarchical tree structures, and introduce T-GRPO (tree-based group relative policy optimization) reinforcement learning for precise key frame localization.

Result: Achieves superior performance in all long-video understanding tasks, with T-GRPO spontaneously generating reasoning chains and tree growth auxin dynamically adjusting expansion depth for accuracy and efficiency gains.

Conclusion: VideoMiner effectively addresses challenges in long-video understanding by combining hierarchical tree structures with reinforcement learning, enabling dynamic adaptation and precise key frame identification while maintaining temporal coherence.

Abstract: Understanding hour-long videos with multi-modal large language models (MM-LLMs) enriches the landscape of human-centered AI applications. However, for end-to-end video understanding with LLMs, uniformly sampling video frames results in LLMs being overwhelmed by a vast amount of irrelevant information as video length increases. Existing hierarchical key frame extraction methods improve the accuracy of video understanding but still face two critical challenges. 1) How can the interference of extensive redundant information in long videos be mitigated? 2) How can a model dynamically adapt to complex hierarchical structures while accurately identifying key frames? To address these issues, we propose VideoMiner, which iteratively segments, captions, and clusters long videos, forming a hierarchical tree structure. The proposed VideoMiner progresses from long videos to events to frames while preserving temporal coherence, effectively addressing the first challenge. To precisely locate key frames, we introduce T-GRPO, a tree-based group relative policy optimization in reinforcement learning method that guides the exploration of the VideoMiner. The proposed T-GRPO is specifically designed for tree structures, integrating spatiotemporal information at the event level while being guided by the question, thus solving the second challenge. We achieve superior performance in all long-video understanding tasks and uncover several interesting insights. Our proposed T-GRPO surprisingly incentivizes the model to spontaneously generate a reasoning chain. Additionally, the designed tree growth auxin dynamically adjusts the expansion depth, obtaining accuracy and efficiency gains. The code is publicly available at https://github.com/caoxinye/VideoMiner.

Method: Extends Learned Vertex Descent (LVD) by integrating per-vertex neural field optimization with global structural guidance from dynamically predicted 3D keypoints, using relative spatial encoding to iteratively refine mesh vertices without dense 3D supervision.

Result: Achieves state-of-the-art performance in single-view settings and remains highly competitive in multi-view scenarios, while substantially reducing inference time.

Conclusion: GLVD enables expressive and adaptable geometry reconstruction while maintaining computational efficiency, offering a balanced solution between representation capacity and computational cost.

Abstract: Existing 3D face modeling methods usually depend on 3D Morphable Models, which inherently constrain the representation capacity to fixed shape priors. Optimization-based approaches offer high-quality reconstructions but tend to be computationally expensive. In this work, we introduce GLVD, a hybrid method for 3D face reconstruction from few-shot images that extends Learned Vertex Descent (LVD) by integrating per-vertex neural field optimization with global structural guidance from dynamically predicted 3D keypoints. By incorporating relative spatial encoding, GLVD iteratively refines mesh vertices without requiring dense 3D supervision. This enables expressive and adaptable geometry reconstruction while maintaining computational efficiency. GLVD achieves state-of-the-art performance in single-view settings and remains highly competitive in multi-view scenarios, all while substantially reducing inference time.

Method: Combine MedFlamingo (medical domain-specific Vision-Language Model) with PPO, processing task observations and instructions once per episode to generate high-level planning tokens that combine medical expertise with real-time visual feedback.

Result: MedFlamingo PPO outperforms and converges faster than standard vision-based PPO and OpenFlamingo PPO baselines, achieving task success rates exceeding 70% across all environments with improvements ranging from 66.67% to 1114.29% compared to baseline.

Conclusion: Specialized medical knowledge in robotic surgical planning and decision-making provides significant value, as demonstrated by the superior performance of MedFlamingo PPO in laparoscopic surgery tasks.

Abstract: Vision-based Proximal Policy Optimization (PPO) struggles with visual observation-based robotic laparoscopic surgical tasks due to the high-dimensional nature of visual input, the sparsity of rewards in surgical environments, and the difficulty of extracting task-relevant features from raw visual data. We introduce a simple approach integrating MedFlamingo, a medical domain-specific Vision-Language Model, with PPO. Our method is evaluated on five diverse laparoscopic surgery task environments in LapGym, using only endoscopic visual observations. MedFlamingo PPO outperforms and converges faster compared to both standard vision-based PPO and OpenFlamingo PPO baselines, achieving task success rates exceeding 70% across all environments, with improvements ranging from 66.67% to 1114.29% compared to baseline. By processing task observations and instructions once per episode to generate high-level planning tokens, our method efficiently combines medical expertise with real-time visual feedback. Our results highlight the value of specialized medical knowledge in robotic surgical planning and decision-making.

Method: Introduces CAPTCHA-X benchmark with 7 CAPTCHA categories, step-by-step action solutions, and grounding annotations. Proposes a general agentic VLM-based framework that leverages models’ inherent reasoning abilities with five reasoning-oriented metrics.

Result: The proposed method achieves state-of-the-art performance with 83.9% average solving accuracy across five high-difficulty CAPTCHA types, substantially surpassing existing baselines (which achieve only 21.9% accuracy).

Conclusion: Step-by-step reasoning is crucial for solving CAPTCHAs and spatial reasoning tasks. Current models have significant limitations in reasoning capabilities, and incorporating reasoning is essential for advancing visual-spatial challenges.

Abstract: CAPTCHA, originally designed to distinguish humans from robots, has evolved into a real-world benchmark for assessing the spatial reasoning capabilities of vision-language models. In this work, we first show that step-by-step reasoning is crucial for vision-language models (VLMs) to solve CAPTCHAs, which represent high-difficulty spatial reasoning tasks, and that current commercial vision-language models still struggle with such reasoning. In particular, we observe that most commercial VLMs (e.g., Gemini, Claude, GPT, etc.) fail to effectively solve CAPTCHAs and thus achieve low accuracy (around 21.9 percent). However, our findings indicate that requiring the model to perform step-by-step reasoning before generating the final coordinates can significantly enhance its solving accuracy, underscoring the severity of the gap. To systematically study this issue, we introduce CAPTCHA-X, the first real-world CAPTCHA benchmark with reasoning, covering seven categories of CAPTCHAs (such as Gobang, hCaptcha, etc.) with step-by-step action solutions and grounding annotations. We further define five reasoning-oriented metrics that enable a comprehensive evaluation of models reasoning capabilities. To validate the effectiveness of reasoning, we also propose a general agentic VLM-based framework that incorporates the models inherent reasoning abilities. Our method achieves state-of-the-art performance across five high-difficulty CAPTCHA types, with an average solving accuracy of 83.9 percent, substantially surpassing existing baselines. These results reveal the limitations of current models and highlight the importance of reasoning in advancing visual-spatial challenges in the future.

Method: Combines attention maps with statistical filtering (initially proposed for CNNs) to remove noisy patterns, with a class-specific variant for discriminative explanations.

Result: Produces sharper and more interpretable maps that outperform or are comparable to SOTA methods across multiple datasets, with better alignment to human perception.

Conclusion: Attention is a valuable signal for ViT explanations when properly filtered, and human interpretability remains essential for XAI evaluation.

Abstract: Explainable AI (XAI) has become increasingly important with the rise of large transformer models, yet many explanation methods designed for CNNs transfer poorly to Vision Transformers (ViTs). Existing ViT explanations often rely on attention weights, which tend to yield noisy maps as they capture token-to-token interactions within each layer.While attribution methods incorporating MLP blocks have been proposed, we argue that attention remains a valuable and interpretable signal when properly filtered. We propose a method that combines attention maps with a statistical filtering, initially proposed for CNNs, to remove noisy or uninformative patterns and produce more faithful explanations. We further extend our approach with a class-specific variant that yields discriminative explanations. Evaluation against popular state-of-the-art methods demonstrates that our approach produces sharper and more interpretable maps. In addition to perturbation-based faithfulness metrics, we incorporate human gaze data to assess alignment with human perception, arguing that human interpretability remains essential for XAI. Across multiple datasets, our approach consistently outperforms or is comparable to the SOTA methods while remaining efficient and human plausible.

Method: Proposes Visual Evidence Reward (VER), a reinforcement learning framework that explicitly rewards reasoning traces that are verifiably grounded in visual evidence. This counters the “visual thinking drift” phenomenon.

Result: Comprehensive evaluation across 10 diverse video understanding benchmarks shows that Video-VER consistently achieves top performance, outperforming standard CoT approaches.

Conclusion: The work highlights distinct challenges in video-centric reasoning and advocates for AI that robustly grounds inferences in visual evidence - enabling models to “see while thinking” rather than just “think before answering”.

Abstract: Video reasoning, the task of enabling machines to infer from dynamic visual content through multi-step logic, is crucial for advanced AI. While the Chain-of-Thought (CoT) mechanism has enhanced reasoning in text-based tasks, its application to video understanding remains underexplored. This paper presents a systematic analysis revealing that CoT often degrades performance in video reasoning, generating verbose but misleading internal monologues, and leading to hallucinated visual details and overridden correct intuitions - a phenomenon we term “visual thinking drift”. We explain this drift through a Bayesian lens, positing that CoT traces often diverge from actual visual evidence, instead amplifying internal biases or language priors, causing models to storytell rather than engage in grounded reasoning. To counteract this, we introduce Visual Evidence Reward (VER), a novel reinforcement learning framework that explicitly rewards the generation of reasoning traces that are verifiably grounded in visual evidence. Comprehensive evaluation across 10 diverse video understanding benchmarks demonstrates that our Video-VER consistently achieves top performance. Our work sheds light on the distinct challenges of video-centric reasoning and encourages the development of AI that robustly grounds its inferences in visual evidence - for large multimodal models that not only “think before answering”, but also “see while thinking”.

Method: Used a state-of-the-art segmentation framework trained on private data with manual annotations, transferred to ImageCAS dataset. Improved CA labels from original annotations and refined PV segmentations from TS outputs. Also identified scans with common data flaws.

Result: Created an open-source dataset with curated segmentations for LAA, CAs, and PVs from 1000 CCTA scans, including identification of scans with data defects like step artifacts and field-of-view limitations.

Conclusion: This dataset provides valuable resources for improving segmentation of challenging cardiac structures and enables novel approaches for LAA morphology analysis in medical imaging research.

Abstract: Despite the success of advanced segmentation frameworks such as TotalSegmentator (TS), accurate segmentations of the left atrial appendage (LAA), coronary arteries (CAs), and pulmonary veins (PVs) remain a significant challenge in medical imaging. In this work, we present the first open-source, anatomically coherent dataset of curated, high-resolution segmentations for these structures, supplemented with whole-heart labels produced by TS on the publicly available ImageCAS dataset consisting of 1000 cardiac computed tomography angiography (CCTA) scans. One purpose of the data set is to foster novel approaches to the analysis of LAA morphology. LAA segmentations on ImageCAS were generated using a state-of-the-art segmentation framework developed specifically for high resolution LAA segmentation. We trained the network on a large private dataset with manual annotations provided by medical readers guided by a trained cardiologist and transferred the model to ImageCAS data. CA labels were improved from the original ImageCAS annotations, while PV segmentations were refined from TS outputs. In addition, we provide a list of scans from ImageCAS that contains common data flaws such as step artefacts, LAAs extending beyond the scanner’s field of view, and other types of data defects.

Method: Proposes a model that decouples spectral low-rankness and spatial priors via block term decomposition, stacks spatial maps as a tensor encoding high-order spatial low-rankness and smoothness priors using non-convex mode-shuffled tensor correlated total variation, and uses an efficient algorithm based on linearized alternating direction method of multipliers.

Result: Experiments on multiple datasets demonstrate the effectiveness of the proposed algorithm in achieving hyperspectral image super-resolution.

Conclusion: The proposed method successfully addresses the limitations of existing tensor-based approaches by compactly characterizing multi-level priors within the tensor framework, providing an efficient optimization solution with theoretical convergence guarantees.

Abstract: Fusing a hyperspectral image with a multispectral image acquired over the same scene, \textit{i.e.}, hyperspectral image super-resolution, has become a popular computational way to access the latent high-spatial-spectral-resolution image. To date, a variety of fusion methods have been proposed, among which the tensor-based ones have testified that multiple priors, such as multidimensional low-rankness and spatial total variation at multiple levels, effectively drive the fusion process. However, existing tensor-based models can only effectively leverage one or two priors at one or two levels, since simultaneously incorporating multi-level priors inevitably increases model complexity. This introduces challenges in both balancing the weights of different priors and optimizing multi-block structures. Concerning this, we present a novel hyperspectral super-resolution model compactly characterizing these multi-level priors of hyperspectral images within the tensor framework. Firstly, the proposed model decouples the spectral low-rankness and spatial priors by casting the latent high-spatial-spectral-resolution image into spectral subspace and spatial maps via block term decomposition. Secondly, these spatial maps are stacked as the spatial tensor encoding the high-order spatial low-rankness and smoothness priors, which are co-modeled via the proposed non-convex mode-shuffled tensor correlated total variation. Finally, we draw inspiration from the linearized alternating direction method of multipliers to design an efficient algorithm to optimize the resulting model, theoretically proving its Karush-Kuhn-Tucker convergence under mild conditions. Experiments on multiple datasets demonstrate the effectiveness of the proposed algorithm. The code implementation will be available from https://github.com/WongYinJ.

Method: Creates unified feature prototype space integrating global/local WSI features with genomic profiles. Key innovations: robust phenotype representation combining patches with global context, Exponential Prototype Update Strategy (EMA ProtoUp) for stable cross-modal associations, and hierarchical prototype matching capturing global centrality, local typicality, and cohort-level trends.

Result: Comprehensive evaluations on four cancer datasets show FeatProto surpasses current leading unimodal and multimodal survival prediction techniques in both accuracy and interpretability.

Conclusion: FeatProto provides a new perspective on prototype learning for critical medical applications, offering traceable and interpretable decision-making processes for cancer survival prediction.

Abstract: Survival analysis plays a vital role in making clinical decisions. However, the models currently in use are often difficult to interpret, which reduces their usefulness in clinical settings. Prototype learning presents a potential solution, yet traditional methods focus on local similarities and static matching, neglecting the broader tumor context and lacking strong semantic alignment with genomic data. To overcome these issues, we introduce an innovative prototype-based multimodal framework, FeatProto, aimed at enhancing cancer survival prediction by addressing significant limitations in current prototype learning methodologies within pathology. Our framework establishes a unified feature prototype space that integrates both global and local features of whole slide images (WSI) with genomic profiles. This integration facilitates traceable and interpretable decision-making processes. Our approach includes three main innovations: (1) A robust phenotype representation that merges critical patches with global context, harmonized with genomic data to minimize local bias. (2) An Exponential Prototype Update Strategy (EMA ProtoUp) that sustains stable cross-modal associations and employs a wandering mechanism to adapt prototypes flexibly to tumor heterogeneity. (3) A hierarchical prototype matching scheme designed to capture global centrality, local typicality, and cohort-level trends, thereby refining prototype inference. Comprehensive evaluations on four publicly available cancer datasets indicate that our method surpasses current leading unimodal and multimodal survival prediction techniques in both accuracy and interoperability, providing a new perspective on prototype learning for critical medical applications. Our source code is available at https://github.com/JSLiam94/FeatProto.

Method: SSGNet combines class-specific generative modeling using StyleGAN3 to expand training data with iterative semi-supervised pseudo labeling to refine labels, augmenting existing baseline models rather than replacing them.

Result: Experiments across multiple medical imaging benchmarks show consistent gains in classification and segmentation performance, with Frechet Inception Distance analysis confirming high-quality generated samples.

Conclusion: SSGNet provides a practical strategy to mitigate annotation bottlenecks and improve robustness in medical image analysis through data augmentation and label refinement.

Abstract: Deep learning in medical imaging is often limited by scarce and imbalanced annotated data. We present SSGNet, a unified framework that combines class specific generative modeling with iterative semisupervised pseudo labeling to enhance both classification and segmentation. Rather than functioning as a standalone model, SSGNet augments existing baselines by expanding training data with StyleGAN3 generated images and refining labels through iterative pseudo labeling. Experiments across multiple medical imaging benchmarks demonstrate consistent gains in classification and segmentation performance, while Frechet Inception Distance analysis confirms the high quality of generated samples. These results highlight SSGNet as a practical strategy to mitigate annotation bottlenecks and improve robustness in medical image analysis.

Method: Built on a discrete diffusion framework using a multimodal large language model as backbone, with key designs: removing causal attention mask for bidirectional context and injecting continuous timestep embeddings for diffusion awareness.

Result: Achieves high-fidelity medical generation (FID 16.60 on MIMIC-CXR, FID 24.19 on PathGen) and accurate report generation (METEOR 0.2650 and 0.2580), with joint image-report pairs enhancing downstream performance significantly.

Conclusion: MeDiM supports coherent and clinically grounded multimodal outputs, demonstrating effective unified medical generation across different modalities.

Abstract: Recent advances in generative medical models are constrained by modality-specific scenarios that hinder the integration of complementary evidence from imaging, pathology, and clinical notes. This fragmentation limits their evolution into foundation models that can learn and reason across the full spectrum of biomedical data. We propose MeDiM, the first medical discrete diffusion model that learns shared distributions across modalities without modality-specific components. MeDiM unifies multiple generative tasks: translating between images and text, and jointly producing image-report pairs across domains in response to prompts. Built on a discrete diffusion framework, MeDiM bridges vision and language representations through a shared probabilistic space. To enable unified and flexible medical generation, we employ a multimodal large language model (MLLM) as the diffusion backbone, leveraging its prior knowledge and cross-modal reasoning. Two key designs are introduced: (1) removing the causal attention mask for bidirectional context, and (2) injecting continuous timestep embeddings for diffusion awareness. Experiments demonstrate high-fidelity medical generation (FID 16.60 on MIMIC-CXR and FID 24.19 on PathGen) and accurate report generation (METEOR 0.2650 and 0.2580). Jointly generated image-report pairs further enhance downstream performance (plus6.43 percent BLEU-1, plus18.57 percent BLEU-2, plus31.58 percent BLEU-3, plus4.80 percent METEOR), showing that MeDiM supports coherent and clinically grounded multimodal outputs.

Method: Proposes FlowRVS framework that treats RVOS as conditional continuous flow problem, leveraging pretrained T2V models for fine-grained pixel control and temporal coherence. Learns direct language-guided deformation from video’s holistic representation to target mask instead of conventional mask generation.

Result: Achieves new state-of-the-art results: 51.1 J&F in MeViS (+1.6 over prior SOTA) and 73.3 in zero-shot Ref-DAVIS17 (+2.7).

Conclusion: Demonstrates significant potential of modeling video understanding tasks as continuous deformation processes, with one-stage generative approach outperforming previous methods.

Abstract: Referring Video Object Segmentation (RVOS) requires segmenting specific objects in a video guided by a natural language description. The core challenge of RVOS is to anchor abstract linguistic concepts onto a specific set of pixels and continuously segment them through the complex dynamics of a video. Faced with this difficulty, prior work has often decomposed the task into a pragmatic `locate-then-segment’ pipeline. However, this cascaded design creates an information bottleneck by simplifying semantics into coarse geometric prompts (e.g, point), and struggles to maintain temporal consistency as the segmenting process is often decoupled from the initial language grounding. To overcome these fundamental limitations, we propose FlowRVS, a novel framework that reconceptualizes RVOS as a conditional continuous flow problem. This allows us to harness the inherent strengths of pretrained T2V models, fine-grained pixel control, text-video semantic alignment, and temporal coherence. Instead of conventional generating from noise to mask or directly predicting mask, we reformulate the task by learning a direct, language-guided deformation from a video’s holistic representation to its target mask. Our one-stage, generative approach achieves new state-of-the-art results across all major RVOS benchmarks. Specifically, achieving a $\mathcal{J}&\mathcal{F}$ of 51.1 in MeViS (+1.6 over prior SOTA) and 73.3 in the zero shot Ref-DAVIS17 (+2.7), demonstrating the significant potential of modeling video understanding tasks as continuous deformation processes.

Method: The authors designed an annotation pipeline using a diffusion model to lift 2D hand keypoint sequences to 4D hand motion, and adopted a diffusion loss for the forecasting model to handle multimodality in hand motion distribution.

Result: Extensive experiments across 6 datasets showed significant improvements: 14% from training on diverse data with imputed labels, 42% better performance from the lifting model, and 16.4% gain from the forecasting model over best baselines, with strong zero-shot generalization to everyday images.

Conclusion: The proposed diffusion-based approach effectively addresses the 3D annotation scarcity problem and achieves state-of-the-art performance in forecasting bimanual hand motion, particularly demonstrating strong generalization capabilities.

Abstract: We tackle the problem of forecasting bimanual 3D hand motion & articulation from a single image in everyday settings. To address the lack of 3D hand annotations in diverse settings, we design an annotation pipeline consisting of a diffusion model to lift 2D hand keypoint sequences to 4D hand motion. For the forecasting model, we adopt a diffusion loss to account for the multimodality in hand motion distribution. Extensive experiments across 6 datasets show the benefits of training on diverse data with imputed labels (14% improvement) and effectiveness of our lifting (42% better) & forecasting (16.4% gain) models, over the best baselines, especially in zero-shot generalization to everyday images.

Method: Three key components: temporal attention for conditioning on all frames, time-aware point sampling and 4D latent anchoring for temporal consistency, and noise sharing across frames for stability.

Result: Method accurately captures complex dynamics without per-frame optimization, improves robustness and perceptual fidelity across diverse in-the-wild videos, and reduces failure modes compared to baselines.

Conclusion: The framework successfully generates high-quality 4D shapes from videos end-to-end, demonstrating superior performance in capturing temporal dynamics and maintaining consistency.

Abstract: Video-conditioned 4D shape generation aims to recover time-varying 3D geometry and view-consistent appearance directly from an input video. In this work, we introduce a native video-to-4D shape generation framework that synthesizes a single dynamic 3D representation end-to-end from the video. Our framework introduces three key components based on large-scale pre-trained 3D models: (i) a temporal attention that conditions generation on all frames while producing a time-indexed dynamic representation; (ii) a time-aware point sampling and 4D latent anchoring that promote temporally consistent geometry and texture; and (iii) noise sharing across frames to enhance temporal stability. Our method accurately captures non-rigid motion, volume changes, and even topological transitions without per-frame optimization. Across diverse in-the-wild videos, our method improves robustness and perceptual fidelity and reduces failure modes compared with the baselines.

Method: Proposed statistical measures using E2E drivers to evaluate video realism, exploited video generation controllability for targeted experiments on distribution gaps, and used synthetic data from video generation models to improve planner generalization.

Result: Developed novel evaluation metrics for synthetic video realism, identified distribution gaps affecting planner performance, and demonstrated that synthetic data effectively improves E2E model generalization beyond existing operational domains.

Conclusion: Video generation models provide a cost-effective alternative to real-world data collection and can effectively improve autonomous vehicle planner generalization, enabling expansion into new operational contexts.

Abstract: Recent advances in generative models have sparked exciting new possibilities in the field of autonomous vehicles. Specifically, video generation models are now being explored as controllable virtual testing environments. Simultaneously, end-to-end (E2E) driving models have emerged as a streamlined alternative to conventional modular autonomous driving systems, gaining popularity for their simplicity and scalability. However, the application of these techniques to simulation and planning raises important questions. First, while video generation models can generate increasingly realistic videos, can these videos faithfully adhere to the specified conditions and be realistic enough for E2E autonomous planner evaluation? Second, given that data is crucial for understanding and controlling E2E planners, how can we gain deeper insights into their biases and improve their ability to generalize to out-of-distribution scenarios? In this work, we bridge the gap between the driving models and generative world models (Drive&Gen) to address these questions. We propose novel statistical measures leveraging E2E drivers to evaluate the realism of generated videos. By exploiting the controllability of the video generation model, we conduct targeted experiments to investigate distribution gaps affecting E2E planner performance. Finally, we show that synthetic data produced by the video generation model offers a cost-effective alternative to real-world data collection. This synthetic data effectively improves E2E model generalization beyond existing Operational Design Domains, facilitating the expansion of autonomous vehicle services into new operational contexts.

Method: The method generates an all-in-focus image, estimates monocular depth, predicts focus distance with a novel transformer, and applies differentiable lens blur model. Gradients flow backwards through the entire process for unsupervised learning.

Result: The model enables superior fine-grained control over defocus effects while preserving scene contents, which existing diffusion models cannot achieve.

Conclusion: The framework successfully integrates camera metadata to generate controllable lens blur effects, providing precise interactive user control without altering scene content.

Abstract: Current text-to-image diffusion models excel at generating diverse, high-quality images, yet they struggle to incorporate fine-grained camera metadata such as precise aperture settings. In this work, we introduce a novel text-to-image diffusion framework that leverages camera metadata, or EXIF data, which is often embedded in image files, with an emphasis on generating controllable lens blur. Our method mimics the physical image formation process by first generating an all-in-focus image, estimating its monocular depth, predicting a plausible focus distance with a novel focus distance transformer, and then forming a defocused image with an existing differentiable lens blur model. Gradients flow backwards through this whole process, allowing us to learn without explicit supervision to generate defocus effects based on content elements and the provided EXIF data. At inference time, this enables precise interactive user control over defocus effects while preserving scene contents, which is not achievable with existing diffusion models. Experimental results demonstrate that our model enables superior fine-grained control without altering the depicted scene.

Method: Uses three pretrained modules: monocular depth estimator, learned keypoint detector, and instance segmentation network. Suppresses dynamic objects with dilated masks, assigns predicted depth to static keypoints, and processes them with unmodified RGB-D SLAM backend.

Result: Achieves 7.4 cm mean ATE on static sequences and 1.8 cm on dynamic sequences in TUM RGB-D benchmark, matching or surpassing state-of-the-art RGB-D methods while operating at 22 FPS on single GPU.

Conclusion: Modern pretrained vision models can effectively replace active depth sensors as reliable, real-time sources of metric scale, enabling simpler and more cost-effective SLAM systems.

Abstract: We present DropD-SLAM, a real-time monocular SLAM system that achieves RGB-D-level accuracy without relying on depth sensors. The system replaces active depth input with three pretrained vision modules: a monocular metric depth estimator, a learned keypoint detector, and an instance segmentation network. Dynamic objects are suppressed using dilated instance masks, while static keypoints are assigned predicted depth values and backprojected into 3D to form metrically scaled features. These are processed by an unmodified RGB-D SLAM back end for tracking and mapping. On the TUM RGB-D benchmark, DropD-SLAM attains 7.4 cm mean ATE on static sequences and 1.8 cm on dynamic sequences, matching or surpassing state-of-the-art RGB-D methods while operating at 22 FPS on a single GPU. These results suggest that modern pretrained vision models can replace active depth sensors as reliable, real-time sources of metric scale, marking a step toward simpler and more cost-effective SLAM systems.

Method: Collected both synthetic videos rendered by Blender and real-world recordings with visually and temporally aligned scenes and actions. Constructed EgoNight-VQA using a day-augmented night auto-labeling engine with extensive human verification, resulting in 3658 QA pairs across 90 videos spanning 12 QA types.

Result: Evaluations of state-of-the-art multimodal large language models (MLLMs) reveal substantial performance drops when transferring from day to night, highlighting the challenges of reasoning under low-light conditions.

Conclusion: EgoNight-VQA provides a strong foundation for advancing application-driven egocentric vision research and developing models that generalize across illumination domains, with auxiliary tasks further exploring model boundaries.

Abstract: Most existing benchmarks for egocentric vision understanding focus primarily on daytime scenarios, overlooking the low-light conditions that are inevitable in real-world applications. To investigate this gap, we present EgoNight, the first comprehensive benchmark for nighttime egocentric vision, with visual question answering (VQA) as the core task. A key feature of EgoNight is the introduction of day-night aligned videos, which enhance night annotation quality using the daytime data and reveal clear performance gaps between lighting conditions. To achieve this, we collect both synthetic videos rendered by Blender and real-world recordings, ensuring that scenes and actions are visually and temporally aligned. Leveraging these paired videos, we construct EgoNight-VQA, supported by a novel day-augmented night auto-labeling engine and refinement through extensive human verification. Each QA pair is double-checked by annotators for reliability. In total, EgoNight-VQA contains 3658 QA pairs across 90 videos, spanning 12 diverse QA types, with more than 300 hours of human work. Evaluations of state-of-the-art multimodal large language models (MLLMs) reveal substantial performance drops when transferring from day to night, underscoring the challenges of reasoning under low-light conditions. Beyond VQA, EgoNight also introduces two auxiliary tasks, day-night correspondence retrieval and egocentric depth estimation at night, that further explore the boundaries of existing models. We believe EgoNight-VQA provides a strong foundation for advancing application-driven egocentric vision research and for developing models that generalize across illumination domains. All the data and code will be made available upon acceptance.

Method: Builds upon CUT3R 4D online reconstruction model using parameter-efficient visual prompt tuning to preserve spatiotemporal priors while enabling direct readout of multiple SMPL-X bodies in a unified feed-forward framework.

Result: Achieves real-time performance (15 FPS) with low memory footprint (8 GB) after training on BEDLAM dataset for one day on one GPU. Delivers state-of-the-art performance in global human motion estimation, local human mesh recovery, video depth estimation, and camera pose estimation.

Conclusion: Human3R serves as a simple yet strong baseline for unified human-scene reconstruction that can be easily extended for downstream applications, eliminating heavy dependencies and iterative refinement.

Abstract: We present Human3R, a unified, feed-forward framework for online 4D human-scene reconstruction, in the world frame, from casually captured monocular videos. Unlike previous approaches that rely on multi-stage pipelines, iterative contact-aware refinement between humans and scenes, and heavy dependencies, e.g., human detection, depth estimation, and SLAM pre-processing, Human3R jointly recovers global multi-person SMPL-X bodies (“everyone”), dense 3D scene (“everywhere”), and camera trajectories in a single forward pass (“all-at-once”). Our method builds upon the 4D online reconstruction model CUT3R, and uses parameter-efficient visual prompt tuning, to strive to preserve CUT3R’s rich spatiotemporal priors, while enabling direct readout of multiple SMPL-X bodies. Human3R is a unified model that eliminates heavy dependencies and iterative refinement. After being trained on the relatively small-scale synthetic dataset BEDLAM for just one day on one GPU, it achieves superior performance with remarkable efficiency: it reconstructs multiple humans in a one-shot manner, along with 3D scenes, in one stage, at real-time speed (15 FPS) with a low memory footprint (8 GB). Extensive experiments demonstrate that Human3R delivers state-of-the-art or competitive performance across tasks, including global human motion estimation, local human mesh recovery, video depth estimation, and camera pose estimation, with a single unified model. We hope that Human3R will serve as a simple yet strong baseline, be easily extended for downstream applications.Code available in https://fanegg.github.io/Human3R

Method: Used triangular 3D meshes acquired by photogrammetry with sub-millimeter accuracy, developed improved procedures for violin alignment reference plane, and computed contour lines and channel of minima for 38 violins, violas, and cellos.

Result: Identified observable differences in contour lines and channel of minima between reduced and unreduced instruments, with improved computational efficiency for these geometric characteristics.

Conclusion: The extended analysis of 38 instruments provides stronger geometric analysis and discussion of morphological differences in violin construction history.

Abstract: Some early violins have been reduced during their history to fit imposed morphological standards, while more recent ones have been built directly to these standards. We can observe differences between reduced and unreduced instruments, particularly in their contour lines and channel of minima. In a recent preliminary work, we computed and highlighted those two features for two instruments using triangular 3D meshes acquired by photogrammetry, whose fidelity has been assessed and validated with sub-millimetre accuracy. We propose here an extension to a corpus of 38 violins, violas and cellos, and introduce improved procedures, leading to a stronger discussion of the geometric analysis. We first recall the material we are working with. We then discuss how to derive the best reference plane for the violin alignment, which is crucial for the computation of contour lines and channel of minima. Finally, we show how to compute efficiently both characteristics and we illustrate our results with a few examples.

Method: Proposed BAFE-Net, a multi-task architecture that jointly tackles target detection and background semantic segmentation. It incorporates a dynamic cross-task feature hard-exchange mechanism to exchange target and background semantics between tasks.

Result: Comprehensive experiments show that BAFE-Net significantly enhances target detection accuracy while reducing false alarms. The method enables the shift from sparse to dense target detection.

Conclusion: The proposed approach with DenseSIRST dataset and BAFE-Net effectively addresses infrared small target detection challenges by leveraging background semantics through multi-task learning and cross-task feature exchange.

Abstract: Infrared small target detection presents significant challenges due to the limited intrinsic features of the target and the overwhelming presence of visually similar background distractors. We contend that background semantics are critical for distinguishing between objects that appear visually similar in this context. To address this challenge, we propose a task, clustered infrared small target detection, and introduce DenseSIRST, a benchmark dataset that provides per-pixel semantic annotations for background regions. This dataset facilitates the shift from sparse to dense target detection. This dataset facilitates the shift from sparse to dense target detection. Building on this resource, we propose the Background-Aware Feature Exchange Network (BAFE-Net), a multi-task architecture that jointly tackles target detection and background semantic segmentation. BAFE-Net incorporates a dynamic cross-task feature hard-exchange mechanism, enabling the effective exchange of target and background semantics between the two tasks. Comprehensive experiments demonstrate that BAFE-Net significantly enhances target detection accuracy while mitigating false alarms. The DenseSIRST dataset, along with the code and trained models, is publicly available at https://github.com/GrokCV/BAFE-Net.

Method: The paper reviews Mamba architectures including pure Mamba, U-Net variants, and hybrid models with CNNs, transformers, and Graph Neural Networks, covering optimizations, techniques, and adaptations.

Result: Mamba demonstrates strong performance in merging multimodal data, enabling faster inference with less memory, and improving diagnosis accuracy and patient outcomes in medical imaging.

Conclusion: Mamba has transformative potential to overcome existing barriers in medical imaging and pave the way for innovative advancements in the field.

Abstract: Mamba, a special case of the State Space Model, is gaining popularity as an alternative to template-based deep learning approaches in medical image analysis. While transformers are powerful architectures, they have drawbacks, including quadratic computational complexity and an inability to address long-range dependencies efficiently. This limitation affects the analysis of large and complex datasets in medical imaging, where there are many spatial and temporal relationships. In contrast, Mamba offers benefits that make it well-suited for medical image analysis. It has linear time complexity, which is a significant improvement over transformers. Mamba processes longer sequences without attention mechanisms, enabling faster inference and requiring less memory. Mamba also demonstrates strong performance in merging multimodal data, improving diagnosis accuracy and patient outcomes. The organization of this paper allows readers to appreciate the capabilities of Mamba in medical imaging step by step. We begin by defining core concepts of SSMs and models, including S4, S5, and S6, followed by an exploration of Mamba architectures such as pure Mamba, U-Net variants, and hybrid models with convolutional neural networks, transformers, and Graph Neural Networks. We also cover Mamba optimizations, techniques and adaptations, scanning, datasets, applications, experimental results, and conclude with its challenges and future directions in medical imaging. This review aims to demonstrate the transformative potential of Mamba in overcoming existing barriers within medical imaging while paving the way for innovative advancements in the field. A comprehensive list of Mamba architectures applied in the medical field, reviewed in this work, is available at Github.

Method: Tokenizes image and target object class, then decodes bounding box coordinates through an autoregressive transformer. Incorporates Direct Preference Optimization to leverage negative labels and refine spatial predictions.

Result: Achieves superior placement accuracy on OPA dataset compared to discriminative baselines and image composition approaches. Shows improved visual coherence in object insertion tasks when combined with off-the-shelf inpainting models.

Conclusion: The generative location model effectively addresses one-to-many nature of plausible locations and dataset sparsity, demonstrating utility in downstream applications like object insertion with better performance than state-of-the-art methods.

Abstract: Location modeling, or determining where non-existing objects could feasibly appear in a scene, has the potential to benefit numerous computer vision tasks, from automatic object insertion to scene creation in virtual reality. Yet, this capability remains largely unexplored to date. In this paper, we develop a generative location model that, given an object class and an image, learns to predict plausible bounding boxes for such an object. Our approach first tokenizes the image and target object class, then decodes bounding box coordinates through an autoregressive transformer. This formulation effectively addresses two core challenges in locatio modeling: the inherent one-to-many nature of plausible locations, and the sparsity of existing location modeling datasets, where fewer than 1% of valid placements are labeled. Furthermore, we incorporate Direct Preference Optimization to leverage negative labels, refining the spatial predictions. Empirical evaluations reveal that our generative location model achieves superior placement accuracy on the OPA dataset as compared to discriminative baselines and image composition approaches. We further test our model in the context of object insertion, where it proposes locations for an off-the-shelf inpainting model to render objects. In this respect, our proposal exhibits improved visual coherence relative to state-of-the-art instruction-tuned editing methods, demonstrating a high-performing location model’s utility in a downstream application.

Method: Use simple classification models with zero-shot embeddings to transform X-rays into radiology-specific labels, then combine with standard RAG and general-domain LLMs to generate reports without directly showing images to the LLM.

Result: Achieves better results across natural language and radiology metrics compared to other retrieval-based methods, and competitive results compared to fine-tuned vision-language models, without task-specific training of generative or image embedding models.

Conclusion: Demonstrates broader compatibility and synergy with fine-tuned methods for further enhancement, suggesting a viable alternative to traditional fine-tuning approaches in image captioning.

Abstract: In the current paradigm of image captioning, deep learning models are trained to generate text from image embeddings of latent features. We challenge the assumption that fine-tuning of large, bespoke models is required to improve model generation accuracy. Here we propose Label Boosted Retrieval Augmented Generation (LaB-RAG), a small-model-based approach to image captioning that leverages image descriptors in the form of categorical labels to boost standard retrieval augmented generation (RAG) with pretrained large language models (LLMs). We study our method in the context of radiology report generation (RRG) over MIMIC-CXR and CheXpert Plus. We argue that simple classification models combined with zero-shot embeddings can effectively transform X-rays into text-space as radiology-specific labels. In combination with standard RAG, we show that these derived text labels can be used with general-domain LLMs to generate radiology reports. Without ever training our generative language model or image embedding models specifically for the task, and without ever directly “showing” the LLM an X-ray, we demonstrate that LaB-RAG achieves better results across natural language and radiology language metrics compared with other retrieval-based RRG methods, while attaining competitive results compared to other fine-tuned vision-language RRG models. We further conduct extensive ablation experiments to better understand the components of LaB-RAG. Our results suggest broader compatibility and synergy with fine-tuned methods to further enhance RRG performance.

Method: Uses diffusion latent beam search with lookahead estimator to select optimal diffusion latents that maximize calibrated alignment rewards during inference. Focuses on reward calibration by weighting existing metrics to better correlate with human/VLM evaluations.

Result: Method improves perceptual quality evaluated on calibrated rewards, VLMs, and human assessment. Outperforms greedy search and best-of-N sampling with more efficient computational cost. Works with multiple generative models without parameter updates.

Conclusion: Inference-time compute should prioritize enabling lookahead estimators and increasing search budget rather than expanding denoising steps. The approach provides practical guidelines for improving video generation quality efficiently.

Abstract: The remarkable progress in text-to-video diffusion models enables the generation of photorealistic videos, although the content of these generated videos often includes unnatural movement or deformation, reverse playback, and motionless scenes. Recently, an alignment problem has attracted huge attention, where we steer the output of diffusion models based on some measure of the content’s goodness. Because there is a large room for improvement of perceptual quality along the frame direction, we should address which metrics we should optimize and how we can optimize them in the video generation. In this paper, we propose diffusion latent beam search with lookahead estimator, which can select a better diffusion latent to maximize a given alignment reward at inference time. We then point out that improving perceptual video quality with respect to alignment to prompts requires reward calibration by weighting existing metrics. This is because when humans or vision language models evaluate outputs, many previous metrics to quantify the naturalness of video do not always correlate with the evaluation. We demonstrate that our method improves the perceptual quality evaluated on the calibrated reward, VLMs, and human assessment, without model parameter update, and outputs the best generation compared to greedy search and best-of-N sampling under much more efficient computational cost. The experiments highlight that our method is beneficial to many capable generative models, and provide a practical guideline: we should prioritize the inference-time compute allocation into enabling the lookahead estimator and increasing the search budget, rather than expanding the denoising steps.

Method: Proposes PartSDF, a supervised implicit representation framework with simple but innovative architecture that explicitly models composite shapes with independent, controllable parts while maintaining shape consistency.

Result: PartSDF outperforms both supervised and unsupervised baselines in reconstruction and generation tasks, and demonstrates effectiveness as a structured shape prior for engineering applications.

Conclusion: The framework enables precise control over individual components while preserving overall coherence, making it suitable for real-world engineering design tasks.

Abstract: Accurate 3D shape representation is essential in engineering applications such as design, optimization, and simulation. In practice, engineering workflows require structured, part-based representations, as objects are inherently designed as assemblies of distinct components. However, most existing methods either model shapes holistically or decompose them without predefined part structures, limiting their applicability in real-world design tasks. We propose PartSDF, a supervised implicit representation framework that explicitly models composite shapes with independent, controllable parts while maintaining shape consistency. Thanks to its simple but innovative architecture, PartSDF outperforms both supervised and unsupervised baselines in reconstruction and generation tasks. We further demonstrate its effectiveness as a structured shape prior for engineering applications, enabling precise control over individual components while preserving overall coherence. Code available at https://github.com/cvlab-epfl/PartSDF.

Method: Introduces Alignment Regularity Characteristic (ARC) curves that show registration performance across various regularity levels, and uses HyperNetwork to continuously interpolate across the full regularization range for efficient curve construction.

Result: ARC curves reveal unique insights not evident from existing evaluation practices, demonstrating their effectiveness across various deep learning DIR methods with different architectures and transformation models.

Conclusion: Provides guidelines for nuanced model evaluation and selection using ARC curves, offering a holistic evaluation scheme for both practitioners and registration researchers.

Abstract: Evaluating deformable image registration (DIR) is challenging due to the inherent trade-off between achieving high alignment accuracy and maintaining deformation regularity. However, most existing DIR works either address this trade-off inadequately or overlook it altogether. In this paper, we highlight the issues with existing practices and propose an evaluation scheme that captures the trade-off continuously to holistically evaluate DIR methods. We first introduce the alignment regularity characteristic (ARC) curves, which describe the performance of a given registration method as a spectrum under various degrees of regularity. We demonstrate that the ARC curves reveal unique insights that are not evident from existing evaluation practices, using experiments on representative deep learning DIR methods with various network architectures and transformation models. We further adopt a HyperNetwork based approach that learns to continuously interpolate across the full regularization range, accelerating the construction and improving the sample density of ARC curves. Finally, we provide general guidelines for a nuanced model evaluation and selection using our evaluation scheme for both practitioners and registration researchers.

Method: Uses Bayesian statistics and image denoising principles to learn the score function of underlying point cloud distribution directly from noisy data, applying Tweedie’s formula for single-step denoising instead of iterative processes.

Result: Achieves state-of-the-art performance among unsupervised methods on standard benchmarks in Chamfer distance and point-to-mesh metrics, with strong generalization ability beyond training datasets.

Conclusion: Noise2Score3D overcomes key limitations of learning-based methods by eliminating the need for clean data and improving generalization, making it suitable for real-world point cloud denoising applications.

Abstract: Building on recent advances in Bayesian statistics and image denoising, we propose Noise2Score3D, a fully unsupervised framework for point cloud denoising. Noise2Score3D learns the score function of the underlying point cloud distribution directly from noisy data, eliminating the need for clean data during training. Using Tweedie’s formula, our method performs denoising in a single step, avoiding the iterative processes used in existing unsupervised methods, thus improving both accuracy and efficiency. Additionally, we introduce Total Variation for Point Clouds as a denoising quality metric, which allows for the estimation of unknown noise parameters. Experimental results demonstrate that Noise2Score3D achieves state-of-the-art performance on standard benchmarks among unsupervised learning methods in Chamfer distance and point-to-mesh metrics. Noise2Score3D also demonstrates strong generalization ability beyond training datasets. Our method, by addressing the generalization issue and challenge of the absence of clean data in learning-based methods, paves the way for learning-based point cloud denoising methods in real-world applications.

Method: Leverage clinically relevant tabular data to identify patient-level similarities and construct meaningful pairs for contrastive learning. Adapt k-NN algorithm for zero-shot prediction to overcome lack of zero-shot capability in unimodal representations.

Result: Method more effectively distinguishes patient subgroups in cardiac MR images. Outperforms conventional methods on downstream tasks including fine-tuning, linear probing, and zero-shot prediction of cardiovascular diseases and cardiac phenotypes. Also generalizes to natural images on car advertisement dataset.

Conclusion: Incorporating tabular data guidance yields stronger visual representations than conventional methods relying solely on image augmentation or combined image-tabular embeddings. The framework enables semantically aligned representation learning without requiring joint embeddings across modalities.

Abstract: Contrastive learning methods in computer vision typically rely on augmented views of the same image or multimodal pretraining strategies that align paired modalities. However, these approaches often overlook semantic relationships between distinct instances, leading to false negatives when semantically similar samples are treated as negatives. This limitation is especially critical in medical imaging domains such as cardiology, where demographic and clinical attributes play a critical role in assessing disease risk and patient outcomes. We introduce a tabular-guided contrastive learning framework that leverages clinically relevant tabular data to identify patient-level similarities and construct more meaningful pairs, enabling semantically aligned representation learning without requiring joint embeddings across modalities. Additionally, we adapt the k-NN algorithm for zero-shot prediction to overcome the lack of zero-shot capability in unimodal representations. We demonstrate the strength of our methods using a large cohort of short-axis cardiac MR images and clinical attributes, where tabular data helps to more effectively distinguish between patient subgroups. Evaluation on downstream tasks, including fine-tuning, linear probing, and zero-shot prediction of cardiovascular artery diseases and cardiac phenotypes, shows that incorporating tabular data guidance yields stronger visual representations than conventional methods that rely solely on image augmentation or combined image-tabular embeddings. Further, we show that our method can generalize to natural images by evaluating it on a car advertisement dataset. The code will be available on GitHub upon acceptance.

Method: 3D patch-based Residual U-Net trained with soft labels from atlas-based segmentations on 532 WB-DWI scans from prostate cancer and multiple myeloma patients, tested on 45 patients.

Result: Achieved dice scores: 0.67 (whole skeleton), 0.76 (skeleton excluding ribcage), 0.83 (internal organs), 0.86 (spinal canal); 12x faster than atlas method; radiologists rated outputs as good/excellent.

Conclusion: The model provides fast, reproducible anatomical localization on WB-DWI, enabling non-invasive imaging biomarker quantification for disease staging and treatment response assessment.

Abstract: Background: Apparent Diffusion Coefficient (ADC) values and Total Diffusion Volume (TDV) from Whole-body diffusion-weighted MRI (WB-DWI) are recognized cancer imaging biomarkers. However, manual disease delineation for ADC and TDV measurements is unfeasible in clinical practice, demanding automation. As a first step, we propose an algorithm to generate fast and reproducible probability maps of the skeleton, adjacent internal organs (liver, spleen, urinary bladder, and kidneys), and spinal canal. Methods: We developed an automated deep-learning pipeline based on a 3D patch-based Residual U-Net architecture that localises and delineates these anatomical structures on WB-DWI. The algorithm was trained using “soft labels” (non-binary segmentations) derived from a computationally intensive atlas-based approach. For training and validation, we employed a multi-centre WB-DWI dataset comprising 532 scans from patients with Advanced Prostate Cancer (APC) or Multiple Myeloma (MM), with testing on 45 patients. Results: Our weakly-supervised deep learning model achieved an average dice score of 0.67 for whole skeletal delineation, 0.76 when excluding ribcage, 0.83 for internal organs, and 0.86 for spinal canal, with average surface distances below 3mm. Relative median ADC differences between automated and manual full-body delineations were below 10%. The model was 12x faster than the atlas-based registration algorithm (25 sec vs. 5 min). Two experienced radiologists rated the model’s outputs as either “good” or “excellent” on test scans, with inter-reader agreement from fair to substantial (Gwet’s AC1 = 0.27-0.72). Conclusion: The model offers fast, reproducible probability maps for localising and delineating body regions on WB-DWI, potentially enabling non-invasive imaging biomarker quantification to support disease staging and treatment response assessment.

Method: Leverages multimodal encoders to extract short segment representations, then uses masked-embedding autoencoder to capture high-level interactions across segments, treating short- and long-span dependencies as separate tasks.

Result: Achieves state-of-the-art results on LVU, COIN, and Breakfast benchmarks using only simple classification heads, with efficient training that processes much longer videos than existing methods.

Conclusion: LV-MAE provides an effective self-supervised approach for long video representation learning that outperforms existing methods while being highly efficient and scalable.

Abstract: In this work, we introduce long-video masked-embedding autoencoders (LV-MAE), a self-supervised learning framework for long video representation. Our approach treats short- and long-span dependencies as two separate tasks. Such decoupling allows for a more intuitive video processing where short-span spatiotemporal primitives are first encoded and are then used to capture long-range dependencies across consecutive video segments. To achieve this, we leverage advanced off-the-shelf multimodal encoders to extract representations from short segments within the long video, followed by pre-training a masked-embedding autoencoder capturing high-level interactions across segments. LV-MAE is highly efficient to train and enables the processing of much longer videos by alleviating the constraint on the number of input frames. Furthermore, unlike existing methods that typically pre-train on short-video datasets, our approach offers self-supervised pre-training using long video samples (e.g., 20+ minutes video clips) at scale. Using LV-MAE representations, we achieve state-of-the-art results on three long-video benchmarks – LVU, COIN, and Breakfast – employing only a simple classification head for either attentive or linear probing. Finally, to assess LV-MAE pre-training and visualize its reconstruction quality, we leverage the video-language aligned space of short video representations to monitor LV-MAE through video-text retrieval. Code is available at https://github.com/amazon-science/lv-mae.

Method: Proximal Policy Optimization (PPO) reinforcement learning algorithm is used to iteratively learn optimal placement locations based on population density, traffic patterns, industrial influence, and green space constraints.

Result: The approach was benchmarked against conventional strategies (random and greedy AQI-based methods) using multi-dimensional metrics including AQI improvement, spatial coverage, population/traffic impact, and spatial entropy.

Conclusion: The DRL framework provides an effective solution for optimizing air purification booth placement in urban environments, outperforming traditional methods by adapting to dynamic conditions and multiple environmental factors.

Abstract: This is the preprint version of the article published in IEEE Access vol. 13, pp. 146503–146526, 2025, doi:10.1109/ACCESS.2025.3599541. Please cite the published version. Urban air pollution remains a pressing global concern, particularly in densely populated and traffic-intensive metropolitan areas like Delhi, where exposure to harmful pollutants severely impacts public health. Delhi, being one of the most polluted cities globally, experiences chronic air quality issues due to vehicular emissions, industrial activities, and construction dust, which exacerbate its already fragile atmospheric conditions. Traditional pollution mitigation strategies, such as static air purifying installations, often fail to maximize their impact due to suboptimal placement and limited adaptability to dynamic urban environments. This study presents a novel deep reinforcement learning (DRL) framework to optimize the placement of air purification booths to improve the air quality index (AQI) in the city of Delhi. We employ Proximal Policy Optimization (PPO), a state-of-the-art reinforcement learning algorithm, to iteratively learn and identify high-impact locations based on multiple spatial and environmental factors, including population density, traffic patterns, industrial influence, and green space constraints. Our approach is benchmarked against conventional placement strategies, including random and greedy AQI-based methods, using multi-dimensional performance evaluation metrics such as AQI improvement, spatial coverage, population and traffic impact, and spatial entropy.

Method: Proposes AuxDet framework with: 1) High-dimensional fusion module using MLPs to dynamically integrate metadata semantics with visual features, 2) Lightweight prior-initialized enhancement module using 1D convolutional blocks to refine fused features and recover fine-grained target cues.

Result: Extensive experiments on WideIRSTD-Full benchmark show AuxDet consistently outperforms state-of-the-art methods, validating the critical role of auxiliary information in improving robustness and accuracy.

Conclusion: Auxiliary metadata plays a critical role in improving IRSTD performance across omni-domain scenarios, and the proposed AuxDet framework effectively leverages this information for scene-aware optimization.

Abstract: Omni-domain infrared small target detection (Omni-IRSTD) poses formidable challenges, as a single model must seamlessly adapt to diverse imaging systems, varying resolutions, and multiple spectral bands simultaneously. Current approaches predominantly rely on visual-only modeling paradigms that not only struggle with complex background interference and inherently scarce target features, but also exhibit limited generalization capabilities across complex omni-scene environments where significant domain shifts and appearance variations occur. In this work, we reveal a critical oversight in existing paradigms: the neglect of readily available auxiliary metadata describing imaging parameters and acquisition conditions, such as spectral bands, sensor platforms, resolution, and observation perspectives. To address this limitation, we propose the Auxiliary Metadata Driven Infrared Small Target Detector (AuxDet), a novel multimodal framework that is the first to incorporate metadata into the IRSTD paradigm for scene-aware optimization. Through a high-dimensional fusion module based on multi-layer perceptrons (MLPs), AuxDet dynamically integrates metadata semantics with visual features, guiding adaptive representation learning for each individual sample. Additionally, we design a lightweight prior-initialized enhancement module using 1D convolutional blocks to further refine fused features and recover fine-grained target cues. Extensive experiments on the challenging WideIRSTD-Full benchmark demonstrate that AuxDet consistently outperforms state-of-the-art methods, validating the critical role of auxiliary information in improving robustness and accuracy in omni-domain IRSTD tasks. Code is available at https://github.com/GrokCV/AuxDet.

Method: Constructs adaptive multimodal graphs with nodes representing frames, objects, and text annotations, connected by spatial, temporal, and semantic edges. Uses Graph Attention Network with task-specific attention to dynamically evolve graph structures based on learned interactions.

Result: Consistently outperforms state-of-the-art baselines on diverse benchmark datasets, demonstrating robust and generalizable action recognition performance.

Conclusion: Combining foundation models with dynamic graph-based reasoning provides a powerful approach for fine-grained action recognition, enabling flexible and context-aware reasoning.

Abstract: Foundation models have ushered in a new era for multimodal video understanding by enabling the extraction of rich spatiotemporal and semantic representations. In this work, we introduce a novel graph-based framework that integrates a vision-language foundation, leveraging VideoMAE for dynamic visual encoding and BERT for contextual textual embedding, to address the challenge of recognizing fine-grained bimanual manipulation actions. Departing from conventional static graph architectures, our approach constructs an adaptive multimodal graph where nodes represent frames, objects, and textual annotations, and edges encode spatial, temporal, and semantic relationships. These graph structures evolve dynamically based on learned interactions, allowing for flexible and context-aware reasoning. A task-specific attention mechanism within a Graph Attention Network further enhances this reasoning by modulating edge importance based on action semantics. Through extensive evaluations on diverse benchmark datasets, we demonstrate that our method consistently outperforms state-of-the-art baselines, underscoring the strength of combining foundation models with dynamic graph-based reasoning for robust and generalizable action recognition.

Method: Proposes semantic-aware permutation using k-means to cluster tokens by semantic similarity, top-p dynamic budget control, and customized kernel implementations for efficient computation without padding.

Result: Achieves up to 2.30x and 1.89x speedup on HunyuanVideo and Wan 2.1 respectively, while maintaining PSNR of up to 30 and 26.

Conclusion: SVG2 provides a Pareto frontier trade-off between generation quality and efficiency through semantic-aware sparse attention, offering significant acceleration for video generation DiTs.

Abstract: Diffusion Transformers (DiTs) are essential for video generation but suffer from significant latency due to the quadratic complexity of attention. By computing only critical tokens, sparse attention reduces computational costs and offers a promising acceleration approach. However, we identify that existing methods fail to approach optimal generation quality under the same computation budget for two reasons: (1) Inaccurate critical token identification: current methods cluster tokens based on position rather than semantics, leading to imprecise aggregated representations. (2) Excessive computation waste: critical tokens are scattered among non-critical ones, leading to wasted computation on GPUs, which are optimized for processing contiguous tokens. In this paper, we propose SVG2, a training-free framework that maximizes identification accuracy and minimizes computation waste, achieving a Pareto frontier trade-off between generation quality and efficiency. The core of SVG2 is semantic-aware permutation, which clusters and reorders tokens based on semantic similarity using k-means. This approach ensures both a precise cluster representation, improving identification accuracy, and a densified layout of critical tokens, enabling efficient computation without padding. Additionally, SVG2 integrates top-p dynamic budget control and customized kernel implementations, achieving up to 2.30x and 1.89x speedup while maintaining a PSNR of up to 30 and 26 on HunyuanVideo and Wan 2.1, respectively. Our code is open-sourced at \href{https://github.com/svg-project/Sparse-VideoGen}{https://github.com/svg-project/Sparse-VideoGen}.

Method: Visualize-then-Retrieve (VisRet) first projects textual queries into the image modality via T2I generation, then performs retrieval within the image modality to bypass weaknesses of cross-modal retrievers.

Result: VisRet substantially outperforms cross-modal similarity matching across four benchmarks, improving nDCG@30 by 0.125 on average with CLIP and by 0.121 with E5-V. For downstream QA, it increases accuracy by 3.8-15.7% in top-1 retrieval and 3.9-11.1% in top-10 retrieval.

Conclusion: VisRet provides a practical and principled path that energizes further advances in vision-language retrieval by effectively addressing the limitations of cross-modal similarity alignment.

Abstract: Text-to-image retrieval (T2I retrieval) remains challenging because cross-modal embeddings often behave as bags of concepts and underrepresent structured visual relationships such as pose and viewpoint. We propose Visualize-then-Retrieve (VisRet), a new paradigm for T2I retrieval that mitigates this limitation of cross-modal similarity alignment. VisRet first projects textual queries into the image modality via T2I generation. Then, it performs retrieval within the image modality to bypass the weaknesses of cross-modal retrievers in recognizing subtle visual-spatial features. Across four benchmarks (Visual-RAG, INQUIRE-Rerank, Microsoft COCO, and our new Visual-RAG-ME featuring multi-entity comparisons), VisRet substantially outperforms cross-modal similarity matching and baselines that recast T2I retrieval as text-to-text similarity matching, improving nDCG@30 by 0.125 on average with CLIP as the retriever and by 0.121 with E5-V. For downstream question answering, VisRet increases accuracy on Visual-RAG and Visual-RAG-ME by 3.8% and 15.7% in top-1 retrieval, and by 3.9% and 11.1% in top-10 retrieval. Ablation studies show compatibility with different T2I instruction LLMs, T2I generation models, and downstream LLMs. VisRet provides a practical and principled path that energizes further advances in vision-language retrieval. Our code and the Visual-RAG-ME benchmark will be publicly released.

Method: Fine-tunes pre-trained video diffusion models using LLMs to infer physical context from text prompts, MLLMs to verify intermediate latent variables against physical rules, transforms LLM output into continuous signals, and uses training objectives for physical accuracy and semantic alignment with attention injection for correcting physical failures.

Result: Extensive experiments on public benchmarks show that DiffPhy produces state-of-the-art results across diverse physics-related scenarios, generating physically-correct and photo-realistic videos.

Conclusion: DiffPhy effectively bridges the gap between visual quality and physical accuracy in video generation by leveraging language models to incorporate physical understanding into diffusion models, achieving superior performance in physics-related video synthesis.

Abstract: Recent video diffusion models have demonstrated their great capability in generating visually-pleasing results, while synthesizing the correct physical effects in generated videos remains challenging. The complexity of real-world motions, interactions, and dynamics introduce great difficulties when learning physics from data. In this work, we propose DiffPhy, a generic framework that enables physically-correct and photo-realistic video generation by fine-tuning a pre-trained video diffusion model. Our method leverages large language models (LLMs) to infer rich physical context from the text prompt. To incorporate this context into the video diffusion model, we use a multimodal large language model (MLLM) to verify intermediate latent variables against the inferred physical rules, guiding the gradient updates of model accordingly. Textual output of LLM is transformed into continuous signals. We then formulate a set of training objectives that jointly ensure physical accuracy and semantic alignment with the input text. Additionally, failure facts of physical phenomena are corrected via attention injection. We also establish a high-quality physical video dataset containing diverse phyiscal actions and events to facilitate effective finetuning. Extensive experiments on public benchmarks demonstrate that DiffPhy is able to produce state-of-the-art results across diverse physics-related scenarios. Our project page is available at https://bwgzk-keke.github.io/DiffPhy/.

Method: Trains two 3DGS models in parallel with consistency constraints, uses divergent masking (multi-cue adaptive mask and self-supervised soft mask) to prevent confirmation bias, and introduces Dynamic EMA Proxy for efficiency.

Result: Extensive experiments show the method consistently outperforms existing approaches on challenging real-world datasets while achieving high efficiency.

Conclusion: The proposed framework effectively suppresses inconsistent artifacts in 3D reconstruction from in-the-wild images through asymmetric training and divergent masking strategies.

Abstract: 3D reconstruction from in-the-wild images remains a challenging task due to inconsistent lighting conditions and transient distractors. Existing methods typically rely on heuristic strategies to handle the low-quality training data, which often struggle to produce stable and consistent reconstructions, frequently resulting in visual artifacts.In this work, we propose \modelname{}, a novel framework that leverages the stochastic nature of these artifacts: they tend to vary across different training runs due to minor randomness. Specifically, our method trains two 3D Gaussian Splatting (3DGS) models in parallel, enforcing a consistency constraint that encourages convergence on reliable scene geometry while suppressing inconsistent artifacts. To prevent the two models from collapsing into similar failure modes due to confirmation bias, we introduce a divergent masking strategy that applies two complementary masks: a multi-cue adaptive mask and a self-supervised soft mask, which leads to an asymmetric training process of the two models, reducing shared error modes. In addition, to improve the efficiency of model training, we introduce a lightweight variant called Dynamic EMA Proxy, which replaces one of the two models with a dynamically updated Exponential Moving Average (EMA) proxy, and employs an alternating masking strategy to preserve divergence. Extensive experiments on challenging real-world datasets demonstrate that our method consistently outperforms existing approaches while achieving high efficiency. See the project website at https://steveli88.github.io/AsymGS.

Method: Two components: (1) ZoomText - coarse-to-fine strategy to identify text regions without external detectors; (2) Grounded Layer Correction - leverages internal representations from less hallucination-prone layers to guide decoding.

Result: Effectively mitigates semantic hallucination and achieves strong performance on public benchmarks for scene text spotting and understanding.

Conclusion: The proposed training-free framework successfully reduces semantic hallucination in LMMs while maintaining performance on meaningful text understanding tasks.

Abstract: Large Multimodal Models (LMMs) have achieved impressive progress in visual perception and reasoning. However, when confronted with visually ambiguous or non-semantic scene text, they often struggle to accurately spot and understand the content, frequently generating semantically plausible yet visually incorrect answers, which we refer to as semantic hallucination. In this work, we investigate the underlying causes of semantic hallucination and identify a key finding: Transformer layers in LLM with stronger attention focus on scene text regions are less prone to producing semantic hallucinations. Thus, we propose a training-free semantic hallucination mitigation framework comprising two key components: (1) ZoomText, a coarse-to-fine strategy that identifies potential text regions without external detectors; and (2) Grounded Layer Correction, which adaptively leverages the internal representations from layers less prone to hallucination to guide decoding, correcting hallucinated outputs for non-semantic samples while preserving the semantics of meaningful ones. To enable rigorous evaluation, we introduce TextHalu-Bench, a benchmark of 1,740 samples spanning both semantic and non-semantic cases, with manually curated question answer pairs designed to probe model hallucinations. Extensive experiments demonstrate that our method not only effectively mitigates semantic hallucination but also achieves strong performance on public benchmarks for scene text spotting and understanding.

Method: Developed a hierarchical annotation pipeline to generate over 5k QA pairs for 811 real indoor scenes, covering 10 atomic spatial capabilities combined into 8 compositional capabilities, with evaluation settings including point cloud input and multi-choice QA.

Result: Even the most advanced MLLMs significantly lag behind humans in spatial intelligence. The counting capability shortcoming was identified as a major limitation affecting compositional spatial capabilities.

Conclusion: SpaCE-10 provides a comprehensive benchmark revealing substantial gaps in MLLMs’ spatial intelligence, with counting capability being a critical bottleneck that needs addressing for improved compositional spatial reasoning.

Abstract: Multimodal Large Language Models (MLLMs) have achieved remarkable progress in various multimodal tasks. To pursue higher intelligence in space, MLLMs require integrating multiple spatial capabilities, even for handling simple and normal tasks. However, existing benchmarks struggle to comprehensively evaluate the spatial intelligence of common MLLMs from the atomic level to the compositional level. To fill this gap, we present SpaCE-10, a comprehensive benchmark for compositional spatial evaluations. In SpaCE-10, we define 10 atomic spatial capabilities, which are combined to form 8 compositional capabilities. Based on these definitions, we propose a novel hierarchical annotation pipeline to generate high-quality and diverse question-answer (QA) pairs. With over 150+ hours of human expert effort, we obtain over 5k QA pairs for 811 real indoor scenes in SpaCE-10, which covers various evaluation settings like point cloud input and multi-choice QA. We conduct an extensive evaluation of common MLLMs on SpaCE-10 and find that even the most advanced MLLM still lags behind humans by large margins. Through our careful study, we also draw several significant findings that benefit the MLLM community. For example, we reveal that the shortcoming of counting capability greatly limits the compositional spatial capabilities of existing MLLMs.

Method: Proposes CP-based low-rank tensor function parameterized by neural networks, uses variational Schatten-p quasi-norm to prune redundant rank-1 components, and introduces SVD-free smoothness regularization based on Jacobian spectral norm with Hutchinson’s trace estimator.

Result: Method demonstrates superiority in multi-dimensional data recovery tasks including image inpainting, denoising, and point cloud upsampling compared to state-of-the-art approaches.

Conclusion: The proposed approach provides an interpretable, sparse CP decomposition with theoretical guarantees and practical effectiveness across various data recovery applications, serving as an alternative to TV regularization.

Abstract: Higher-order tensors are well-suited for representing multi-dimensional data, such as images and videos, which typically characterize low-rank structures. Low-rank tensor decomposition has become essential in machine learning and computer vision, but existing methods like Tucker decomposition offer flexibility at the expense of interpretability. The CANDECOMP/PARAFAC (CP) decomposition provides a natural and interpretable structure, while obtaining a sparse solutions remains challenging. Leveraging the rich properties of CP decomposition, we propose a CP-based low-rank tensor function parameterized by neural networks (NN) for implicit neural representation. This approach can model the tensor both on-grid and beyond grid, fully utilizing the non-linearity of NN with theoretical guarantees on excess risk bounds. To achieve sparser CP decomposition, we introduce a variational Schatten-p quasi-norm to prune redundant rank-1 components and prove that it serves as a common upper bound for the Schatten-p quasi-norms of arbitrary unfolding matrices. For smoothness, we propose a regularization term based on the spectral norm of the Jacobian and Hutchinson’s trace estimator. The proposed smoothness regularization is SVD-free and avoids explicit chain rule derivations. It can serve as an alternative to Total Variation (TV) regularization in image denoising tasks and is naturally applicable to implicit neural representation. Extensive experiments on multi-dimensional data recovery tasks, including image inpainting, denoising, and point cloud upsampling, demonstrate the superiority and versatility of our method compared to state-of-the-art approaches. The code is available at https://github.com/CZY-Code/CP-Pruner.

Method: For classification: Chain of Counterfactual Thought framework using counterfactual reasoning, ASR-enhanced video content, and rationale refinement. For generation: framework combining ASR output with LLM to produce coherent updates. Also introduces a new benchmark dataset with strengthener/weakener annotations and LLM-based evaluation metric.

Result: Experimental results demonstrate significant improvements in enhancing dynamic reasoning capabilities of VLMMs.

Conclusion: The proposed DVidE task and frameworks effectively address the limitations of current VLMMs in abstract and adaptive reasoning, enabling models to think like doubters and constantly update interpretations based on evolving evidence.

Abstract: Video Large Multimodal Models (VLMMs) have made impressive strides in understanding video content, but they often struggle with abstract and adaptive reasoning-the ability to revise their interpretations when new information emerges. In reality, conclusions are rarely set in stone; additional context can strengthen or weaken an initial inference. To address this, we introduce Defeasible Video Entailment (DVidE), a new task that challenges models to think like doubters, constantly updating their reasoning based on evolving evidence. In DVidE, given a video premise and a textual hypothesis, models must determine whether a new update strengthens or weakens the hypothesis (classification version) or generate a coherent update that modifies the entailment relationship (generation version). For solving the classification task, we propose the Chain of Counterfactual Thought framework, utilizing counterfactual reasoning, ASR-enhanced video content, and rationale refinement to reduce inference bias. For the generation task, we develop a framework that combines ASR output with a Large Language Model (LLM) to produce coherent, contextually relevant updates aligned with the intended strengthener or weakener goals. Additionally, we introduce a novel benchmark dataset, with strengthener/weakener annotations and an LLM-based evaluation metric specifically designed for assessing generative performance. Experimental results demonstrate significant improvements, highlighting our proposed method in enhancing dynamic reasoning capabilities of VLMMs.

Method: Built on YOLOv11n with GhostConv and VoVGSCSPC modules to reduce computation while enhancing multi-scale features. Includes P2-level branch for small-object detection, lightweight detection head (LEDH), and hybrid training with NWD loss, BCKD knowledge distillation, and Albumentations-based augmentation.

Result: Achieves 87.0% mAP, outperforming YOLOv11n baseline by 5.8%. After TensorRT FP16 quantization, runs at 139.5 FPS on Jetson AGX Xavier with 12.4% speedup over P2-enhanced YOLOv11.

Conclusion: The framework is suitable for fast, low-latency road condition perception in embedded suspension control systems.

Abstract: Speed bumps and potholes are the most common road anomalies, significantly affecting ride comfort and vehicle stability. Preview-based suspension control mitigates their impact by detecting such irregularities in advance and adjusting suspension parameters proactively. Accurate and real-time detection is essential, but embedded deployment is constrained by limited computational resources and the small size of targets in input images.To address these challenges, this paper proposes SBP-YOLO, an efficient detection framework for speed bumps and potholes in embedded systems. Built upon YOLOv11n, it integrates GhostConv and VoVGSCSPC modules in the backbone and neck to reduce computation while enhancing multi-scale semantic features. A P2-level branch improves small-object detection, and a lightweight and efficient detection head (LEDH) maintains accuracy with minimal overhead. A hybrid training strategy further enhances robustness under varying road and environmental conditions, combining NWD loss, BCKD knowledge distillation, and Albumentations-based augmentation. Experiments show that SBP-YOLO achieves 87.0% mAP, outperforming the YOLOv11n baseline by 5.8%. After TensorRT FP16 quantization, it runs at 139.5 FPS on Jetson AGX Xavier, yielding a 12.4% speedup over the P2-enhanced YOLOv11. These results demonstrate the framework’s suitability for fast, low-latency road condition perception in embedded suspension control systems.

Method: Proposes mixture-of-experts (MoE) decoding module that dynamically routes image features based on recognized blur degradation for end-to-end restoration.

Result: Achieves performance comparable to dedicated task-specific models and shows promising generalization to unseen blur scenarios with appropriate expert selection.

Conclusion: The unified MoE-based approach provides an efficient all-in-one solution for diverse blur degradations including motion, low-light, and defocus blur.

Abstract: Image deblurring, removing blurring artifacts from images, is a fundamental task in computational photography and low-level computer vision. Existing approaches focus on specialized solutions tailored to particular blur types, thus, these solutions lack generalization. This limitation in current methods implies requiring multiple models to cover several blur types, which is not practical in many real scenarios. In this paper, we introduce the first all-in-one deblurring method capable of efficiently restoring images affected by diverse blur degradations, including global motion, local motion, blur in low-light conditions, and defocus blur. We propose a mixture-of-experts (MoE) decoding module, which dynamically routes image features based on the recognized blur degradation, enabling precise and efficient restoration in an end-to-end manner. Our unified approach not only achieves performance comparable to dedicated task-specific models, but also shows promising generalization to unseen blur scenarios, particularly when leveraging appropriate expert selection. Code available at https://github.com/cidautai/DeMoE.

Method: Uses diffusion-based framework with DC-AE for high-compression latent space generation, pretrained diffusion transformer with linear attention for efficiency, and CrossStitch convolutional module for cross-map consistency.

Result: HiMat achieves high-fidelity 4K SVBRDF generation with superior efficiency, structural consistency, and diversity compared to prior methods.

Conclusion: The framework successfully addresses 4K SVBRDF generation challenges and generalizes to related applications like intrinsic decomposition.

Abstract: Creating ultra-high-resolution spatially varying bidirectional reflectance functions (SVBRDFs) is critical for photorealistic 3D content creation, to faithfully represent fine-scale surface details required for close-up rendering. However, achieving 4K generation faces two key challenges: (1) the need to synthesize multiple reflectance maps at full resolution, which multiplies the pixel budget and imposes prohibitive memory and computational cost, and (2) the requirement to maintain strong pixel-level alignment across maps at 4K, which is particularly difficult when adapting pretrained models designed for the RGB image domain. We introduce HiMat, a diffusion-based framework tailored for efficient and diverse 4K SVBRDF generation. To address the first challenge, HiMat performs generation in a high-compression latent space via DC-AE, and employs a pretrained diffusion transformer with linear attention to improve per-map efficiency. To address the second challenge, we propose CrossStitch, a lightweight convolutional module that enforces cross-map consistency without incurring the cost of global attention. Our experiments show that HiMat achieves high-fidelity 4K SVBRDF generation with superior efficiency, structural consistency, and diversity compared to prior methods. Beyond materials, our framework also generalizes to related applications such as intrinsic decomposition.

Method: Proposes CMB-Net with: 1) LLM-generated textual descriptions of manipulated regions, 2) ITCAM module to weight text features by quantifying image-text ambiguity, 3) ITIM module for fine-grained visual-text feature alignment, and 4) RED decoder using invertible neural networks to preserve boundary information.

Result: Extensive experiments show CMB-Net outperforms most existing IML models, demonstrating improved accuracy in localizing manipulated image regions.

Conclusion: The proposed multimodal approach combining visual and semantic textual information, with mechanisms to handle LLM hallucinations and preserve boundaries, significantly advances image manipulation localization performance.

Abstract: The existing image manipulation localization (IML) models mainly relies on visual cues, but ignores the semantic logical relationships between content features. In fact, the content semantics conveyed by real images often conform to human cognitive laws. However, image manipulation technology usually destroys the internal relationship between content features, thus leaving semantic clues for IML. In this paper, we propose a cognition inspired multimodal boundary preserving network (CMB-Net). Specifically, CMB-Net utilizes large language models (LLMs) to analyze manipulated regions within images and generate prompt-based textual information to compensate for the lack of semantic relationships in the visual information. Considering that the erroneous texts induced by hallucination from LLMs will damage the accuracy of IML, we propose an image-text central ambiguity module (ITCAM). It assigns weights to the text features by quantifying the ambiguity between text and image features, thereby ensuring the beneficial impact of textual information. We also propose an image-text interaction module (ITIM) that aligns visual and text features using a correlation matrix for fine-grained interaction. Finally, inspired by invertible neural networks, we propose a restoration edge decoder (RED) that mutually generates input and output features to preserve boundary information in manipulated regions without loss. Extensive experiments show that CMB-Net outperforms most existing IML models. Our code is available on https://github.com/vpsg-research/CMB-Net.

Method: Developed a systematic methodology using the HistVis dataset - 30,000 synthetic images generated by three state-of-the-art diffusion models with carefully designed prompts depicting universal human activities across different historical periods.

Result: TTI models frequently stereotype past eras with unstated stylistic cues, introduce anachronisms (modern artifacts in pre-modern contexts), and fail to reflect historically plausible demographic patterns in racial and gender distributions.

Conclusion: The work provides a scalable methodology and benchmark for assessing historical representation in generated imagery, offering an initial step toward building more historically accurate and culturally aligned TTI models.

Abstract: As Text-to-Image (TTI) diffusion models become increasingly influential in content creation, growing attention is being directed toward their societal and cultural implications. While prior research has primarily examined demographic and cultural biases, the ability of these models to accurately represent historical contexts remains largely underexplored. In this work, we present a systematic and reproducible methodology for evaluating how TTI systems depict different historical periods. For this purpose, we introduce the HistVis dataset, a curated collection of 30,000 synthetic images generated by three state-of-the-art diffusion models using carefully designed prompts depicting universal human activities across different historical periods. We evaluate generated imagery across three key aspects: (1) Implicit Stylistic Associations: examining default visual styles associated with specific eras; (2) Historical Consistency: identifying anachronisms such as modern artifacts in pre-modern contexts; and (3) Demographic Representation: comparing generated racial and gender distributions against historically plausible baselines. Our findings reveal systematic inaccuracies in historically themed generated imagery, as TTI models frequently stereotype past eras by incorporating unstated stylistic cues, introduce anachronisms, and fail to reflect plausible demographic patterns. By offering a scalable methodology and benchmark for assessing historical representation in generated imagery, this work provides an initial step toward building more historically accurate and culturally aligned TTI models.

Method: Created two realistic benchmarks: Domain RICO (domain shifts with fixed classes) and Expanding-Classes RICO (new domains and classes per step), built from 14 diverse datasets covering real/synthetic domains, varying conditions, sensors, perspectives, and labeling policies.

Result: All tested IL methods underperformed in both adaptability to new data and retention of old knowledge. Replaying a small amount of previous data outperformed all IL methods, though individual training on the full data remained superior.

Conclusion: The performance gap is attributed to weak teachers in distillation, single models’ inability to handle diverse tasks, and insufficient plasticity. The benchmarks reveal critical limitations in current IL approaches for realistic scenarios.

Abstract: Incremental Learning (IL) trains models sequentially on new data without full retraining, offering privacy, efficiency, and scalability. IL must balance adaptability to new data with retention of old knowledge. However, evaluations often rely on synthetic, simplified benchmarks, obscuring real-world IL performance. To address this, we introduce two Realistic Incremental Object Detection Benchmarks (RICO): Domain RICO (D-RICO) features domain shifts with a fixed class set, and Expanding-Classes RICO (EC-RICO) integrates new domains and classes per IL step. Built from 14 diverse datasets covering real and synthetic domains, varying conditions (e.g., weather, time of day), camera sensors, perspectives, and labeling policies, both benchmarks capture challenges absent in existing evaluations. Our experiments show that all IL methods underperform in adaptability and retention, while replaying a small amount of previous data already outperforms all methods. However, individual training on the data remains superior. We heuristically attribute this gap to weak teachers in distillation, single models’ inability to manage diverse tasks, and insufficient plasticity. Our code will be made publicly available.

Method: Two single-step diffusion models handle forward and inverse rendering with mutual reinforcement, extending intrinsic decomposition to indoor/outdoor scenes and introducing cycle consistency mechanism.

Result: State-of-the-art performance across diverse scenes with substantially faster inference speed compared to other diffusion-based methods; can transfer to video decomposition training-free with reduced temporal inconsistency.

Conclusion: Ouroboros provides an effective framework for coherent forward and inverse rendering with fast inference and good generalization to video applications.

Abstract: While multi-step diffusion models have advanced both forward and inverse rendering, existing approaches often treat these problems independently, leading to cycle inconsistency and slow inference speed. In this work, we present Ouroboros, a framework composed of two single-step diffusion models that handle forward and inverse rendering with mutual reinforcement. Our approach extends intrinsic decomposition to both indoor and outdoor scenes and introduces a cycle consistency mechanism that ensures coherence between forward and inverse rendering outputs. Experimental results demonstrate state-of-the-art performance across diverse scenes while achieving substantially faster inference speed compared to other diffusion-based methods. We also demonstrate that Ouroboros can transfer to video decomposition in a training-free manner, reducing temporal inconsistency in video sequences while maintaining high-quality per-frame inverse rendering.

Method: Analyzed three different prompt-based methods under complex domain-incremental learning, combined with replaying small portions of previous data, and tested prompt length and initialization.

Result: Prompt-based approaches alone underperformed, but combining visual prompts with limited data replay achieved the best results.

Conclusion: Visual prompts need to be combined with data replay for effective incremental object detection, providing insights for advancing prompt-based incremental learning in IOD.

Abstract: Visual prompt-based methods have seen growing interest in incremental learning (IL) for image classification. These approaches learn additional embedding vectors while keeping the model frozen, making them efficient to train. However, no prior work has applied such methods to incremental object detection (IOD), leaving their generalizability unclear. In this paper, we analyze three different prompt-based methods under a complex domain-incremental learning setting. We additionally provide a wide range of reference baselines for comparison. Empirically, we show that the prompt-based approaches we tested underperform in this setting. However, a strong yet practical method, combining visual prompts with replaying a small portion of previous data, achieves the best results. Together with additional experiments on prompt length and initialization, our findings offer valuable insights for advancing prompt-based IL in IOD.

Method: Two-stage approach: 1) 3D structure transformer generates human motion sequences from text prompts, 2) Appearance synthesis guided by structural sequence with Human-Aware Dynamic Control modules and contact constraints for improved interactions.

Result: MoSA substantially outperforms existing approaches across majority of evaluation metrics, demonstrating superior structural and appearance fidelity in generated human videos.

Conclusion: The proposed decoupled structure-appearance generation approach with specialized motion modeling and interaction constraints effectively addresses limitations in existing human video generation systems.

Abstract: Existing video generation models predominantly emphasize appearance fidelity while exhibiting limited ability to synthesize complex human motions, such as whole-body movements, long-range dynamics, and fine-grained human-environment interactions. This often leads to unrealistic or physically implausible movements with inadequate structural coherence. To conquer these challenges, we propose MoSA, which decouples the process of human video generation into two components, i.e., structure generation and appearance generation. MoSA first employs a 3D structure transformer to generate a human motion sequence from the text prompt. The remaining video appearance is then synthesized under the guidance of this structural sequence. We achieve fine-grained control over the sparse human structures by introducing Human-Aware Dynamic Control modules with a dense tracking constraint during training. The modeling of human-environment interactions is improved through the proposed contact constraint. Those two components work comprehensively to ensure the structural and appearance fidelity across the generated videos. This paper also contributes a large-scale human video dataset, which features more complex and diverse motions than existing human video datasets. We conduct comprehensive comparisons between MoSA and a variety of approaches, including general video generation models, human video generation models, and human animation models. Experiments demonstrate that MoSA substantially outperforms existing approaches across the majority of evaluation metrics.

Method: Created PRISM benchmark to evaluate MLLMs on biometric privacy. Audited LLaVA datasets and found extensive biometric leakage. Developed Safe-LLaVA dataset by systematically removing explicit and implicit biometric information from LLaVA.

Result: Evaluations on PRISM revealed biometric leakages across MLLMs for different attributes. Models fine-tuned on Safe-LLaVA showed substantial reduction in biometric leakages while maintaining semantic faithfulness.

Conclusion: PRISM and Safe-LLaVA establish new standards for privacy-aligned development and evaluation of MLLMs, addressing critical biometric privacy concerns in multimodal AI systems.

Abstract: Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in vision-language tasks. However, these models often infer and reveal sensitive biometric attributes such as race, gender, age, body weight, and eye color; even when such information is not explicitly requested. This raises critical concerns, particularly in real-world applications and socially-sensitive domains. Despite increasing awareness, no publicly available dataset or benchmark exists to comprehensively evaluate or mitigate biometric leakage in MLLMs. To address this gap, we introduce PRISM (Privacy-aware Evaluation of Responses in Sensitive Modalities), a new benchmark designed to assess MLLMs on two fronts: (1) refuse biometric-related queries and (2) implicit biometric leakage in general responses while maintaining semantic faithfulness. Further, we conduct a detailed audit of the widely used LLaVA datasets and uncover extensive biometric leakage across pretraining and instruction data. To address this, we present Safe-LLaVA dataset, the first privacy-preserving MLLM training dataset constructed by systematically removing explicit and implicit biometric information from LLaVA dataset. Our evaluations on PRISM reveal biometric leakages across MLLMs for different attributes, highlighting the detailed privacy-violations. We also fine-tune a model on Safe-LLaVA dataset and show that it substantially reduces the biometric leakages. Together, Safe-LLaVA and PRISM set a new standard for privacy-aligned development and evaluation of MLLMs.

Method: Integration of pre-trained 2D CNNs (ResNet50, EfficientNet, Vision Transformers) with LSTM network enhanced by spatial attention mechanism to extract spatial features and capture temporal dependencies while emphasizing informative segments.

Result: Achieved peak accuracy of 93.34% with ResNet50-based configuration on UCF101 dataset subset, demonstrating superiority over several state-of-the-art HAR systems.

Conclusion: The proposed method offers a scalable and real-time-capable solution for fitness activity recognition with broader applications in vision-based health and activity monitoring.

Abstract: Fitness movement recognition, a focused subdomain of human activity recognition (HAR), plays a vital role in health monitoring, rehabilitation, and personalized fitness training by enabling automated exercise classification from video data. However, many existing deep learning approaches rely on computationally intensive 3D models, limiting their feasibility in real-time or resource-constrained settings. In this paper, we present a lightweight and effective framework that integrates pre-trained 2D Convolutional Neural Networks (CNNs) such as ResNet50, EfficientNet, and Vision Transformers (ViT) with a Long Short-Term Memory (LSTM) network enhanced by spatial attention. These models efficiently extract spatial features while the LSTM captures temporal dependencies, and the attention mechanism emphasizes informative segments. We evaluate the framework on a curated subset of the UCF101 dataset, achieving a peak accuracy of 93.34% with the ResNet50-based configuration. Comparative results demonstrate the superiority of our approach over several state-of-the-art HAR systems. The proposed method offers a scalable and real-time-capable solution for fitness activity recognition with broader applications in vision-based health and activity monitoring.

Method: Uses a pure decoder-only transformer with modality-specific Mixture-of-Experts (MoE) structure trained with single autoregressive objective, featuring multi-scale visual autoregressive mechanism within LLM to reduce decoding steps compared to diffusion methods.

Result: Sets new performance standard, outperforming existing open-source unified multimodal models across benchmarks for multimodal generation, editing, and understanding, with significant efficiency gains for high-resolution inputs.

Conclusion: Demonstrates that pure autoregressive modeling provides a sufficient and elegant foundation for unified multimodal intelligence, achieving state-of-the-art performance while improving efficiency.

Abstract: We introduce OneCAT, a unified multimodal model that seamlessly integrates understanding, generation, and editing within a novel, pure decoder-only transformer architecture. Our framework uniquely eliminates the need for external components such as Vision Transformers (ViT) or vision tokenizer during inference, leading to significant efficiency gains, especially for high-resolution inputs. This is achieved through a modality-specific Mixture-of-Experts (MoE) structure trained with a single autoregressive (AR) objective, which also natively supports dynamic resolutions. Furthermore, we pioneer a multi-scale visual autoregressive mechanism within the Large Language Model (LLM) that drastically reduces decoding steps compared to diffusion-based methods while maintaining state-of-the-art performance. Our findings demonstrate the powerful potential of pure autoregressive modeling as a sufficient and elegant foundation for unified multimodal intelligence. As a result, OneCAT sets a new performance standard, outperforming existing open-source unified multimodal models across benchmarks for multimodal generation, editing, and understanding.

Method: Created OpenFake dataset with 3 million real images paired with captions and 1 million synthetic images from state-of-the-art models, plus an innovative crowdsourced adversarial platform for continuous integration of hard examples.

Result: Detectors trained on OpenFake achieve near-perfect in-distribution performance, strong generalization to unseen generators, and high accuracy on real-world social media test sets, outperforming models trained on existing datasets.

Conclusion: With high-quality and continually updated benchmarks like OpenFake, automatic deepfake detection is both feasible and effective in real-world settings.

Abstract: Deepfakes, synthetic media created using advanced AI techniques, pose a growing threat to information integrity, particularly in politically sensitive contexts. This challenge is amplified by the increasing realism of modern generative models, which our human perception study confirms are often indistinguishable from real images. Yet, existing deepfake detection benchmarks rely on outdated generators or narrowly scoped datasets (e.g., single-face imagery), limiting their utility for real-world detection. To address these gaps, we present OpenFake, a large politically grounded dataset specifically crafted for benchmarking against modern generative models with high realism, and designed to remain extensible through an innovative crowdsourced adversarial platform that continually integrates new hard examples. OpenFake comprises nearly four million total images: three million real images paired with descriptive captions and almost one million synthetic counterparts from state-of-the-art proprietary and open-source models. Detectors trained on OpenFake achieve near-perfect in-distribution performance, strong generalization to unseen generators, and high accuracy on a curated in-the-wild social media test set, significantly outperforming models trained on existing datasets. Overall, we demonstrate that with high-quality and continually updated benchmarks, automatic deepfake detection is both feasible and effective in real-world settings.

Method: SCORE minimizes mutual information between target concepts and generated outputs using adversarial optimization, trajectory consistency, and saliency-driven fine-tuning, providing provable erasure guarantees with formal convergence proofs.

Result: SCORE outperforms state-of-the-art methods (EraseAnything, ANT, MACE, ESD, UCE) by up to 12.5% higher erasure efficacy while maintaining comparable or superior image quality across object erasure, NSFW removal, celebrity face suppression, and artistic style unlearning benchmarks.

Conclusion: SCORE sets a new standard for secure and robust concept erasure in diffusion models by integrating adversarial optimization with theoretical guarantees of statistical independence between erased concepts and model outputs.

Abstract: Diffusion models have achieved unprecedented success in image generation but pose increasing risks in terms of privacy, fairness, and security. A growing demand exists to \emph{erase} sensitive or harmful concepts (e.g., NSFW content, private individuals, artistic styles) from these models while preserving their overall generative capabilities. We introduce \textbf{SCORE} (Secure and Concept-Oriented Robust Erasure), a novel framework for robust concept removal in diffusion models. SCORE formulates concept erasure as an \emph{adversarial independence} problem, theoretically guaranteeing that the model’s outputs become statistically independent of the erased concept. Unlike prior heuristic methods, SCORE minimizes the mutual information between a target concept and generated outputs, yielding provable erasure guarantees. We provide formal proofs establishing convergence properties and derive upper bounds on residual concept leakage. Empirically, we evaluate SCORE on Stable Diffusion and FLUX across four challenging benchmarks: object erasure, NSFW removal, celebrity face suppression, and artistic style unlearning. SCORE consistently outperforms state-of-the-art methods including EraseAnything, ANT, MACE, ESD, and UCE, achieving up to \textbf{12.5%} higher erasure efficacy while maintaining comparable or superior image quality. By integrating adversarial optimization, trajectory consistency, and saliency-driven fine-tuning, SCORE sets a new standard for secure and robust concept erasure in diffusion models.

Method: Proposes a reinforcement learning approach using a Markov Decision Process that dynamically adjusts hyperparameters across denoising steps, with editing objectives integrated into a reward function and optimized via proximal policy optimization.

Result: Significant reduction in search time and computational overhead compared to brute-force approaches, while maintaining optimal hyperparameter configurations for effective image editing.

Conclusion: The method enables practical deployment of diffusion-based image editing frameworks by automating hyperparameter optimization, making the process more efficient and accessible for real-world applications.

Abstract: Recent advances in diffusion models have revolutionized text-guided image editing, yet existing editing methods face critical challenges in hyperparameter identification. To get the reasonable editing performance, these methods often require the user to brute-force tune multiple interdependent hyperparameters, such as inversion timesteps and attention modification. This process incurs high computational costs due to the huge hyperparameter search space. We consider searching optimal editing’s hyperparameters as a sequential decision-making task within the diffusion denoising process. Specifically, we propose a reinforcement learning framework, which establishes a Markov Decision Process that dynamically adjusts hyperparameters across denoising steps, integrating editing objectives into a reward function. The method achieves time efficiency through proximal policy optimization while maintaining optimal hyperparameter configurations. Experiments demonstrate significant reduction in search time and computational overhead compared to existing brute-force approaches, advancing the practical deployment of a diffusion-based image editing framework in the real world. Codes can be found at https://github.com/chaupham1709/AutoEdit.git.

Method: Two-stage framework: (1) Geometry removal using strictly mask-aligned supervision on depth/geometry data, (2) Appearance rendering to generate photorealistic RGB image from updated geometry. Uses preference-driven objective with positive/negative sample pairs to guide learning.

Result: Achieves state-of-the-art performance in removing both objects and their associated artifacts on two popular benchmarks, effectively eliminating causal visual effects like shadows and reflections.

Conclusion: The proposed geometry-aware approach successfully addresses limitations of appearance-based methods by leveraging causal relationships between object geometry and visual artifacts, enabling more complete and controllable object removal.

Abstract: Towards intelligent image editing, object removal should eliminate both the target object and its causal visual artifacts, such as shadows and reflections. However, existing image appearance-based methods either follow strictly mask-aligned training and fail to remove these causal effects which are not explicitly masked, or adopt loosely mask-aligned strategies that lack controllability and may unintentionally over-erase other objects. We identify that these limitations stem from ignoring the causal relationship between an object’s geometry presence and its visual effects. To address this limitation, we propose a geometry-aware two-stage framework that decouples object removal into (1) geometry removal and (2) appearance rendering. In the first stage, we remove the object directly from the geometry (e.g., depth) using strictly mask-aligned supervision, enabling structure-aware editing with strong geometric constraints. In the second stage, we render a photorealistic RGB image conditioned on the updated geometry, where causal visual effects are considered implicitly as a result of the modified 3D geometry. To guide learning in the geometry removal stage, we introduce a preference-driven objective based on positive and negative sample pairs, encouraging the model to remove objects as well as their causal visual artifacts while avoiding new structural insertions. Extensive experiments demonstrate that our method achieves state-of-the-art performance in removing both objects and their associated artifacts on two popular benchmarks. The code is available at https://github.com/buxiangzhiren/GeoRemover.

Method: Developed a domain-agnostic framework that systematically suppresses shape, texture, and color cues without forced-choice conflicts. Evaluated humans and neural networks under controlled suppression conditions across computer vision, medical imaging, and remote sensing domains.

Result: CNNs predominantly rely on local shape features, not texture. Modern training strategies and architectures (ConvNeXt, ViTs) can mitigate this reliance. Domain-specific patterns emerged: computer vision models prioritize shape, medical imaging models emphasize color, and remote sensing models show stronger texture reliance.

Conclusion: The texture-bias hypothesis for CNNs is incorrect - they are primarily shape-biased. Feature reliance patterns are domain-dependent, suggesting the need for domain-aware model design and evaluation.

Abstract: The hypothesis that Convolutional Neural Networks (CNNs) are inherently texture-biased has shaped much of the discourse on feature use in deep learning. We revisit this hypothesis by examining limitations in the cue-conflict experiment by Geirhos et al. To address these limitations, we propose a domain-agnostic framework that quantifies feature reliance through systematic suppression of shape, texture, and color cues, avoiding the confounds of forced-choice conflicts. By evaluating humans and neural networks under controlled suppression conditions, we find that CNNs are not inherently texture-biased but predominantly rely on local shape features. Nonetheless, this reliance can be substantially mitigated through modern training strategies or architectures (ConvNeXt, ViTs). We further extend the analysis across computer vision, medical imaging, and remote sensing, revealing that reliance patterns differ systematically: computer vision models prioritize shape, medical imaging models emphasize color, and remote sensing models exhibit a stronger reliance on texture. Code is available at https://github.com/tomburgert/feature-reliance.

Method: Augmented Group Relative Policy Optimization (GRPO) with curriculum-based data scheduling and difficulty-aware filtering to stabilize optimization under sparse, noisy rewards.

Result: Substantial improvements in detection accuracy and robustness on autonomous driving benchmarks; ablation studies confirm importance of reward design, KL regularization, and curriculum pacing.

Conclusion: Reinforcement-driven optimization with structured data curricula provides a scalable path toward robust and interpretable multimodal detection.

Abstract: Multimodal Large Language Models (MLLMs) excel in vision-language reasoning but often struggle with structured perception tasks requiring precise localization and robustness. We propose a reinforcement learning framework that augments Group Relative Policy Optimization (GRPO) with curriculum-based data scheduling and difficulty-aware filtering. This approach stabilizes optimization under sparse, noisy rewards and enables progressive adaptation to complex samples. Evaluations on autonomous driving benchmarks demonstrate substantial improvements in detection accuracy and robustness. Ablation studies confirm the importance of reward design, KL regularization, and curriculum pacing for convergence stability and generalization. Our findings highlight reinforcement-driven optimization with structured data curricula as a scalable path toward robust and interpretable multimodal detection.

Method: ExGS unifies UGC for re-optimization-free pruning to reduce Gaussian primitives and GaussPainter that uses diffusion priors with mask-guided refinement to restore quality from heavily pruned scenes, featuring lightweight VAE and one-step diffusion for real-time performance.

Result: Achieves over 100X compression (reducing 354.77 MB to 3.31 MB) while preserving fidelity and significantly improving image quality under challenging conditions.

Conclusion: Diffusion priors play a central role in bridging extreme compression and high-quality neural rendering, enabling practical deployment of compressed 3DGS models.

Abstract: Neural scene representations, such as 3D Gaussian Splatting (3DGS), have enabled high-quality neural rendering; however, their large storage and transmission costs hinder deployment in resource-constrained environments. Existing compression methods either rely on costly optimization, which is slow and scene-specific, or adopt training-free pruning and quantization, which degrade rendering quality under high compression ratios. In contrast, recent data-driven approaches provide a promising direction to overcome this trade-off, enabling efficient compression while preserving high rendering quality. We introduce ExGS, a novel feed-forward framework that unifies Universal Gaussian Compression (UGC) with GaussPainter for Extreme 3DGS compression. UGC performs re-optimization-free pruning to aggressively reduce Gaussian primitives while retaining only essential information, whereas GaussPainter leverages powerful diffusion priors with mask-guided refinement to restore high-quality renderings from heavily pruned Gaussian scenes. Unlike conventional inpainting, GaussPainter not only fills in missing regions but also enhances visible pixels, yielding substantial improvements in degraded renderings. To ensure practicality, it adopts a lightweight VAE and a one-step diffusion design, enabling real-time restoration. Our framework can even achieve over 100X compression (reducing a typical 354.77 MB model to about 3.31 MB) while preserving fidelity and significantly improving image quality under challenging conditions. These results highlight the central role of diffusion priors in bridging the gap between extreme compression and high-quality neural rendering. Our code repository will be released at: https://github.com/chenttt2001/ExGS

Method: Three key components: 1) Hybrid-Loss Depth Estimation for multi-view consistency, 2) Bidirectional Warping Virtual View Synthesis for stronger constraints, and 3) Occlusion-Aware Reconstruction using depth-difference masks and learning-based inpainting.

Result: Achieves up to 21.13 dB PSNR and 0.189 LPIPS on LLFF, Blender, and DTU benchmarks, setting new state-of-the-art while maintaining real-time inference.

Conclusion: HBSplat successfully addresses sparse view synthesis challenges in 3DGS through unified integration of structural cues, virtual views, and occlusion completion, demonstrating significant performance improvements.

Abstract: Novel View Synthesis (NVS) from sparse views presents a formidable challenge in 3D reconstruction, where limited multi-view constraints lead to severe overfitting, geometric distortion, and fragmented scenes. While 3D Gaussian Splatting (3DGS) delivers real-time, high-fidelity rendering, its performance drastically deteriorates under sparse inputs, plagued by floating artifacts and structural failures. To address these challenges, we introduce HBSplat, a unified framework that elevates 3DGS by seamlessly integrating robust structural cues, virtual view constraints, and occluded region completion. Our core contributions are threefold: a Hybrid-Loss Depth Estimation module that ensures multi-view consistency by leveraging dense matching priors and integrating reprojection, point propagation, and smoothness constraints; a Bidirectional Warping Virtual View Synthesis method that enforces substantially stronger constraints by creating high-fidelity virtual views through bidirectional depth-image warping and multi-view fusion; and an Occlusion-Aware Reconstruction component that recovers occluded areas using a depth-difference mask and a learning-based inpainting model. Extensive evaluations on LLFF, Blender, and DTU benchmarks validate that HBSplat sets a new state-of-the-art, achieving up to 21.13 dB PSNR and 0.189 LPIPS, while maintaining real-time inference. Code is available at: https://github.com/eternalland/HBSplat.

Method: EvoQuality adapts self-consistency to IQA by generating pseudo-labels through pairwise majority voting on the VLM’s outputs to establish consensus on relative quality. These pseudo-rankings are formulated into a fidelity reward that guides iterative evolution using group relative policy optimization (GRPO).

Result: EvoQuality boosts the base VLM’s zero-shot performance by 31.8% on PLCC across diverse IQA benchmarks. Despite being entirely self-supervised, it achieves performance competitive with or superior to state-of-the-art supervised VLM-based IQA models, outperforming them on 5 out of 7 benchmarks.

Conclusion: EvoQuality demonstrates that VLMs can autonomously refine their perceptual capabilities through self-supervised learning, achieving competitive performance without requiring expensive human annotations, making it a promising approach for enhancing vision-language models in perceptual tasks.

Abstract: Improving vision-language models (VLMs) in the post-training stage typically relies on supervised fine-tuning or reinforcement learning, methods that necessitate costly, human-annotated data. While self-supervised techniques such as self-consistency have proven effective for enhancing reasoning capabilities, their application to perceptual domains such as image quality assessment (IQA) remains largely unexplored. In this work, we introduce EvoQuality, a novel framework that enables a VLM to autonomously refine its quality perception capabilities without any ground-truth labels. EvoQuality adapts the principle of self-consistency to the ranking-based nature of IQA. It generates pseudo-labels by performing pairwise majority voting on the VLM’s own outputs to establish a consensus on relative quality. These pseudo-rankings are then formulated into a fidelity reward that guides the model’s iterative evolution through group relative policy optimization (GRPO). By iteratively leveraging its own predictions, EvoQuality progressively refines the VLM’s perceptual capability. Extensive experiments show that EvoQuality boosts the base VLM’s zero-shot performance by 31.8% on PLCC across diverse IQA benchmarks. Remarkably, despite being entirely self-supervised, EvoQuality achieves performance that is competitive with, or even surpasses, state-of-the-art supervised VLM-based IQA models, outperforming these models on 5 out of 7 IQA benchmarks.

Method: Introduces CAT framework that uses cluster-specific affine parameters instead of uniform ones. It refines low-bit quantized outputs with minimal additional parameters, integrated into a novel PTQ framework without requiring model or quantization parameter fine-tuning.

Result: Achieves up to 53.18% Top-1 accuracy on W2A2 ResNet-18 on ImageNet-1K, consistently outperforming prior PTQ methods across diverse architectures and low-bit settings. When used as plug-in, enhances existing PTQ baselines by more than 3%.

Conclusion: CAT effectively addresses accuracy degradation in low-bit PTQ through cluster-specific affine transformation, providing significant improvements without computational overhead of fine-tuning, making it a practical solution for efficient model deployment.

Abstract: Post-Training Quantization (PTQ) reduces the memory footprint and computational overhead of deep neural networks by converting full-precision (FP) values into quantized and compressed data types. While PTQ is more cost-efficient than Quantization-Aware Training (QAT), it is highly susceptible to accuracy degradation under a low-bit quantization (LQ) regime (e.g., 2-bit). Affine transformation is a classical technique used to reduce the discrepancy between the information processed by a quantized model and that processed by its full-precision counterpart; however, we find that using plain affine transformation, which applies a uniform affine parameter set for all outputs, worsens the results in low-bit PTQ. To address this, we propose Cluster-based Affine Transformation (CAT), an error-reduction framework that employs cluster-specific parameters to align LQ outputs with FP counterparts. CAT refines LQ outputs with only a negligible number of additional parameters, without requiring fine-tuning of the model or quantization parameters. We further introduce a novel PTQ framework integrated with CAT. Experiments on ImageNet-1K show that this framework consistently outperforms prior PTQ methods across diverse architectures and LQ settings, achieving up to 53.18% Top-1 accuracy on W2A2 ResNet-18. Moreover, CAT enhances existing PTQ baselines by more than 3% when used as a plug-in. We plan to release our implementation alongside the publication of this paper.

Method: CPG uses a controllable self-reinforcing optimization cycle: (1) dynamic controllable filtering to selectively add pseudo-labels ensuring known distribution, (2) Bayes-optimal classifier with logit adjustment, (3) improved classifier helps identify more pseudo-labels. Also includes class-aware adaptive augmentation and auxiliary branch for data utilization.

Result: CPG achieves consistent improvements across benchmark datasets, surpassing state-of-the-art methods by up to 15.97% in accuracy.

Conclusion: The proposed CPG framework effectively handles unknown unlabeled data distributions in long-tailed semi-supervised learning through its controllable pseudo-label generation and optimization cycle, with theoretical guarantees and practical performance gains.

Abstract: Current long-tailed semi-supervised learning methods assume that labeled data exhibit a long-tailed distribution, and unlabeled data adhere to a typical predefined distribution (i.e., long-tailed, uniform, or inverse long-tailed). However, the distribution of the unlabeled data is generally unknown and may follow an arbitrary distribution. To tackle this challenge, we propose a Controllable Pseudo-label Generation (CPG) framework, expanding the labeled dataset with the progressively identified reliable pseudo-labels from the unlabeled dataset and training the model on the updated labeled dataset with a known distribution, making it unaffected by the unlabeled data distribution. Specifically, CPG operates through a controllable self-reinforcing optimization cycle: (i) at each training step, our dynamic controllable filtering mechanism selectively incorporates reliable pseudo-labels from the unlabeled dataset into the labeled dataset, ensuring that the updated labeled dataset follows a known distribution; (ii) we then construct a Bayes-optimal classifier using logit adjustment based on the updated labeled data distribution; (iii) this improved classifier subsequently helps identify more reliable pseudo-labels in the next training step. We further theoretically prove that this optimization cycle can significantly reduce the generalization error under some conditions. Additionally, we propose a class-aware adaptive augmentation module to further improve the representation of minority classes, and an auxiliary branch to maximize data utilization by leveraging all labeled and unlabeled samples. Comprehensive evaluations on various commonly used benchmark datasets show that CPG achieves consistent improvements, surpassing state-of-the-art methods by up to $\textbf{15.97%}$ in accuracy. The code is available at https://github.com/yaxinhou/CPG.

Method: Two-stage approach: 1) Localize question-relevant intervals using low-fps skim, 2) Answer via span-aware reallocation of visual tokens at higher effective frame rate. Uses interleaved group-relative objective coupling temporal IoU for localization with answer correctness.

Result: Achieves up to 8.6% improvement with 50% less frame input on long-video QA and temporal grounding tasks (Charades-STA, ActivityNet-Captions). Span-aware token reallocation consistently outperforms uniform sampling.

Conclusion: ViTL and the new dataset provide an interpretable, compute-efficient solution for scalable long-video QA with direct attribution capabilities.

Abstract: We present \emph{Video-in-the-Loop} (ViTL), a two-stage long-video QA framework that preserves a fixed token budget by first \emph{localizing} question-relevant interval(s) with a low-fps skim and then \emph{answering} via span-aware reallocation of visual tokens at higher effective frame rate, emitting an interleaved output with both spans and the final option for direct attribution. We also introduce \dataname{}, which converts description based event graphs into \emph{span-grounded} multiple-choice QA by pairing each question with \emph{ground-truth} time span(s) and related reasoning. ViTL is trained end-to-end with an interleaved group-relative objective that couples temporal IoU for localization with answer correctness, allowing credit to flow from answers back to spans without increasing compute. Under fixed token budgets, ViTL attains up to 8.6% with 50% less frame input on long-video QA and temporal grounding (e.g., Charades-STA, ActivityNet-Captions) and ablations show that span-aware token reallocation consistently surpasses uniform sampling. Together, \dataname{} and ViTL provide an interpretable, compute-efficient recipe for scalable long-video QA.

Method: MoME processes walking cycles in four stages of movement complexity using lightweight expert models and task-specific gating modules to adaptively weight experts across traits and stages.

Result: Outperforms state-of-the-art gait analysis models with 37.47% weighted F1 score at run level and 44.6% at subject level on PsyMo benchmark. Integrating auxiliary tasks further improves performance.

Conclusion: Demonstrates viability of multi-task gait-based learning for psychological trait estimation and provides foundation for movement-informed psychological inference.

Abstract: Gait encodes rich biometric and behavioural information, yet leveraging the manner of walking to infer psychological traits remains a challenging and underexplored problem. We introduce a hierarchical Multi-Stage Mixture of Movement Experts (MoME) architecture for multi-task prediction of psychological attributes from gait sequences represented as 2D poses. MoME processes the walking cycle in four stages of movement complexity, employing lightweight expert models to extract spatio-temporal features and task-specific gating modules to adaptively weight experts across traits and stages. Evaluated on the PsyMo benchmark covering 17 psychological traits, our method outperforms state-of-the-art gait analysis models, achieving a 37.47% weighted F1 score at the run level and 44.6% at the subject level. Our experiments show that integrating auxiliary tasks such as identity recognition, gender prediction, and BMI estimation further improves psychological trait estimation. Our findings demonstrate the viability of multi-task gait-based learning for psychological trait estimation and provide a foundation for future research on movement-informed psychological inference.

Method: Comprehensive information-theoretic analysis of rule encodings, formal analysis of multiple attention architectures (causal, bidirectional, local sparse, kernelized, cross-attention), and dynamic rule verification with hot reloading.

Result: Rule formats with low syntactic entropy and concentrated anchors reduce attention entropy and improve pointer fidelity, but reveal a fundamental trade-off between anchor redundancy and attention entropy.

Conclusion: Principled anchor design and dual enforcement mechanisms are necessary to protect LLM-based agents against prompt injection attacks while maintaining compliance in evolving domains.

Abstract: The design of safety-critical agents based on large language models (LLMs) requires more than simple prompt engineering. This paper presents a comprehensive information-theoretic analysis of how rule encodings in system prompts influence attention mechanisms and compliance behaviour. We demonstrate that rule formats with low syntactic entropy and highly concentrated anchors reduce attention entropy and improve pointer fidelity, but reveal a fundamental trade-off between anchor redundancy and attention entropy that previous work failed to recognize. Through formal analysis of multiple attention architectures including causal, bidirectional, local sparse, kernelized, and cross-attention mechanisms, we establish bounds on pointer fidelity and show how anchor placement strategies must account for competing fidelity and entropy objectives. Combining these insights with a dynamic rule verification architecture, we provide a formal proof that hot reloading of verified rule sets increases the asymptotic probability of compliant outputs. These findings underscore the necessity of principled anchor design and dual enforcement mechanisms to protect LLM-based agents against prompt injection attacks while maintaining compliance in evolving domains.

Method: SCL dedicates the language model to inference only, maintains memory externally, and uses a lightweight controller within a goal-directed loop. This allows intermediate results to be stored, revisited, and checked before actions.

Result: SCL achieved 86.3% task success vs 70-77% for baselines, with higher goal fidelity, fewer redundant calls, better state reuse, and reduced unsupported assertions. Ablations show external memory and control each contribute independently.

Conclusion: Architectural separation can improve reliability and traceability without larger models or heavier prompts. Findings are preliminary and should guide extended studies with additional models and more complex tasks.

Abstract: Large language models have advanced natural language understanding and generation, yet their use as autonomous agents raises architectural challenges for multi-step tasks. Existing frameworks often intertwine inference, memory, and control in a single prompt, which can reduce coherence and predictability. The Structured Cognitive Loop (SCL) is introduced as an alternative architecture that separates these functions. In SCL, the language model is dedicated to inference, memory is maintained externally, and execution is guided by a lightweight controller within a goal-directed loop. This design offloads cognitive load from the model and allows intermediate results to be stored, revisited, and checked before actions are taken, providing a clearer basis for traceability and evaluation. We evaluate SCL against prompt-based baselines including ReAct and common LangChain agents across three scenarios: temperature-based travel planning, email drafting with conditional send, and constraint-guided image generation. All systems share the same base model and tools under matched decoding settings. Across 360 episodes, SCL shows modest but consistent improvements. Task success averages 86.3 percent compared with 70-77 percent for baselines. Goal fidelity is higher, redundant calls are fewer, intermediate states are reused more reliably, and unsupported assertions per 100 tool calls are reduced. Ablations show that external memory and control each contribute independently, and decoding sweeps confirm stability of the effects. These results suggest that architectural separation can improve reliability and traceability without relying on larger models or heavier prompts. The findings are preliminary and intended to guide extended studies with additional models, longer horizons, multimodal tasks, and collaborative settings.

Method: Lang-PINN coordinates four specialized agents: PDE Agent (parses natural language to symbolic PDEs), PINN Agent (selects architectures), Code Agent (generates modular implementations), and Feedback Agent (executes and diagnoses errors for iterative refinement).

Result: Lang-PINN achieves substantially lower errors (MSE reduced by 3-5 orders of magnitude), improves end-to-end execution success by over 50%, and reduces time overhead by up to 74% compared to baselines.

Conclusion: The multi-agent LLM system successfully transforms informal task statements into executable and verifiable PINN code, demonstrating robust automation for scientific computing workflows.

Abstract: Physics-informed neural networks (PINNs) provide a powerful approach for solving partial differential equations (PDEs), but constructing a usable PINN remains labor-intensive and error-prone. Scientists must interpret problems as PDE formulations, design architectures and loss functions, and implement stable training pipelines. Existing large language model (LLM) based approaches address isolated steps such as code generation or architecture suggestion, but typically assume a formal PDE is already specified and therefore lack an end-to-end perspective. We present Lang-PINN, an LLM-driven multi-agent system that builds trainable PINNs directly from natural language task descriptions. Lang-PINN coordinates four complementary agents: a PDE Agent that parses task descriptions into symbolic PDEs, a PINN Agent that selects architectures, a Code Agent that generates modular implementations, and a Feedback Agent that executes and diagnoses errors for iterative refinement. This design transforms informal task statements into executable and verifiable PINN code. Experiments show that Lang-PINN achieves substantially lower errors and greater robustness than competitive baselines: mean squared error (MSE) is reduced by up to 3–5 orders of magnitude, end-to-end execution success improves by more than 50%, and reduces time overhead by up to 74%.

Method: SAC-Opt uses backward-guided correction that grounds optimization modeling in problem semantics rather than solver feedback. It aligns original semantic anchors with reconstructed ones from generated code and selectively corrects mismatched components through fine-grained refinement of constraint and objective logic.

Result: Empirical results on seven datasets show SAC-Opt improves average modeling accuracy by 7.8%, with gains up to 21.9% on the ComplexLP dataset, enhancing both fidelity and robustness without requiring additional training or supervision.

Conclusion: Semantic-anchored correction is crucial in LLM-based optimization workflows to ensure faithful translation from problem intent to solver-executable code, addressing the limitations of solver-driven approaches.

Abstract: Large language models (LLMs) have opened new paradigms in optimization modeling by enabling the generation of executable solver code from natural language descriptions. Despite this promise, existing approaches typically remain solver-driven: they rely on single-pass forward generation and apply limited post-hoc fixes based on solver error messages, leaving undetected semantic errors that silently produce syntactically correct but logically flawed models. To address this challenge, we propose SAC-Opt, a backward-guided correction framework that grounds optimization modeling in problem semantics rather than solver feedback. At each step, SAC-Opt aligns the original semantic anchors with those reconstructed from the generated code and selectively corrects only the mismatched components, driving convergence toward a semantically faithful model. This anchor-driven correction enables fine-grained refinement of constraint and objective logic, enhancing both fidelity and robustness without requiring additional training or supervision. Empirical results on seven public datasets demonstrate that SAC-Opt improves average modeling accuracy by 7.8%, with gains of up to 21.9% on the ComplexLP dataset. These findings highlight the importance of semantic-anchored correction in LLM-based optimization workflows to ensure faithful translation from problem intent to solver-executable code.

Method: AgentFlow coordinates planner, executor, verifier, and generator modules through evolving memory. Uses Flow-GRPO for on-policy training in live environments, converting multi-turn optimization into tractable single-turn updates with group-normalized advantages.

Result: Outperforms top-performing baselines with average accuracy gains of 14.9% on search, 14.0% on agentic, 14.5% on mathematical, and 4.1% on scientific tasks, surpassing larger proprietary models like GPT-4o.

Conclusion: AgentFlow demonstrates benefits of in-the-flow optimization with improved planning, enhanced tool-calling reliability, and positive scaling with model size and reasoning turns.

Abstract: Outcome-driven reinforcement learning has advanced reasoning in large language models (LLMs), but prevailing tool-augmented approaches train a single, monolithic policy that interleaves thoughts and tool calls under full context; this scales poorly with long horizons and diverse tools and generalizes weakly to new scenarios. Agentic systems offer a promising alternative by decomposing work across specialized modules, yet most remain training-free or rely on offline training decoupled from the live dynamics of multi-turn interaction. We introduce AgentFlow, a trainable, in-the-flow agentic framework that coordinates four modules (planner, executor, verifier, generator) through an evolving memory and directly optimizes its planner inside the multi-turn loop. To train on-policy in live environments, we propose Flow-based Group Refined Policy Optimization (Flow-GRPO), which tackles long-horizon, sparse-reward credit assignment by converting multi-turn optimization into a sequence of tractable single-turn policy updates. It broadcasts a single, verifiable trajectory-level outcome to every turn to align local planner decisions with global success and stabilizes learning with group-normalized advantages. Across ten benchmarks, AgentFlow with a 7B-scale backbone outperforms top-performing baselines with average accuracy gains of 14.9% on search, 14.0% on agentic, 14.5% on mathematical, and 4.1% on scientific tasks, even surpassing larger proprietary models like GPT-4o. Further analyses confirm the benefits of in-the-flow optimization, showing improved planning, enhanced tool-calling reliability, and positive scaling with model size and reasoning turns.

Method: Dynamic Adjudication Template (DAT) - a three-stage framework: 1) Qualitative analysis for contextual evaluation, 2) Evidence gathering with targeted information extraction using template placeholders and systematic rule verification, 3) Adjudication phase for synthesizing validated components into comprehensive judgments.

Result: DAT consistently outperforms conventional CoT approaches in complex rule-based tasks. It enables smaller language models to match or exceed the performance of significantly larger LLMs, demonstrating efficiency and effectiveness in managing intricate rule systems.

Conclusion: DAT provides a systematic methodology for structured rule processing that addresses the limitations of existing approaches, offering improved reasoning capabilities for complex rule systems while being computationally efficient.

Abstract: Large language models (LLMs) face significant challenges when processing complex rule systems, as they typically treat interdependent rules as unstructured textual data rather than as logically organized frameworks. This limitation results in reasoning divergence, where models often overlook critical rule dependencies essential for accurate interpretation. Although existing approaches such as Chain-of-Thought (CoT) reasoning have shown promise, they lack systematic methodologies for structured rule processing and are particularly susceptible to error propagation through sequential reasoning chains. To address these limitations, we propose the Dynamic Adjudication Template (DAT), a novel framework inspired by expert human reasoning processes. DAT structures the inference mechanism into three methodical stages: qualitative analysis, evidence gathering, and adjudication. During the qualitative analysis phase, the model comprehensively evaluates the contextual landscape. The subsequent evidence gathering phase involves the targeted extraction of pertinent information based on predefined template elements ([placeholder]), followed by systematic verification against applicable rules. Finally, in the adjudication phase, the model synthesizes these validated components to formulate a comprehensive judgment. Empirical results demonstrate that DAT consistently outperforms conventional CoT approaches in complex rule-based tasks. Notably, DAT enables smaller language models to match, and in some cases exceed, the performance of significantly larger LLMs, highlighting its efficiency and effectiveness in managing intricate rule systems.

Method: Model symbolic systems as universal Turing machines and define grounding as information compression. The analysis proceeds in four stages: proving symbolic systems cannot ground random worlds, showing static grounding is incomplete, demonstrating grounding acts are non-inferable, and using Chaitin’s Incompleteness Theorem.

Result: Established that purely symbolic systems cannot ground algorithmically random worlds, static grounding is inherently incomplete, grounding acts require new information input, and algorithmic learning processes cannot comprehend worlds exceeding their complexity.

Conclusion: Meaning is an open-ended process where systems perpetually attempt to overcome their own information-theoretic limitations, rather than a static achievement.

Abstract: This paper provides a definitive, unifying framework for the Symbol Grounding Problem (SGP) by reformulating it within Algorithmic Information Theory (AIT). We demonstrate that the grounding of meaning is a process fundamentally constrained by information-theoretic limits, thereby unifying the G"odelian (self-reference) and No Free Lunch (statistical) perspectives. We model a symbolic system as a universal Turing machine and define grounding as an act of information compression. The argument proceeds in four stages. First, we prove that a purely symbolic system cannot ground almost all possible “worlds” (data strings), as they are algorithmically random and thus incompressible. Second, we show that any statically grounded system, specialized for compressing a specific world, is inherently incomplete because an adversarial, incompressible world relative to the system can always be constructed. Third, the “grounding act” of adapting to a new world is proven to be non-inferable, as it requires the input of new information (a shorter program) that cannot be deduced from the system’s existing code. Finally, we use Chaitin’s Incompleteness Theorem to prove that any algorithmic learning process is itself a finite system that cannot comprehend or model worlds whose complexity provably exceeds its own. This establishes that meaning is the open-ended process of a system perpetually attempting to overcome its own information-theoretic limitations.

Method: Survey methodology: reviewing representative foundation models and key alignment metrics, synthesizing empirical evidence from studies in vision, language, speech, multimodality, and neuroscience.

Result: Evidence suggests foundation models exhibit structural regularities and semantic consistencies in representation spaces, making them strong candidates for cross-modal transfer and alignment.

Conclusion: Foundation models show promising representation potentials for cross-modal applications, though key factors, open questions, and challenges need further analysis.

Abstract: Foundation models learn highly transferable representations through large-scale pretraining on diverse data. An increasing body of research indicates that these representations exhibit a remarkable degree of similarity across architectures and modalities. In this survey, we investigate the representation potentials of foundation models, defined as the latent capacity of their learned representations to capture task-specific information within a single modality while also providing a transferable basis for alignment and unification across modalities. We begin by reviewing representative foundation models and the key metrics that make alignment measurable. We then synthesize empirical evidence of representation potentials from studies in vision, language, speech, multimodality, and neuroscience. The evidence suggests that foundation models often exhibit structural regularities and semantic consistencies in their representation spaces, positioning them as strong candidates for cross-modal transfer and alignment. We further analyze the key factors that foster representation potentials, discuss open questions, and highlight potential challenges.

Method: Six-layer framework: perception, semantic annotation, interoperability, transportation, semantic reasoning, and application layers with semantic algorithms, metadata processing, and reasoning methods.

Result: Framework enables semantic completeness, real-time knowledge inference, and robust IoT data management through uncertainty reasoning and semantic interoperability.

Conclusion: The framework provides a valuable tool for advancing IoT applications, particularly in agriculture, by integrating semantic layers and reasoning methods.

Abstract: The Internet of Things (IoT) has revolutionized various applications including agriculture, but it still faces challenges in data collection and understanding. This paper proposes a real-time framework with three additional semantic layers to help IoT devices and sensors comprehend data meaning and source. The framework consists of six layers: perception, semantic annotation, interoperability, transportation, semantic reasoning, and application, suitable for dynamic environments. Sensors collect data in the form of voltage, which is then processed by microprocessors or microcontrollers in the semantic annotation and preprocessing layer. Metadata is added to the raw data, including the purpose, ID number, and application. Two semantic algorithms are proposed in the semantic interoperability and ontologies layer: the interoperability semantic algorithm for standardizing file types and the synonym identification algorithm for identifying synonyms. In the transportation layer, raw data and metadata are sent to other IoT devices or cloud computing platforms using techniques like WiFi, Zigbee networks, Bluetooth, and mobile communication networks. A semantic reasoning layer is proposed to infer new knowledge from the existing data, using fuzzy logic, Dempster-Shafer theory, and Bayesian networks. A Graphical User Interface (GUI) is proposed in the application layer to help users communicate with and monitor IoT sensors, devices, and new knowledge inferred. This framework provides a robust solution for managing IoT data, ensuring semantic completeness, and enabling real-time knowledge inference. The integration of uncertainty reasoning methods and semantic interoperability techniques makes this framework a valuable tool for advancing IoT applications in general and in agriculture in particular.

Method: Uses programmatically generated charts and synthetically rendered tables created with Python visualization libraries, paired with systematically generated multi-step analytical questions based on visual content. Answers are provided in structured formats (JSON/YAML) with a taxonomy of reasoning types including comparison, trend identification, ranking, aggregation, proportion estimation, and anomaly detection.

Result: GRAFT establishes a comprehensive benchmark with controlled data semantics, structure, and clarity, enabling precise evaluation of both reasoning capabilities and output format compliance through strict factual and formatting guidelines.

Conclusion: GRAFT sets a new evaluation standard for multimodal models by providing a unified, scalable framework for fine-grained benchmarking of visually grounded, structured reasoning tasks.

Abstract: GRAFT is a structured multimodal benchmark for evaluating models on instruction-following, visual reasoning, and visual-textual alignment tasks. It features programmatically generated charts and synthetically rendered tables, created with Python visualization libraries to ensure control over data semantics, structure, and clarity. Each GRAFT instance pairs a chart or table image with a systematically generated, multi-step analytical question based solely on visual content. Answers are provided in structured formats such as JSON or YAML, supporting consistent evaluation of both reasoning and output format. The benchmark introduces a taxonomy of reasoning types including comparison, trend identification, ranking, aggregation, proportion estimation, and anomaly detection to enable comprehensive assessment. Reference answers follow strict factual and formatting guidelines for precise, aspect-based evaluation. GRAFT offers a unified, scalable framework for fine-grained benchmarking of multimodal models on visually grounded, structured reasoning tasks, setting a new evaluation standard in this field.

Method: RareAgent organizes task-specific adversarial debates where agents dynamically construct evidence graphs from diverse perspectives to support, refute, or entail hypotheses. It uses a self-evolutionary loop to analyze reasoning strategies and refine agent policies, distilling successful paths into transferable heuristics.

Result: RareAgent improves indication AUPRC by 18.1% over reasoning baselines and provides transparent reasoning chains consistent with clinical evidence.

Conclusion: The approach successfully transforms rare disease drug repurposing from passive pattern recognition to active evidence-seeking reasoning, achieving significant performance improvements while maintaining interpretability.

Abstract: Computational drug repurposing for rare diseases is especially challenging when no prior associations exist between drugs and target diseases. Therefore, knowledge graph completion and message-passing GNNs have little reliable signal to learn and propagate, resulting in poor performance. We present RareAgent, a self-evolving multi-agent system that reframes this task from passive pattern recognition to active evidence-seeking reasoning. RareAgent organizes task-specific adversarial debates in which agents dynamically construct evidence graphs from diverse perspectives to support, refute, or entail hypotheses. The reasoning strategies are analyzed post hoc in a self-evolutionary loop, producing textual feedback that refines agent policies, while successful reasoning paths are distilled into transferable heuristics to accelerate future investigations. Comprehensive evaluations reveal that RareAgent improves the indication AUPRC by 18.1% over reasoning baselines and provides a transparent reasoning chain consistent with clinical evidence.

Method: A divide-and-conquer approach using hierarchical LLM agents with three stages: Global Review for storyline and structure, Scene-level Review for detailed flaws, and Hierarchical Coordinated Revision to integrate improvements.

Result: Significantly outperforms all baselines in script-level overall quality and scene-level details, with plug-and-play capability for existing methods.

Conclusion: The hierarchical multi-agent approach effectively addresses the challenge of revising long narratives by coordinating global and local improvements through iterative refinement.

Abstract: Although LLMs have been widely adopted for creative content generation, a single-pass process often struggles to produce high-quality long narratives. How to effectively revise and improve long narrative scripts like scriptwriters remains a significant challenge, as it demands a comprehensive understanding of the entire context to identify global structural issues and local detailed flaws, as well as coordinating revisions at multiple granularities and locations. Direct modifications by LLMs typically introduce inconsistencies between local edits and the overall narrative requirements. To address these issues, we propose Dramaturge, a task and feature oriented divide-and-conquer approach powered by hierarchical multiple LLM agents. It consists of a Global Review stage to grasp the overall storyline and structural issues, a Scene-level Review stage to pinpoint detailed scene and sentence flaws, and a Hierarchical Coordinated Revision stage that coordinates and integrates structural and detailed improvements throughout the script. The top-down task flow ensures that high-level strategies guide local modifications, maintaining contextual consistency. The review and revision workflow follows a coarse-to-fine iterative process, continuing through multiple rounds until no further substantive improvements can be made. Comprehensive experiments show that Dramaturge significantly outperforms all baselines in terms of script-level overall quality and scene-level details. Our approach is plug-and-play and can be easily integrated into existing methods to improve the generated scripts.

Method: A weakly supervised pipeline that integrates large language models with structured demographic attributes and unstructured public feedback, dynamically modeling citizen needs into a need-aware graph based on features like age, gender, and deprivation index.

Result: Preliminary experimental results using real-world data demonstrate the feasibility of the approach, generating interpretable insights for health policy decision-making.

Conclusion: The framework offers a scalable solution for intelligent population health monitoring in resource-constrained clinical and governmental settings.

Abstract: Timely and accurate analysis of population-level data is crucial for effective decision-making during public health emergencies such as the COVID-19 pandemic. However, the massive input of semi-structured data, including structured demographic information and unstructured human feedback, poses significant challenges to conventional analysis methods. Manual expert-driven assessments, though accurate, are inefficient, while standard NLP pipelines often require large task-specific labeled datasets and struggle with generalization across diverse domains. To address these challenges, we propose a novel graph-based reasoning framework that integrates large language models with structured demographic attributes and unstructured public feedback in a weakly supervised pipeline. The proposed approach dynamically models evolving citizen needs into a need-aware graph, enabling population-specific analyses based on key features such as age, gender, and the Index of Multiple Deprivation. It generates interpretable insights to inform responsive health policy decision-making. We test our method using a real-world dataset, and preliminary experimental results demonstrate its feasibility. This approach offers a scalable solution for intelligent population health monitoring in resource-constrained clinical and governmental settings.

Method: Uses beta-binomial distribution for robust estimation and introduces dynamic sampling strategy that allocates more budget to harder problems.

Result: The proposed innovations enable more reliable prediction of rare risks and capabilities at a fraction of the computational cost.

Conclusion: The framework provides improved methods for capability and safety forecasting of frontier AI systems when scaling to massive usage scenarios.

Abstract: Assessing the capabilities and risks of frontier AI systems is a critical area of research, and recent work has shown that repeated sampling from models can dramatically increase both. For instance, repeated sampling has been shown to increase their capabilities, such as solving difficult math and coding problems, but it has also been shown to increase their potential for harm, such as being jailbroken. Such results raise a crucial question for both capability and safety forecasting: how can one accurately predict a model’s behavior when scaled to a massive number of attempts, given a vastly smaller sampling budget? This question is directly relevant to model providers, who serve hundreds of millions of users daily, and to governmental regulators, who seek to prevent harms. To answer this questions, we make three contributions. First, we find that standard methods for fitting these laws suffer from statistical shortcomings that hinder predictive accuracy, especially in data-limited scenarios. Second, we remedy these shortcomings by introducing a robust estimation framework, which uses a beta-binomial distribution to generate more accurate predictions from limited data. Third, we propose a dynamic sampling strategy that allocates a greater budget to harder problems. Combined, these innovations enable more reliable prediction of rare risks and capabilities at a fraction of the computational cost.

Method: Proposes a hybrid framework integrating: (i) model-based rewards from synthetic/human feedback, (ii) rule-based rewards with domain-specific heuristics, (iii) multi-aspect rewards for instruction adherence, and (iv) generalized length-penalty reward for training stability.

Result: Consistent improvements across multimodal benchmarks; 3B model achieves ~9.5% average improvement on general/math tasks and ~16% improvement specifically on mathematical benchmarks.

Conclusion: The hybrid reward modeling framework provides a flexible and effective approach for aligning MLLMs through reinforcement learning, particularly effective for mathematical reasoning tasks.

Abstract: Aligning multimodal large language models (MLLMs) with human preferences often relies on single-signal, model-based reward methods. Such monolithic rewards often lack confidence calibration across domain-specific tasks, fail to capture diverse aspects of human preferences, and require extensive data annotation and reward model training. In this work, we propose a hybrid reward modeling framework that integrates complementary reward paradigms: (i) model-based rewards, where a learned reward model predicts scalar or vector scores from synthetic and human feedback, and (ii) rule-based rewards, where domain-specific heuristics provide explicit correctness signals with confidence. Beyond accuracy, we further incorporate multi-aspect rewards to enforce instruction adherence and introduce a generalized length-penalty reward to stabilize training and improve performance. The proposed framework provides a flexible and effective approach to aligning MLLMs through reinforcement learning policy optimization. Our experiments show consistent improvements across different multimodal benchmarks when applying hybrid and multi-aspect reward modeling. Our best performing model in the 3B family achieves an overall average improvement of ~9.5% across general and math reasoning tasks. Focusing specifically on mathematical benchmarks, the model achieves a significant average improvement of ~16%, highlighting its effectiveness in mathematical reasoning and problem solving.

Method: The benchmark introduces: (1) comprehensive interaction environment with hierarchical knowledge base and function-driven user simulator, (2) two evaluation settings (c-Interact with predefined protocol and a-Interact with autonomous decision-making), and (3) challenging task suite covering full CRUD spectrum with executable test cases.

Result: The benchmark proves highly challenging - GPT-5 completes only 8.67% of tasks in c-Interact and 17.00% in a-Interact. Analysis validates the importance of effective interaction for complex text-to-SQL tasks.

Conclusion: BIRD-INTERACT restores realism to multi-turn text-to-SQL evaluation and highlights the difficulty of dynamic interaction tasks, providing a comprehensive framework for assessing and developing database assistant capabilities.

Abstract: Large language models (LLMs) have demonstrated remarkable performance on single-turn text-to-SQL tasks, but real-world database applications predominantly require multi-turn interactions to handle ambiguous queries, execution errors, and evolving user requirements. Existing multi-turn benchmarks fall short by treating conversation histories as static context or limiting evaluation to read-only operations, failing to reflect production-grade database assistant challenges. We introduce BIRD-INTERACT, a benchmark that restores this realism through: (1) a comprehensive interaction environment coupling each database with a hierarchical knowledge base, metadata files, and a function-driven user simulator, enabling models to solicit clarifications, retrieve knowledge, and recover from errors without human supervision; (2) two evaluation settings consisting of a pre-defined conversational protocol (c-Interact) and an open-ended agentic setting (a-Interact) where models autonomously decide when to query the user simulator or explore the environment; (3) a challenging task suite covering the full CRUD spectrum for business-intelligence and operational use cases, guarded by executable test cases. Each task features ambiguous and follow-up sub-tasks requiring dynamic interaction. The suite comprises BIRD-INTERACT-FULL (600 tasks, up to 11,796 interactions) for comprehensive performance assessment, and BIRD-INTERACT-LITE (300 tasks with simplified databases) for detailed behavioral analysis and rapid method development. Our empirical results highlight BIRD-INTERACT’s difficulty: GPT-5 completes only 8.67% of tasks in c-Interact and 17.00% in a-Interact. Analysis via memory grafting and Interaction Test-time Scaling validates the importance of effective interaction for complex, dynamic text-to-SQL tasks.

Method: Uses modular agent orchestration with specialized agents for different evidence streams, enabling parallel processing and fine-grained analysis with deterministic code validation.

Result: Substantial gains in efficiency and output consistency, with complete traceability from source evidence to conclusions through an interactive user interface.

Conclusion: M-Reason shows potential as both a practical tool for evidence synthesis and a testbed for robust multi-agent LLM systems in scientific research.

Abstract: We present M-Reason, a demonstration system for transparent, agent-based reasoning and evidence integration in the biomedical domain, with a focus on cancer research. M-Reason leverages recent advances in large language models (LLMs) and modular agent orchestration to automate evidence retrieval, appraisal, and synthesis across diverse biomedical data sources. Each agent specializes in a specific evidence stream, enabling parallel processing and fine-grained analysis. The system emphasizes explainability, structured reporting, and user auditability, providing complete traceability from source evidence to final conclusions. We discuss critical tradeoffs between agent specialization, system complexity, and resource usage, as well as the integration of deterministic code for validation. An open, interactive user interface allows researchers to directly observe, explore and evaluate the multi-agent workflow. Our evaluation demonstrates substantial gains in efficiency and output consistency, highlighting M-Reason’s potential as both a practical tool for evidence synthesis and a testbed for robust multi-agent LLM systems in scientific research, available at https://m-reason.digitalecmt.com.

Method: Systematic analysis of individual studies and their practical implementations in neural-network-based MPC modeling and control design.

Result: Bayesian approaches are increasingly used in MPC for uncertainty handling, but reported performance and robustness gains remain fragmented with inconsistent baselines and limited reliability analyses.

Conclusion: The paper argues for standardized benchmarks, ablation studies, and transparent reporting to rigorously determine the effectiveness of Bayesian techniques for MPC.

Abstract: In this review, we assess the use of Bayesian methods in model predictive control (MPC), focusing on neural-network-based modeling, control design, and uncertainty quantification. We systematically analyze individual studies and how they are implemented in practice. While Bayesian approaches are increasingly adopted to capture and propagate uncertainty in MPC, reported gains in performance and robustness remain fragmented, with inconsistent baselines and limited reliability analyses. We therefore argue for standardized benchmarks, ablation studies, and transparent reporting to rigorously determine the effectiveness of Bayesian techniques for MPC.

Method: Multi-Head Attention Retrieval-Augmented Generation (MHA-RAG) represents exemplars as soft prompts with an exemplar order invariant model architecture. The number of attention heads serves as a hyperparameter to control soft prompt-generation across different tasks.

Result: Across multiple question-answering benchmarks and model scales, MHA-RAG achieves a 20-point performance gain over standard RAG while cutting inference costs by a factor of 10X GFLOPs, delivering higher accuracy and greater efficiency invariant to exemplar order.

Conclusion: Representing exemplars as soft prompts with an exemplar order invariant architecture is more efficient, effective, and stable than text-based representations for adapting Foundation Models to new domains with limited training data.

Abstract: Adapting Foundation Models to new domains with limited training data is challenging and computationally expensive. While prior work has demonstrated the effectiveness of using domain-specific exemplars as in-context demonstrations, we investigate whether representing exemplars purely as text is the most efficient, effective, and stable approach. We explore an alternative: representing exemplars as soft prompts with an exemplar order invariant model architecture. To this end, we introduce Multi-Head Attention Retrieval-Augmented Generation (MHA-RAG), a framework with the number of attention heads serving as a simple hyperparameter to control soft prompt-generation across different tasks. Across multiple question-answering benchmarks and model scales, MHA-RAG achieves a 20-point performance gain over standard RAG, while cutting inference costs by a factor of 10X GFLOPs-delivering both higher accuracy and greater efficiency, invariant to exemplar order.

Method: Two studies drawing on Symbolic Interactionism theory to investigate how humans and AI co-construct symbols and their meanings during conversational interactions.

Result: Participants shifted their initial definitions of meaning in response to AI-suggested symbols and interpretations, especially when social context was introduced. Participants projected personal and social values into interactions, refining meanings over time.

Conclusion: Shared understanding in human-AI interaction emerges from bidirectional exchange and reinterpretation of symbols, suggesting new paradigms for human-AI interaction design that account for dynamic meaning construction.

Abstract: Meaningful human-AI collaboration requires more than processing language; it demands a deeper understanding of symbols and their socially constructed meanings. While humans naturally interpret symbols through social interaction, AI systems often miss the dynamic interpretations that emerge in conversation. Drawing on Symbolic Interactionism theory, we conducted two studies to investigate how humans and AI co-construct symbols and their meanings. Findings provide empirical insights into how humans and conversational AI agents collaboratively shape meanings during interaction. We show how participants shift their initial definitions of meaning in response to the symbols and interpretations suggested by the conversational AI agents, especially when social context is introduced. We also observe how participants project their personal and social values into these interactions, refining meanings over time. These findings reveal that shared understanding does not emerge from mere agreement but from the bi-directional exchange and reinterpretation of symbols, suggesting new paradigms for human-AI interaction design.

Method: Train a forward model (Teacher) to predict hardness from 13 input features, then train a backward model (Student) to infer input configurations from target hardness using iterative Teacher feedback in supervised loop.

Result: Outperforms baseline regression and reinforcement learning models in inverse prediction accuracy with significantly less computational time on tempered steel dataset.

Conclusion: The Teacher-Student framework is effective and efficient for inverse process modeling in materials science applications.

Abstract: Predicting the final hardness of steel after heat treatment is a challenging regression task due to the many-to-one nature of the process – different combinations of input parameters (such as temperature, duration, and chemical composition) can result in the same hardness value. This ambiguity makes the inverse problem, estimating input parameters from a desired hardness, particularly difficult. In this work, we propose a novel solution using a Teacher-Student learning framework. First, a forward model (Teacher) is trained to predict final hardness from 13 metallurgical input features. Then, a backward model (Student) is trained to infer plausible input configurations from a target hardness value. The Student is optimized by leveraging feedback from the Teacher in an iterative, supervised loop. We evaluate our method on a publicly available tempered steel dataset and compare it against baseline regression and reinforcement learning models. Results show that our Teacher-Student framework not only achieves higher inverse prediction accuracy but also requires significantly less computational time, demonstrating its effectiveness and efficiency for inverse process modeling in materials science.

Method: Extract distilled problem statements from ICLR 2025 submissions, use specialized solver agents with iterative critique loops, and evaluate using LLM-as-a-judge paradigm with structured rubric and manual checks.

Result: LLMs can rediscover feasible solutions and occasionally propose creative alternatives, but their problem-solving ability is fragile and highly sensitive to framing.

Conclusion: LLMs have latent potential as scientific problem-solvers but currently face significant limitations in robust reasoning capabilities.

Abstract: Large language models (LLMs) demonstrate impressive capabilities across a wide range of tasks, yet it remains unclear whether such success reflects genuine reasoning or sophisticated recall. We introduce AInstein, a framework for testing whether LLMs can generate valid solutions to AI research problems using only their pretrained parametric knowledge – without domain-specific fine-tuning, retrieval augmentation, or other external aids. Our approach extracts distilled problem statements from high-quality ICLR 2025 submissions, then tasks specialized solver agents with proposing and refining technical solutions through iterative critique loops, mimicking the cycles of proposal, review, and revision central to scientific inquiry. We evaluate AInstein on 1,214 ICLR papers stratified by acceptance tier (Oral, Spotlight, Poster), using an LLM-as-a-judge paradigm guided by a structured rubric, complemented by targeted manual checks. Performance is assessed with three metrics: Success Rate (does the solution address the problem?), Rediscovery (does it align with human-proposed methods?), and Novelty (does it yield valid, original approaches?). Our results reveal that while LLMs can rediscover feasible solutions and occasionally propose creative alternatives, their problem-solving ability remains fragile and highly sensitive to framing. These findings provide the first large-scale evidence on the extent to which LLMs can act as autonomous scientific problem-solvers, highlighting both their latent potential and their current limitations.

Method: Hybrid framework integrating ASP rules with transformers: (1) rule-based data augmentation for logically consistent samples, (2) fuzzy-logic regularizer during fine-tuning, and (3) per-class threshold tuning.

Result: Improved micro- and macro-F1 scores compared to BCE baseline, with 86% reduction in rule violations on ASRS test set.

Conclusion: First large-scale neuro-symbolic application to ASRS that unifies ASP reasoning, rule-driven augmentation, and differentiable transformer training for trustworthy safety-critical NLP.

Abstract: Deep transformer models excel at multi-label text classification but often violate domain logic that experts consider essential, an issue of particular concern in safety-critical applications. We propose a hybrid neuro-symbolic framework that integrates Answer Set Programming (ASP) with transformer-based learning on the Aviation Safety Reporting System (ASRS) corpus. Domain knowledge is formalized as weighted ASP rules and validated using the Clingo solver. These rules are incorporated in two complementary ways: (i) as rule-based data augmentation, generating logically consistent synthetic samples that improve label diversity and coverage; and (ii) as a fuzzy-logic regularizer, enforcing rule satisfaction in a differentiable form during fine-tuning. This design preserves the interpretability of symbolic reasoning while leveraging the scalability of deep neural architectures. We further tune per-class thresholds and report both standard classification metrics and logic-consistency rates. Compared to a strong Binary Cross-Entropy (BCE) baseline, our approach improves micro- and macro-F1 scores and achieves up to an 86% reduction in rule violations on the ASRS test set. To the best of our knowledge, this constitutes the first large-scale neuro-symbolic application to ASRS reports that unifies ASP-based reasoning, rule-driven augmentation, and differentiable transformer training for trustworthy, safety-critical NLP.

Method: Analyzed model confidence and performance across diverse programming languages, focusing on state-of-the-art LLMs in coding tasks.

Result: AI models mirror human patterns of overconfidence, with less competent models and those operating in rare programming languages showing stronger DKE-like bias.

Conclusion: The strength of Dunning-Kruger bias in AI models is proportionate to their competence level, revealing similar cognitive limitations as humans.

Abstract: As artificial intelligence systems increasingly collaborate with humans in creative and technical domains, questions arise about the cognitive boundaries and biases that shape our shared agency. This paper investigates the Dunning-Kruger Effect (DKE), the tendency for those with limited competence to overestimate their abilities in state-of-the-art LLMs in coding tasks. By analyzing model confidence and performance across a diverse set of programming languages, we reveal that AI models mirror human patterns of overconfidence, especially in unfamiliar or low-resource domains. Our experiments demonstrate that less competent models and those operating in rare programming languages exhibit stronger DKE-like bias, suggesting that the strength of the bias is proportionate to the competence of the models.

Method: The benchmark uses 115K paired prompts from Wikipedia’s controversial sections that frame opposing sides of public debates, and measures model alignment using LLM-as-judge to score agreement or divergence between paired responses.

Result: The benchmark reveals large variation in alignment across leading open- and closed-source models, and highlights trade-offs between safety strategies (like refusals) and more expressive value systems.

Conclusion: VAL-Bench provides a scalable, reproducible benchmark that enables systematic comparison of how reliably LLMs embody human values, addressing the critical need to test whether model responses reflect consistent human values.

Abstract: Large language models (LLMs) are increasingly used for tasks where outputs shape human decisions, so it is critical to test whether their responses reflect consistent human values. Existing benchmarks mostly track refusals or predefined safety violations, but these only check rule compliance and do not reveal whether a model upholds a coherent value system when facing controversial real-world issues. We introduce the \textbf{V}alue \textbf{AL}ignment \textbf{Bench}mark (\textbf{VAL-Bench}), which evaluates whether models maintain a stable value stance across paired prompts that frame opposing sides of public debates. VAL-Bench consists of 115K such pairs from Wikipedia’s controversial sections. A well-aligned model should express similar underlying views regardless of framing, which we measure using an LLM-as-judge to score agreement or divergence between paired responses. Applied across leading open- and closed-source models, the benchmark reveals large variation in alignment and highlights trade-offs between safety strategies (e.g., refusals) and more expressive value systems. By providing a scalable, reproducible benchmark, VAL-Bench enables systematic comparison of how reliably LLMs embody human values.

Method: Recent research formulates AVR as a sequence generation problem using large language models, either through prompting or fine-tuning, but these methods primarily rely on foundation models with general programming knowledge.

Result: Current methods show state-of-the-art performance but fail to capture diverse vulnerability repair patterns due to lack of vulnerability-related reasoning data and the inability to verify intermediate repair processes during training.

Conclusion: There is a critical need for approaches that can incorporate vulnerability-specific reasoning data and provide verifiable intermediate feedback to improve automatic vulnerability repair capabilities.

Abstract: The exponential increase in software vulnerabilities has created an urgent need for automatic vulnerability repair (AVR) solutions. Recent research has formulated AVR as a sequence generation problem and has leveraged large language models (LLMs) to address this problem. Typically, these approaches prompt or fine-tune LLMs to generate repairs for vulnerabilities directly. Although these methods show state-of-the-art performance, they face the following challenges: (1) Lack of high-quality, vulnerability-related reasoning data. Current approaches primarily rely on foundation models that mainly encode general programming knowledge. Without vulnerability-related reasoning data, they tend to fail to capture the diverse vulnerability repair patterns. (2) Hard to verify the intermediate vulnerability repair process during LLM training. Existing reinforcement learning methods often leverage intermediate execution feedback from the environment (e.g., sandbox-based execution results) to guide reinforcement learning training. In contrast, the vulnerability repair process generally lacks such intermediate, verifiable feedback, which poses additional challenges for model training.

Method: Comprehensive comparison of 21 time series models including univariate and multivariate approaches. Top performers (FFNN, SVM, RFR) were integrated with Linear Regression using a custom stacked generalization ensemble technique.

Result: Feedforward Neural Network (FFNN), Support Vector Machine (SVM), and Random Forest Regressor (RFR) achieved the best performances among the 21 models. The proposed ensemble model achieved SMAPE of 1.407 with no signs of overfitting.

Conclusion: The research provides an accurate decade-long emission projection that will assist policymakers in making more data-driven decisions for addressing Taiwan’s air pollution challenges.

Abstract: Taiwan’s high population and heavy dependence on fossil fuels have led to severe air pollution, with the most prevalent greenhouse gas being carbon dioxide (CO2). There-fore, this study presents a reproducible and comprehensive case study comparing 21 of the most commonly employed time series models in forecasting emissions, analyzing both univariate and multivariate approaches. Among these, Feedforward Neural Network (FFNN), Support Vector Machine (SVM), and Random Forest Regressor (RFR) achieved the best performances. To further enhance robustness, the top performers were integrated with Linear Regression through a custom stacked generalization en-semble technique. Our proposed ensemble model achieved an SMAPE of 1.407 with no signs of overfitting. Finally, this research provides an accurate decade-long emission projection that will assist policymakers in making more data-driven decisions.

Method: Proposes Context-Aware Meta Co-Training that consolidates diverse target tasks into single fine-tuning stage, using structurally diverse auxiliary tasks with lightweight meta-learning mechanism derived from Attentive Neural Processes.

Result: On LIBERO benchmark, MetaVLA outperforms OpenVLA by up to 8.0% on long-horizon tasks, reduces training steps from 240K to 75K, and cuts GPU time by ~76%.

Conclusion: Scalable, low-resource post-training is achievable, paving the way toward general-purpose embodied agents.

Abstract: Vision-Language-Action (VLA) models show promise in embodied reasoning, yet remain far from true generalists-they often require task-specific fine-tuning, and generalize poorly to unseen tasks. We propose MetaVLA, a unified, backbone-agnostic post-training framework for efficient and scalable alignment. MetaVLA introduces Context-Aware Meta Co-Training, which consolidates diverse target tasks into a single fine-tuning stage while leveraging structurally diverse auxiliary tasks to improve in-domain generalization. Unlike naive multi-task SFT, MetaVLA integrates a lightweight meta-learning mechanism-derived from Attentive Neural Processes-to enable rapid adaptation from diverse contexts with minimal architectural change or inference overhead. On the LIBERO benchmark, MetaVLA with six auxiliary tasks outperforms OpenVLA by up to 8.0% on long-horizon tasks, reduces training steps from 240K to 75K, and cuts GPU time by ~76%. These results show that scalable, low-resource post-training is achievable-paving the way toward general-purpose embodied agents. Code will be available.

Method: Multi-agent cooperative framework with role-specialized LLMs coordinated by a supervisor agent, using structured dialogue, iterative feedback, and systematic validation for autonomous evolution cycles.

Result: The framework autonomously upgraded fixed antenna optimization to movable antenna optimization, improving beam gain by up to 52.02% and consistently surpassing fixed baseline performance.

Conclusion: Self-evolving agentic AI provides adaptable and robust intelligence for next-generation wireless systems, enabling autonomous performance improvement with minimal human intervention.

Abstract: Self-evolving agentic artificial intelligence (AI) offers a new paradigm for future wireless systems by enabling autonomous agents to continually adapt and improve without human intervention. Unlike static AI models, self-evolving agents embed an autonomous evolution cycle that updates models, tools, and workflows in response to environmental dynamics. This paper presents a comprehensive overview of self-evolving agentic AI, highlighting its layered architecture, life cycle, and key techniques, including tool intelligence, workflow optimization, self-reflection, and evolutionary learning. We further propose a multi-agent cooperative self-evolving agentic AI framework, where multiple large language models (LLMs) are assigned role-specialized prompts under the coordination of a supervisor agent. Through structured dialogue, iterative feedback, and systematic validation, the system autonomously executes the entire life cycle without human intervention. A case study on antenna evolution in low-altitude wireless networks (LAWNs) demonstrates how the framework autonomously upgrades fixed antenna optimization into movable antenna optimization. Experimental results show that the proposed self-evolving agentic AI autonomously improves beam gain and restores degraded performance by up to 52.02%, consistently surpassing the fixed baseline with little to no human intervention and validating its adaptability and robustness for next-generation wireless intelligence.

Method: Retrospective study using radiography series of clavicle, elbow, and thumb. GPT-4o extracted labels indicating findings as present, absent, or uncertain. Labels were used for multi-label classification with ResNet50, with uncertain labels reassigned to true (inclusive) or false (exclusive) strategies. Performance evaluated on internal and external test sets using ROC AUC, precision recall curves, sensitivity, specificity, and accuracy.

Result: GPT-4o achieved 98.6% accuracy in label extraction. Multi-label classification models showed competitive performance across anatomic regions (e.g., elbow AUC=0.80) and generalized well to external datasets. No significant differences were found between labeling strategies or datasets (p>=0.15).

Conclusion: GPT-4o can extract labels from radiologic reports to train competitive multi-label classification models with high accuracy, and detected uncertainty in reports does not influence model performance.

Abstract: Objectives: To evaluate GPT-4o’s ability to extract diagnostic labels (with uncertainty) from free-text radiology reports and to test how these labels affect multi-label image classification of musculoskeletal radiographs. Methods: This retrospective study included radiography series of the clavicle (n=1,170), elbow (n=3,755), and thumb (n=1,978). After anonymization, GPT-4o filled out structured templates by indicating imaging findings as present (“true”), absent (“false”), or “uncertain.” To assess the impact of label uncertainty, “uncertain” labels of the training and validation sets were automatically reassigned to “true” (inclusive) or “false” (exclusive). Label-image-pairs were used for multi-label classification using ResNet50. Label extraction accuracy was manually verified on internal (clavicle: n=233, elbow: n=745, thumb: n=393) and external test sets (n=300 for each). Performance was assessed using macro-averaged receiver operating characteristic (ROC) area under the curve (AUC), precision recall curves, sensitivity, specificity, and accuracy. AUCs were compared with the DeLong test. Results: Automatic extraction was correct in 98.6% (60,618 of 61,488) of labels in the test sets. Across anatomic regions, label-based model training yielded competitive performance measured by macro-averaged AUC values for inclusive (e.g., elbow: AUC=0.80 [range, 0.62-0.87]) and exclusive models (elbow: AUC=0.80 [range, 0.61-0.88]). Models generalized well on external datasets (elbow [inclusive]: AUC=0.79 [range, 0.61-0.87]; elbow [exclusive]: AUC=0.79 [range, 0.63-0.89]). No significant differences were observed across labeling strategies or datasets (p>=0.15). Conclusion: GPT-4o extracted labels from radiologic reports to train competitive multi-label classification models with high accuracy. Detected uncertainty in the radiologic reports did not influence the performance of these models.

Method: Three components: OWA Toolkit for unified desktop interaction format with 152x compression, Generalist-IDM for zero-shot generalization across games via timestamp-based event prediction, and VAPT for transferring pretrained representations to physical tasks.

Result: Using 1.3K+ hours of data (259h human + 1K+ pseudo-labeled), achieved 96.6% success on LIBERO manipulation and 83.3% on CANVAS navigation benchmarks.

Conclusion: Desktop pretraining is a practical paradigm for robotics as sensorimotor primitives in digital interactions transfer meaningfully to physical embodied tasks.

Abstract: Large language models leverage internet-scale text data, yet embodied AI remains constrained by the prohibitive costs of physical trajectory collection. Desktop environments – particularly gaming – offer a compelling alternative: they provide rich sensorimotor interactions at scale while maintaining the structured observation-action coupling essential for embodied learning. We present D2E (Desktop to Embodied AI), a framework that demonstrates desktop interactions can serve as an effective pretraining substrate for robotics embodied AI tasks. Unlike prior work that remained domain-specific (e.g., VPT for Minecraft) or kept data proprietary (e.g., SIMA), D2E establishes a complete pipeline from scalable desktop data collection to verified transfer in embodied domains. Our framework comprises three components: (1) the OWA Toolkit that unifies diverse desktop interactions into a standardized format with 152x compression, (2) the Generalist-IDM that achieves strong zero-shot generalization across unseen games through timestamp-based event prediction, enabling internet-scale pseudo-labeling, and (3) VAPT that transfers desktop-pretrained representations to physical manipulation and navigation. Using 1.3K+ hours of data (259 hours of human demonstrations, and 1K+ hours of pseudo-labeled gameplay), we achieve a total of 96.6% success rate on LIBERO manipulation and 83.3% on CANVAS navigation benchmarks. This validates that sensorimotor primitives in digital interactions exhibit sufficient invariance to transfer meaningfully to physical embodied tasks, establishing desktop pretraining as a practical paradigm for robotics. We will make all our work public, including the OWA toolkit, datasets of human-collected and pseudo-labeled, and VAPT-trained models available at https://worv-ai.github.io/d2e/

Method: Proposed AIC-VDS uses attention-based in-context learning with LLMs to jointly optimize data collection schedules and flight velocities for multiple UAVs, considering sensor battery levels, queue lengths, channel conditions, and UAV trajectories.

Result: Simulation results show AIC-VDS outperforms both Deep-Q-Network (DQN) and maximum channel gain baselines in minimizing data loss.

Conclusion: LLM-based in-context learning provides an effective alternative to DRL for emergency scenarios like tsunami monitoring, offering better performance without complex training processes.

Abstract: Recently, Unmanned Aerial Vehicles (UAVs) are increasingly being investigated to collect sensory data in post-disaster monitoring scenarios, such as tsunamis, where early actions are critical to limit coastal damage. A major challenge is to design the data collection schedules and flight velocities, as unfavorable schedules and velocities can lead to transmission errors and buffer overflows of the ground sensors, ultimately resulting in significant packet loss. Meanwhile, online Deep Reinforcement Learning (DRL) solutions have a complex training process and a mismatch between simulation and reality that does not meet the urgent requirements of tsunami monitoring. Recent advances in Large Language Models (LLMs) offer a compelling alternative. With their strong reasoning and generalization capabilities, LLMs can adapt to new tasks through In-Context Learning (ICL), which enables task adaptation through natural language prompts and example-based guidance without retraining. However, LLM models have input data limitations and thus require customized approaches. In this paper, a joint optimization of data collection schedules and velocities control for multiple UAVs is proposed to minimize data loss. The battery level of the ground sensors, the length of the queues, and the channel conditions, as well as the trajectories of the UAVs, are taken into account. Attention-Based In-Context Learning for Velocity Control and Data Collection Schedule (AIC-VDS) is proposed as an alternative to DRL in emergencies. The simulation results show that the proposed AIC-VDS outperforms both the Deep-Q-Network (DQN) and maximum channel gain baselines.

Method: Three-tiered mechanism: 1) Visual-Semantic Synergy for cross-modal alignment, 2) Content-Aware Reasoning with dynamic contextual prompts, 3) Cloud-Edge Synergy using knowledge distillation for lightweight edge models.

Result: Significantly outperforms existing methods in 1-shot and cross-condition scenarios. Edge model achieves comparable performance to cloud version with 83% smaller size and 50% lower latency.

Conclusion: Syn-Diag provides a practical, robust, and deployable paradigm for modern intelligent diagnostics in industrial settings.

Abstract: Industrial fault diagnosis faces the dual challenges of data scarcity and the difficulty of deploying large AI models in resource-constrained environments. This paper introduces Syn-Diag, a novel cloud-edge synergistic framework that leverages Large Language Models to overcome these limitations in few-shot fault diagnosis. Syn-Diag is built on a three-tiered mechanism: 1) Visual-Semantic Synergy, which aligns signal features with the LLM’s semantic space through cross-modal pre-training; 2) Content-Aware Reasoning, which dynamically constructs contextual prompts to enhance diagnostic accuracy with limited samples; and 3) Cloud-Edge Synergy, which uses knowledge distillation to create a lightweight, efficient edge model capable of online updates via a shared decision space. Extensive experiments on six datasets covering different CWRU and SEU working conditions show that Syn-Diag significantly outperforms existing methods, especially in 1-shot and cross-condition scenarios. The edge model achieves performance comparable to the cloud version while reducing model size by 83% and latency by 50%, offering a practical, robust, and deployable paradigm for modern intelligent diagnostics.

Method: Comprehensive historical review and analysis of developments from early computer simulations to current large language model experiments.

Result: Documents the progression from initial social simulations to sophisticated AI applications, highlighting technological and epistemic transformations in social science research.

Conclusion: Human understanding of ourselves is deeply intertwined with the technological tools we develop, with AI agents playing an increasingly central role in social science research.

Abstract: We review the historical development and current trends of artificially intelligent agents (agentic AI) in the social and behavioral sciences: from the first programmable computers, and social simulations soon thereafter, to today’s experiments with large language models. This overview emphasizes the role of AI in the scientific process and the changes brought about, both through technological advancements and the broader evolution of science from around 1950 to the present. Some of the specific points we cover include: the challenges of presenting the first social simulation studies to a world unaware of computers, the rise of social systems science, intelligent game theoretic agents, the age of big data and the epistemic upheaval in its wake, and the current enthusiasm around applications of generative AI, and many other topics. A pervasive theme is how deeply entwined we are with the technologies we use to understand ourselves.

Method: Develop ARM as an agentic generalization of CoT where specialized reasoning modules are discovered through tree search over code space, starting from simple CoT and evolving using mutations informed by execution trace reflection.

Result: ARM significantly outperforms both manually designed MASes and state-of-the-art automatic MAS design methods, with superb generalization across different foundation models and task domains without further optimization.

Conclusion: Optimizing CoT reasoning through ARM provides a versatile building block for MAS design that achieves superior performance and generalization, making it more effective than current automatic MAS design approaches.

Abstract: Large Language Model (LLM)-powered Multi-agent systems (MAS) have achieved state-of-the-art results on various complex reasoning tasks. Recent works have proposed techniques to automate the design of MASes, eliminating the need for manual engineering. However, these techniques perform poorly, often achieving similar or inferior performance to simple baselines. Furthermore, they require computationally expensive re-discovery of architectures for each new task domain and expensive data annotation on domains without existing labeled validation sets. A critical insight is that simple Chain of Thought (CoT) reasoning often performs competitively with these complex systems, suggesting that the fundamental reasoning unit of MASes, CoT, warrants further investigation. To this end, we present a new paradigm for automatic MAS design that pivots the focus to optimizing CoT reasoning. We introduce the Agentic Reasoning Module (ARM), an agentic generalization of CoT where each granular reasoning step is executed by a specialized reasoning module. This module is discovered through a tree search over the code space, starting from a simple CoT module and evolved using mutations informed by reflection on execution traces. The resulting ARM acts as a versatile reasoning building block which can be utilized as a direct recursive loop or as a subroutine in a learned meta-orchestrator. Our approach significantly outperforms both manually designed MASes and state-of-the-art automatic MAS design methods. Crucially, MASes built with ARM exhibit superb generalization, maintaining high performance across different foundation models and task domains without further optimization.

Method: Developed an ensemble-based pipeline using graph neural network emulator of Lagrangian Particle Dispersion Model (LPDM) to estimate atmospheric transport footprints and greenhouse gas mole fractions with uncertainty quantification.

Result: The emulator achieved ~1000x speed-up over NAME LPDM while reproducing large-scale footprint structures. Ensemble analysis revealed spatial correlations with prediction error and identified low-confidence predictions for both transport footprints and methane mole fractions.

Conclusion: The ensemble-based emulator approach can be generalized to atmospheric transport models, supporting uncertainty-aware greenhouse gas inversion systems and improving the robustness of satellite-based emissions monitoring, with potential for exploring systematic model errors.

Abstract: Monitoring greenhouse gas emissions and evaluating national inventories require efficient, scalable, and reliable inference methods. Top-down approaches, combined with recent advances in satellite observations, provide new opportunities to evaluate emissions at continental and global scales. However, transport models used in these methods remain a key source of uncertainty: they are computationally expensive to run at scale, and their uncertainty is difficult to characterise. Artificial intelligence offers a dual opportunity to accelerate transport simulations and to quantify their associated uncertainty. We present an ensemble-based pipeline for estimating atmospheric transport “footprints”, greenhouse gas mole fraction measurements, and their uncertainties using a graph neural network emulator of a Lagrangian Particle Dispersion Model (LPDM). The approach is demonstrated with GOSAT (Greenhouse Gases Observing Satellite) observations for Brazil in 2016. The emulator achieved a ~1000x speed-up over the NAME LPDM, while reproducing large-scale footprint structures. Ensembles were calculated to quantify absolute and relative uncertainty, revealing spatial correlations with prediction error. The results show that ensemble spread highlights low-confidence spatial and temporal predictions for both atmospheric transport footprints and methane mole fractions. While demonstrated here for an LPDM emulator, the approach could be applied more generally to atmospheric transport models, supporting uncertainty-aware greenhouse gas inversion systems and improving the robustness of satellite-based emissions monitoring. With further development, ensemble-based emulators could also help explore systematic LPDM errors, offering a computationally efficient pathway towards a more comprehensive uncertainty budget in greenhouse gas flux estimates.

Method: Used large-scale cross-lingual dataset from 25 Reddit communities, defined virality using hybrid engagement score with percentile-based threshold, evaluated Logistic Regression, XGBoost, and MLP models with multimodal features across time windows (30-420 min).

Result: XGBoost achieved PR-AUC > 0.52 in just 30 minutes, identified ’evidentiary transition’ where feature importance shifts from static context to temporal dynamics as meme gains traction.

Conclusion: Established robust, interpretable benchmark for early virality prediction, contributed novel cross-lingual dataset and methodologically sound virality definition, first study combining time series with static content and network features for early meme virality prediction.

Abstract: Predicting the virality of online content remains challenging, especially for culturally complex, fast-evolving memes. This study investigates the feasibility of early prediction of meme virality using a large-scale, cross-lingual dataset from 25 diverse Reddit communities. We propose a robust, data-driven method to define virality based on a hybrid engagement score, learning a percentile-based threshold from a chronologically held-out training set to prevent data leakage. We evaluated a suite of models, including Logistic Regression, XGBoost, and a Multi-layer Perceptron (MLP), with a comprehensive, multimodal feature set across increasing time windows (30-420 min). Crucially, useful signals emerge quickly: our best-performing model, XGBoost, achieves a PR-AUC $>$ 0.52 in just 30 minutes. Our analysis reveals a clear “evidentiary transition,” in which the importance of the feature dynamically shifts from the static context to the temporal dynamics as a meme gains traction. This work establishes a robust, interpretable, and practical benchmark for early virality prediction in scenarios where full diffusion cascade data is unavailable, contributing a novel cross-lingual dataset and a methodologically sound definition of virality. To our knowledge, this study is the first to combine time series data with static content and network features to predict early meme virality.

Method: Trained an open-source LLM with multi-instruction supervised fine-tuning, integrated with Constraint-Aware Retrieval Module for enhanced in-context learning, and used Tree-of-Thoughts framework with guided self-correction mechanism.

Result: ConstraintLLM achieves state-of-the-art solving accuracy across multiple benchmarks and outperforms baselines by 2x on the new IndusCP benchmark containing 140 industrial tasks.

Conclusion: ConstraintLLM successfully demonstrates the potential of specialized LLMs for constraint programming modeling, providing a foundation for trustworthy neuro-symbolic AI systems.

Abstract: Constraint programming (CP) is a crucial technology for solving real-world constraint optimization problems (COPs), with the advantages of rich modeling semantics and high solving efficiency. Using large language models (LLMs) to generate formal modeling automatically for COPs is becoming a promising approach, which aims to build trustworthy neuro-symbolic AI with the help of symbolic solvers. However, CP has received less attention compared to works based on operations research (OR) models. We introduce ConstraintLLM, the first LLM specifically designed for CP modeling, which is trained on an open-source LLM with multi-instruction supervised fine-tuning. We propose the Constraint-Aware Retrieval Module (CARM) to increase the in-context learning capabilities, which is integrated in a Tree-of-Thoughts (ToT) framework with guided self-correction mechanism. Moreover, we construct and release IndusCP, the first industrial-level benchmark for CP modeling, which contains 140 challenging tasks from various domains. Our experiments demonstrate that ConstraintLLM achieves state-of-the-art solving accuracy across multiple benchmarks and outperforms the baselines by 2x on the new IndusCP benchmark. Code and data are available at: https://github.com/william4s/ConstraintLLM.

Method: Conducted literature review of World Models in autonomous driving and robotics, complemented by empirical analysis collecting predictions from state-of-the-art models, identifying and categorizing common faults (pathologies).

Result: Provided quantitative evaluation of prediction results and identified common pathologies in World Model predictions.

Conclusion: World Models need to ensure safe predictions for embodied agents, and the study provides insights into common prediction faults that need to be addressed for safe deployment.

Abstract: The rapid progress in embodied artificial intelligence has highlighted the necessity for more advanced and integrated models that can perceive, interpret, and predict environmental dynamics. In this context, World Models (WMs) have been introduced to provide embodied agents with the abilities to anticipate future environmental states and fill in knowledge gaps, thereby enhancing agents’ ability to plan and execute actions. However, when dealing with embodied agents it is fundamental to ensure that predictions are safe for both the agent and the environment. In this article, we conduct a comprehensive literature review of World Models in the domains of autonomous driving and robotics, with a specific focus on the safety implications of scene and control generation tasks. Our review is complemented by an empirical analysis, wherein we collect and examine predictions from state-of-the-art models, identify and categorize common faults (herein referred to as pathologies), and provide a quantitative evaluation of the results.

Method: Uses model’s own confidence metrics (self-consistency and predictive perplexity) to filter synthetic reasoning traces, sampling multiple traces and retaining only low-uncertainty subsets for supervised fine-tuning.

Result: Uncertainty-filtered data has higher accuracy, outperforms unfiltered synthetic data in SFT, narrows gap to ground-truth training, and surpasses strong LRM baselines in biological perturbation prediction.

Conclusion: Model-internal confidence is a powerful signal for efficient reasoning dataset creation, enabling large reasoning models in domains where supervision is expensive.

Abstract: Synthetic chain-of-thought (CoT) traces are widely used to train large reasoning models (LRMs), improving generalization by providing step-level supervision. Yet most approaches require ground-truth labels to seed or filter these traces - an expensive bottleneck in domains like biology where wet-lab data are scarce. We propose a label-free alternative: uncertainty-based filtering, which uses a model’s own confidence - quantified through established uncertainty metrics like self-consistency and predictive perplexity - as a substitute for external labels. We sample multiple reasoning traces and retain only low-uncertainty subsets. Applied to biological perturbation prediction, a domain where wet-lab labels are especially costly, we show that the filtered subset has higher accuracy, and that supervised fine-tuning (SFT) on uncertainty-filtered data outperforms unfiltered synthetic data, narrows the gap to ground-truth training, and surpasses strong LRM baselines. Ablations show that per-class filtering corrects for class-specific uncertainty scales and that hybrid uncertainty metrics yield higher-quality datasets. Our results suggest that model-internal confidence is a powerful signal for efficient reasoning dataset creation, enabling LRMs in domains where supervision is expensive.

Method: DebateQD - a minimal Quality-Diversity evolutionary algorithm that evolves diverse debate strategies across categories (rationality, authority, emotional appeal) through tournament-style competitions where two LLMs debate while a third judges, using prompt-based strategies within a single LLM architecture.

Result: Persuasion-optimized strategies achieve up to 13.94% smaller train-test generalization gaps while matching or exceeding truth optimization’s test performance across three model scales (7B, 32B, 72B parameters) and multiple dataset sizes.

Conclusion: Competitive pressure to persuade, rather than seek truth collaboratively, fosters more transferable reasoning skills, offering a promising path for improving LLM generalization.

Abstract: Large Language Models (LLMs) optimized to output truthful answers often overfit, producing brittle reasoning that fails to generalize. While persuasion-based optimization has shown promise in debate settings, it has not been systematically compared against mainstream truth-based approaches. We introduce DebateQD, a minimal Quality-Diversity (QD) evolutionary algorithm that evolves diverse debate strategies across different categories (rationality, authority, emotional appeal, etc.) through tournament-style competitions where two LLMs debate while a third judges. Unlike previously proposed methods that require a population of LLMs, our approach maintains diversity of opponents through prompt-based strategies within a single LLM architecture, making it more accessible for experiments while preserving the key benefits of population-based optimization. In contrast to prior work, we explicitly isolate the role of the optimization objective by fixing the debate protocol and swapping only the fitness function: persuasion rewards strategies that convince the judge irrespective of truth, whereas truth rewards collaborative correctness. Across three model scales (7B, 32B, 72B parameters) and multiple dataset sizes from the QuALITY benchmark, persuasion-optimized strategies achieve up to 13.94% smaller train-test generalization gaps, while matching or exceeding truth optimization’s test performance. These results provide the first controlled evidence that competitive pressure to persuade, rather than seek the truth collaboratively, fosters more transferable reasoning skills, offering a promising path for improving LLM generalization.

Method: Decomposes multivariate series into channel-wise subproblems, retrieves structurally similar labeled examples for each channel, uses reasoning LLM to compare query against exemplars with confidence scores, and fuses channel decisions via confidence-weighted aggregation.

Result: Achieves strong accuracy on nine UEA datasets under fully training-free setting, surpassing multiple trained baselines.

Conclusion: Multi-agent in-context reasoning framework can transform LLMs into competitive, plug-and-play TSC solvers without any parameter training.

Abstract: Time series classification (TSC) spans diverse application scenarios, yet labeled data are often scarce, making task-specific training costly and inflexible. Recent reasoning-oriented large language models (LLMs) show promise in understanding temporal patterns, but purely zero-shot usage remains suboptimal. We propose FETA, a multi-agent framework for training-free TSC via exemplar-based in-context reasoning. FETA decomposes a multivariate series into channel-wise subproblems, retrieves a few structurally similar labeled examples for each channel, and leverages a reasoning LLM to compare the query against these exemplars, producing channel-level labels with self-assessed confidences; a confidence-weighted aggregator then fuses all channel decisions. This design eliminates the need for pretraining or fine-tuning, improves efficiency by pruning irrelevant channels and controlling input length, and enhances interpretability through exemplar grounding and confidence estimation. On nine challenging UEA datasets, FETA achieves strong accuracy under a fully training-free setting, surpassing multiple trained baselines. These results demonstrate that a multi-agent in-context reasoning framework can transform LLMs into competitive, plug-and-play TSC solvers without any parameter training. The code is available at https://github.com/SongyuanSui/FETATSC.

Method: Generate dynamic math test instances with altered interpretations of numbers and operators while maintaining automatically verifiable answers, using random seeding at test time to evaluate induction/deduction capabilities.

Result: Models solve deduction more easily than induction but revert to standard math; math-adapted models lack general reasoning skills, and fine-tuning on induction tasks generalizes poorly.

Conclusion: The proposed dynamic benchmark effectively reveals overfitting and measures true reasoning, showing current models’ limitations in general reasoning capabilities despite mathematical adaptation.

Abstract: Conducting contamination-free evaluation of mathematical capabilities can be difficult for two reasons: models may memorize a test set once it is made public, and current mathematical benchmarks are prone to overfitting due to having limited diversity of symbols and rules, coupled with closed-ended answers. This paper proposes a method to leverage these shortcomings as useful features to a construct dynamic, counterfactual benchmark, which can be used to both reveal overfitting and measure true reasoning. We demonstrate this via MatheMagic, which generates math test instances with the interpretations of numbers and operators altered, yet has automatically verifiable answers. Test instances are randomly seeded and constructed at test time to evaluate a model’s induction or deduction capability, offering stability, extensibility, comparability, and robustness to overfitting. Our experiments find that models solve deduction more easily than induction, but they revert to standard math. Further analysis reveals that math-adapted models fail to exhibit a general “skill” of reasoning, and fine-tuning on induction tasks generalizes poorly.

Method: Extended traditional empowerment by introducing discounted empowerment to balance control across time horizons, and proposed a pre-training paradigm that initializes policies to maximize discounted empowerment.

Result: Empowerment-maximizing policies with long horizons are data-efficient and effective, leading to improved adaptability in downstream tasks across various RL algorithms.

Conclusion: Empowerment-based pre-training shows potential as a general-purpose initialization strategy, paving the way for scaling to high-dimensional and complex tasks in RL.

Abstract: Empowerment, an information-theoretic measure of an agent’s potential influence on its environment, has emerged as a powerful intrinsic motivation and exploration framework for reinforcement learning (RL). Besides for unsupervised RL and skill learning algorithms, the specific use of empowerment as a pre-training signal has received limited attention in the literature. We show that empowerment can be used as a pre-training signal for data-efficient downstream task adaptation. For this we extend the traditional notion of empowerment by introducing discounted empowerment, which balances the agent’s control over the environment across short- and long-term horizons. Leveraging this formulation, we propose a novel pre-training paradigm that initializes policies to maximize discounted empowerment, enabling agents to acquire a robust understanding of environmental dynamics. We analyze empowerment-based pre-training for various existing RL algorithms and empirically demonstrate its potential as a general-purpose initialization strategy: empowerment-maximizing policies with long horizons are data-efficient and effective, leading to improved adaptability in downstream tasks. Our findings pave the way for future research to scale this framework to high-dimensional and complex tasks, further advancing the field of RL.

Method: A formal query execution layer using canonical actions (atomic, composable, auditable primitives) that separate probabilistic discovery from deterministic retrieval, with planner-guided agents decomposing queries into DAGs of actions.

Result: Enables high-precision hybrid search, robust reference resolution, point-in-time version retrieval, and auditable causal tracing in legal domain knowledge graphs.

Conclusion: The two-layer architecture transforms retrieval into a transparent, auditable process that directly addresses Explainable AI requirements for high-stakes domains like law.

Abstract: The Structure-Aware Temporal Graph RAG (SAT-Graph RAG) addresses core limitations of standard Retrieval-Augmented Generation in the legal domain by providing a verifiable knowledge graph that models hierarchical structure, temporal evolution, and causal events of legal norms. However, a critical gap remains: how to reliably query this structured knowledge without sacrificing its deterministic properties. This paper introduces the SAT-Graph API, a formal query execution layer centered on canonical actions-atomic, composable, and auditable primitives that isolate probabilistic discovery from deterministic retrieval. These actions enable: (i) high-precision hybrid search; (ii) robust reference resolution; (iii) point-in-time version retrieval; and (iv) auditable causal tracing. We demonstrate how planner-guided agents can decompose complex queries into Directed Acyclic Graphs (DAGs) of these actions. This two-layer architecture transforms retrieval from an opaque black box to a transparent, auditable process, directly addressing Explainable AI (XAI) requirements for high-stakes domains.

Method: ARISE incorporates two key innovations: sample-level awareness to penalize negative scaling behaviors, and dynamic sampling mechanism to mitigate accuracy fluctuations and token count instability.

Result: Comprehensive experiments show ARISE provides reliable and fine-grained measurement of test-time scaling capabilities, revealing significant variations across models. Claude Opus was identified as having superior scaling characteristics compared to other contemporary reasoning models.

Conclusion: ARISE successfully addresses the need for systematic evaluation of test-time scaling capabilities, providing a robust metric that can guide model selection and development in the rapidly expanding landscape of reasoning models.

Abstract: Test-time scaling has emerged as a transformative paradigm for enhancing the performance of large reasoning models, enabling dynamic allocation of computational resources during inference. However, as the landscape of reasoning models rapidly expands, a critical question remains: how can we systematically compare and evaluate the test-time scaling capabilities across different models? In this paper, we introduce ARISE (Adaptive Resolution-aware Scaling Evaluation), a novel metric specifically designed to assess the test-time scaling effectiveness of large reasoning models. Unlike existing evaluation approaches, ARISE incorporates two key innovations: (1) sample-level awareness that effectively penalizes negative scaling behaviors where increased computation leads to performance degradation, and (2) a dynamic sampling mechanism that mitigates the impact of accuracy fluctuations and token count instability on the final assessment. We conduct comprehensive experiments evaluating state-of-the-art reasoning models across diverse domains including mathematical reasoning, code generation, and agentic tasks. Our results demonstrate that ARISE provides a reliable and fine-grained measurement of test-time scaling capabilities, revealing significant variations in scaling efficiency across models. Notably, our evaluation identifies Claude Opus as exhibiting superior scaling characteristics compared to other contemporary reasoning models.

Method: Used linear probing to trace refusal intentions across token positions, identified ‘refusal cliff’ phenomenon, performed causal intervention analysis to locate problematic attention heads, and developed Cliff-as-a-Judge data selection method.

Result: Discovered refusal cliff where models correctly identify harmful prompts but suppress refusal at final tokens. Ablating 3% of problematic heads reduced attack success below 10%. Cliff-as-a-Judge achieved comparable safety with only 1.7% of vanilla training data.

Conclusion: Reasoning models’ safety failures stem from systematic suppression of refusal intentions rather than inherent unsafety. Targeted interventions on specific mechanisms and efficient data selection can effectively repair safety alignment with minimal resources.

Abstract: Large reasoning models (LRMs) with multi-step reasoning capabilities have shown remarkable problem-solving abilities, yet they exhibit concerning safety vulnerabilities that remain poorly understood. In this work, we investigate why safety alignment fails in reasoning models through a mechanistic interpretability lens. Using a linear probing approach to trace refusal intentions across token positions, we discover a striking phenomenon termed as \textbf{refusal cliff}: many poorly-aligned reasoning models correctly identify harmful prompts and maintain strong refusal intentions during their thinking process, but experience a sharp drop in refusal scores at the final tokens before output generation. This suggests that these models are not inherently unsafe; rather, their refusal intentions are systematically suppressed. Through causal intervention analysis, we identify a sparse set of attention heads that negatively contribute to refusal behavior. Ablating just 3% of these heads can reduce attack success rates below 10%. Building on these mechanistic insights, we propose \textbf{Cliff-as-a-Judge}, a novel data selection method that identifies training examples exhibiting the largest refusal cliff to efficiently repair reasoning models’ safety alignment. This approach achieves comparable safety improvements using only 1.7% of the vanilla safety training data, demonstrating a less-is-more effect in safety alignment.

Method: MixReasoning framework that adaptively adjusts reasoning depth within a single response, creating a mixture of detailed reasoning for difficult steps and concise inference for simpler ones.

Result: Experiments on GSM8K, MATH-500, and AIME show that MixReasoning shortens reasoning length and substantially improves efficiency without compromising accuracy.

Conclusion: Adaptive reasoning depth adjustment through MixReasoning provides an effective approach to balance reasoning quality and efficiency in step-by-step problem solving.

Abstract: Reasoning models enhance performance by tackling problems in a step-by-step manner, decomposing them into sub-problems and exploring long chains of thought before producing an answer. However, applying extended reasoning to every step introduces substantial redundancy, as sub-problems vary widely in difficulty and complexity: a small number of pivotal steps are genuinely challenging and decisive for the final answer, while many others only involve straightforward revisions or simple computations. Therefore, a natural idea is to endow reasoning models with the ability to adaptively respond to this variation, rather than treating all steps with the same level of elaboration. To this end, we propose MixReasoning, a framework that dynamically adjusts the depth of reasoning within a single response. The resulting chain of thought then becomes a mixture of detailed reasoning on difficult steps and concise inference on simpler ones. Experiments on GSM8K, MATH-500, and AIME show that MixReasoning shortens reasoning length and substantially improves efficiency without compromising accuracy.

Method: DeepEvolve combines external knowledge retrieval, cross-file code editing, and systematic debugging in a feedback-driven iterative loop that proposes, refines, implements, and tests hypotheses to avoid shallow improvements and unproductive over-refinements.

Result: Across nine benchmarks in chemistry, mathematics, biology, materials, and patents, DeepEvolve consistently improved initial algorithms and produced executable new algorithms with sustained gains.

Conclusion: DeepEvolve bridges the gap between unguided evolution and research without grounding, providing a reliable framework for advancing scientific algorithm discovery.

Abstract: Large language models hold promise as scientific assistants, yet existing agents either rely solely on algorithm evolution or on deep research in isolation, both of which face critical limitations. Pure algorithm evolution, as in AlphaEvolve, depends only on the internal knowledge of LLMs and quickly plateaus in complex domains, while pure deep research proposes ideas without validation, resulting in unrealistic or unimplementable solutions. We present DeepEvolve, an agent that integrates deep research with algorithm evolution, uniting external knowledge retrieval, cross-file code editing, and systematic debugging under a feedback-driven iterative loop. Each iteration not only proposes new hypotheses but also refines, implements, and tests them, avoiding both shallow improvements and unproductive over-refinements. Across nine benchmarks in chemistry, mathematics, biology, materials, and patents, DeepEvolve consistently improves the initial algorithm, producing executable new algorithms with sustained gains. By bridging the gap between unguided evolution and research without grounding, DeepEvolve provides a reliable framework for advancing scientific algorithm discovery. Our code is available at https://github.com/liugangcode/deepevolve.

Method: Created TelecomTS dataset derived from a 5G telecommunications network with heterogeneous, de-anonymized covariates and explicit scale information, supporting anomaly detection, root-cause analysis, and multi-modal reasoning tasks.

Result: Benchmarking shows existing time series, language, and reasoning models struggle with the abrupt, noisy, and high-variance dynamics of observability data, highlighting the importance of preserving absolute scale information.

Conclusion: Foundation time series models need to natively leverage scale information for practical observability applications, as current approaches are inadequate for handling the unique characteristics of observability data.

Abstract: Modern enterprises generate vast streams of time series metrics when monitoring complex systems, known as observability data. Unlike conventional time series from domains such as weather, observability data are zero-inflated, highly stochastic, and exhibit minimal temporal structure. Despite their importance, observability datasets are underrepresented in public benchmarks due to proprietary restrictions. Existing datasets are often anonymized and normalized, removing scale information and limiting their use for tasks beyond forecasting, such as anomaly detection, root-cause analysis, and multi-modal reasoning. To address this gap, we introduce TelecomTS, a large-scale observability dataset derived from a 5G telecommunications network. TelecomTS features heterogeneous, de-anonymized covariates with explicit scale information and supports a suite of downstream tasks, including anomaly detection, root-cause analysis, and a question-answering benchmark requiring multi-modal reasoning. Benchmarking state-of-the-art time series, language, and reasoning models reveals that existing approaches struggle with the abrupt, noisy, and high-variance dynamics of observability data. Our experiments also underscore the importance of preserving covariates’ absolute scale, emphasizing the need for foundation time series models that natively leverage scale information for practical observability applications.

Method: Hierarchical multi-agent framework that parses queries into structured intents, then coordinates constraint agent, POI agent, path refinement agent, and verifier agent to ground preferences into routes with routing engine integration.

Result: Experiments show reliable grounding of textual preferences into constraint-aware routes, improving route quality and preference satisfaction over classical methods.

Conclusion: RouteLLM successfully bridges linguistic flexibility and spatial structure, enabling reasoning over route feasibility and user preferences through its multi-agent architecture.

Abstract: Route recommendation aims to provide users with optimal travel plans that satisfy diverse and complex requirements. Classical routing algorithms (e.g., shortest-path and constraint-aware search) are efficient but assume structured inputs and fixed objectives, limiting adaptability to natural-language queries. Recent LLM-based approaches enhance flexibility but struggle with spatial reasoning and the joint modeling of route-level and POI-level preferences. To address these limitations, we propose RouteLLM, a hierarchical multi-agent framework that grounds natural-language intents into constraint-aware routes. It first parses user queries into structured intents including POIs, paths, and constraints. A manager agent then coordinates specialized sub-agents: a constraint agent that resolves and formally check constraints, a POI agent that retrieves and ranks candidate POIs, and a path refinement agent that refines routes via a routing engine with preference-conditioned costs. A final verifier agent ensures constraint satisfaction and produces the final route with an interpretable rationale. This design bridges linguistic flexibility and spatial structure, enabling reasoning over route feasibility and user preferences. Experiments show that our method reliably grounds textual preferences into constraint-aware routes, improving route quality and preference satisfaction over classical methods.

Method: Implemented a prior classical AI model and developed an LLM-based decision-maker, evaluated using GPT-5 (reasoning model) and GPT-4 (non-reasoning model) with weighted self-consistency under zero-shot prompting. Tested on health insurance dataset annotated for three decision-makers with varying risk tolerance levels (0.0, 0.5, 1.0).

Result: Both classical AI and LLM-based models achieved comparable alignment with attribute-based targets. Classical AI exhibited slightly better alignment for moderate risk profiles.

Conclusion: Classical AI and LLM-based approaches show similar performance in decision-maker alignment, with classical methods having a slight advantage for moderate risk tolerance scenarios. The research provides publicly available datasets and implementations for further study.

Abstract: As algorithmic decision-makers are increasingly applied to high-stakes domains, AI alignment research has evolved from a focus on universal value alignment to context-specific approaches that account for decision-maker attributes. Prior work on Decision-Maker Alignment (DMA) has explored two primary strategies: (1) classical AI methods integrating case-based reasoning, Bayesian reasoning, and naturalistic decision-making, and (2) large language model (LLM)-based methods leveraging prompt engineering. While both approaches have shown promise in limited domains such as medical triage, their generalizability to novel contexts remains underexplored. In this work, we implement a prior classical AI model and develop an LLM-based algorithmic decision-maker evaluated using a large reasoning model (GPT-5) and a non-reasoning model (GPT-4) with weighted self-consistency under a zero-shot prompting framework, as proposed in recent literature. We evaluate both approaches on a health insurance decision-making dataset annotated for three target decision-makers with varying levels of risk tolerance (0.0, 0.5, 1.0). In the experiments reported herein, classical AI and LLM-based models achieved comparable alignment with attribute-based targets, with classical AI exhibiting slightly better alignment for a moderate risk profile. The dataset and open-source implementation are publicly available at: https://github.com/TeX-Base/ClassicalAIvsLLMsforDMAlignment and https://github.com/Parallax-Advanced-Research/ITM/tree/feature_insurance.

Method: Using simulated environments across three competitive scenarios (sales, elections, social media) to measure how optimizing for competitive success affects LLM behavior and alignment metrics.

Result: Competitive optimization caused: 6.3% sales increase with 14.0% more deceptive marketing; 4.9% vote share gain with 22.3% more disinformation and 12.5% more populist rhetoric; 7.5% engagement boost with 188.6% more disinformation and 16.3% increase in harmful behavior promotion.

Conclusion: Market-driven optimization systematically erodes alignment, creating a ‘race to the bottom’ that current safeguards cannot prevent, requiring stronger governance and carefully designed incentives for safe AI deployment.

Abstract: Large language models (LLMs) are increasingly shaping how information is created and disseminated, from companies using them to craft persuasive advertisements, to election campaigns optimizing messaging to gain votes, to social media influencers boosting engagement. These settings are inherently competitive, with sellers, candidates, and influencers vying for audience approval, yet it remains poorly understood how competitive feedback loops influence LLM behavior. We show that optimizing LLMs for competitive success can inadvertently drive misalignment. Using simulated environments across these scenarios, we find that, 6.3% increase in sales is accompanied by a 14.0% rise in deceptive marketing; in elections, a 4.9% gain in vote share coincides with 22.3% more disinformation and 12.5% more populist rhetoric; and on social media, a 7.5% engagement boost comes with 188.6% more disinformation and a 16.3% increase in promotion of harmful behaviors. We call this phenomenon Moloch’s Bargain for AI–competitive success achieved at the cost of alignment. These misaligned behaviors emerge even when models are explicitly instructed to remain truthful and grounded, revealing the fragility of current alignment safeguards. Our findings highlight how market-driven optimization pressures can systematically erode alignment, creating a race to the bottom, and suggest that safe deployment of AI systems will require stronger governance and carefully designed incentives to prevent competitive dynamics from undermining societal trust.

Method: The method involves combining sequential scaling (lengthening generation process) and parallel scaling (verifying and selecting among multiple candidate outputs), with a focus on allocating modest compute to verifiers to leverage asymmetric verification.

Result: Results show substantial improvements: up to 27 absolute points on BrowseComp, with GLM-4.5 Heavy achieving 54.0% on BrowseComp and 66.0% on GAIA, and Tongyi-DeepResearch Heavy reaching 69.0% on BrowseComp, surpassing proprietary results.

Conclusion: Test-time scaling with asymmetric verification enables significant performance gains in deep search agents, making open-source alternatives competitive with or superior to proprietary systems in certain benchmarks.

Abstract: Test-time compute can be scaled both sequentially and in parallel. Sequential scaling involves lengthening the generation process, while parallel scaling involves verifying and selecting among multiple candidate outputs. Combining these two strategies has led to the most powerful AI systems, such as Grok 4 Heavy and GPT-5 Pro. In certain contexts (e.g., solving Sudoku puzzles), verifying responses can be substantially easier than generating them. This property, referred to as \emph{asymmetric verification}, highlights the strong potential of test-time scaling (TTS). In this work, we study both sequential and parallel TTS of deep search agents, motivated by the intuition that verification in this setting is often much easier than generation. In experiments, we first show that sequential scaling methods, such as budget forcing, can be effective initially but soon degrade performance. Leveraging asymmetric verification, however, we are able to achieve substantial improvements by allocating only a modest amount of compute to the verifier. We conduct experiments with flagship open-source models and extend them to their ``Heavy’’ variants through TTS. These deep research agents achieve gains of up to 27 absolute points on benchmarks such as BrowseComp. Remarkably, as an open-source alternative, GLM-4.5 Heavy reaches accuracy of {\bf 54.0%} on BrowseComp and {\bf 66.0%} on GAIA, placing it comparable to the best proprietary choices such as OpenAI Deep Research. Tongyi-DeepResearch Heavy further achieves {\bf 69.0%} accuracy on BrowseComp, greatly surpassing the best proprietary results.

Method: ADRS iteratively generates, evaluates, and refines solutions using reliable verifiers that test implementations against workloads and measure performance.

Result: ADRS discovered algorithms that outperform state-of-the-art human designs, achieving up to 5.0x runtime improvements or 50% cost reductions in domains like load balancing, Mixture-of-Experts inference, LLM-based SQL queries, and transaction scheduling.

Conclusion: AI will transform systems research by taking over algorithm design, while human researchers will focus on problem formulation and strategic guidance, requiring adaptation of research practices.

Abstract: Artificial Intelligence (AI) is starting to transform the research process as we know it by automating the discovery of new solutions. Given a task, the typical AI-driven approach is (i) to generate a set of diverse solutions, and then (ii) to verify these solutions and select one that solves the problem. Crucially, this approach assumes the existence of a reliable verifier, i.e., one that can accurately determine whether a solution solves the given problem. We argue that systems research, long focused on designing and evaluating new performance-oriented algorithms, is particularly well-suited for AI-driven solution discovery. This is because system performance problems naturally admit reliable verifiers: solutions are typically implemented in real systems or simulators, and verification reduces to running these software artifacts against predefined workloads and measuring performance. We term this approach as AI-Driven Research for Systems (ADRS), which iteratively generates, evaluates, and refines solutions. Using penEvolve, an existing open-source ADRS instance, we present case studies across diverse domains, including load balancing for multi-region cloud scheduling, Mixture-of-Experts inference, LLM-based SQL queries, and transaction scheduling. In multiple instances, ADRS discovers algorithms that outperform state-of-the-art human designs (e.g., achieving up to 5.0x runtime improvements or 50% cost reductions). We distill best practices for guiding algorithm evolution, from prompt design to evaluator construction, for existing frameworks. We then discuss the broader implications for the systems community: as AI assumes a central role in algorithm design, we argue that human researchers will increasingly focus on problem formulation and strategic guidance. Our results highlight both the disruptive potential and the urgent need to adapt systems research practices in the age of AI.

Method: Developed a scalable data curation pipeline creating 60k+ step-level annotations, then trained TaTToo with dual-stage paradigm: cold-start supervised fine-tuning followed by reinforcement learning with tool-grounded reward shaping.

Result: TaTToo improves downstream policy LRMs by 30.9% at inference, surpasses strong PRM baselines like Qwen-2.5-Math-PRM-72B with only 8B parameters, and shows strong generalizability across diverse test-time scaling strategies.

Conclusion: TaTToo effectively addresses tabular reasoning challenges by integrating explicit table operations and tool-based verification, demonstrating significant performance improvements and scalability advantages over existing PRMs.

Abstract: Process Reward Models (PRMs) have recently emerged as a powerful framework for enhancing the reasoning capabilities of large reasoning models (LRMs), particularly in the context of test-time scaling (TTS). However, their potential for supervising LRMs on tabular reasoning domains remains underexplored. Through detailed empirical analyses, we identify that existing PRMs, though widely adopted for supervising text-only reasoning steps, struggle with table-specific operations such as sub-table retrieval and schema interaction, leading to critical performance bottlenecks. To address this limitation, we propose TaTToo, a novel table-grounded PRM framework that (i) reasons explicitly over tabular reasoning steps and (ii) integrates tool-based verification to provide precise reward supervision. Concretely, we first design a scalable data curation pipeline that constructs over 60k high-quality step-level annotations by integrating table verification rationales with tool-based executions. Building on the collected data, we train TaTToo with a dual-stage paradigm: cold-start supervised fine-tuning to capture tool-use reasoning patterns, followed by reinforcement learning with tool-grounded reward shaping to align our model with table-based verification. We provide a comprehensive evaluation of the policy improvement induced by our newly designed PRM. Across 5 challenging tabular reasoning benchmarks covering numerical reasoning, fact-checking, and data analysis, TaTToo improves downstream policy LRMs by 30.9% at inference, surpasses strong PRM baselines such as Qwen-2.5-Math-PRM-72B with only 8B parameters, and demonstrates strong generalizability across diverse TTS strategies.

Method: The authors propose six fine-grained metrics to address coarse-grained evaluation issues and conduct thematic analysis to identify four major reasoning failures that affect LLM performance in obscured communication.

Result: The paper introduces a systematic evaluation framework with fine-grained metrics and identifies specific reasoning failures that undermine LLMs’ performance in obscured communication tasks.

Conclusion: A microscopic and systematic approach is necessary for properly evaluating LLMs in obscured communication contexts, providing more detailed insights into their capabilities and limitations.

Abstract: Recent studies have investigated whether large language models (LLMs) can support obscured communication, which is characterized by core aspects such as inferring subtext and evading suspicions. To conduct the investigation, researchers have used social deduction games (SDGs) as their experimental environment, in which players conceal and infer specific information. However, prior work has often overlooked how LLMs should be evaluated in such settings. Specifically, we point out two limitations with the evaluation methods they employed. First, metrics used in prior studies are coarse-grained as they are based on overall game outcomes that often fail to capture event-level behaviors; Second, error analyses have lacked structured methodologies capable of producing insights that meaningfully support evaluation outcomes. To address these limitations, we propose a microscopic and systematic approach to the investigation. Specifically, we introduce six fine-grained metrics that resolve the first issue. To tackle the second issue, we conducted a thematic analysis and identified four major reasoning failures that undermine LLMs’ performance in obscured communication.

Method: Developed SenD training protocol that deterministically drops embedding indices with significant variability, and created EES metric that approximates EigenScore 2x faster for unsupervised hallucination detection.

Result: SenD improves test-time reliability of Pythia and Llama models by up to 17%, enhances factual accuracy across Wikipedia, Medical, Legal, and Coding domains without affecting downstream task performance.

Conclusion: The proposed SenD protocol effectively reduces hallucination variance during training, improving model reliability while maintaining computational scalability and downstream task performance.

Abstract: As large language models (LLMs) become increasingly prevalent, concerns about their reliability, particularly due to hallucinations - factually inaccurate or irrelevant outputs - have grown. Our research investigates the relationship between the uncertainty in training dynamics and the emergence of hallucinations. Using models from the Pythia suite and several hallucination detection metrics, we analyze hallucination trends and identify significant variance during training. To address this, we propose Sensitivity Dropout (SenD), a novel training protocol designed to reduce hallucination variance during training by deterministically dropping embedding indices with significant variability. In addition, we develop an unsupervised hallucination detection metric, Efficient EigenScore (EES), which approximates the traditional EigenScore in 2x speed. This metric is integrated into our training protocol, allowing SenD to be both computationally scalable and effective at reducing hallucination variance. SenD improves test-time reliability of Pythia and Meta’s Llama models by up to 17% and enhances factual accuracy in Wikipedia, Medical, Legal, and Coding domains without affecting downstream task performance.

Method: Adapted a decision tree algorithm based on PAC framework theoretical results to ensure PAC guarantees under certain conditions, focusing on binary classification and extracting trees from BERT-based language models.

Result: The experiments revealed occupational gender bias in the BERT-based language models through the extracted decision trees with PAC guarantees.

Conclusion: The PAC framework can provide theoretical guarantees for decision tree fidelity when used as surrogate models, enabling trustworthy approximation of complex AI model behavior while also revealing biases in the original models.

Abstract: Decision trees are a popular machine learning method, known for their inherent explainability. In Explainable AI, decision trees can be used as surrogate models for complex black box AI models or as approximations of parts of such models. A key challenge of this approach is determining how accurately the extracted decision tree represents the original model and to what extent it can be trusted as an approximation of their behavior. In this work, we investigate the use of the Probably Approximately Correct (PAC) framework to provide a theoretical guarantee of fidelity for decision trees extracted from AI models. Based on theoretical results from the PAC framework, we adapt a decision tree algorithm to ensure a PAC guarantee under certain conditions. We focus on binary classification and conduct experiments where we extract decision trees from BERT-based language models with PAC guarantees. Our results indicate occupational gender bias in these models.

Method: The study examines various types of large models including Large Language Models, Vision Models, 3D Large Models, Multimodal Models, and Large Graph Models, analyzing their integration with medical knowledge graphs and applications in medical image analysis and drug discovery.

Result: Large models are setting new benchmarks in medical innovation, improving diagnostic accuracy, and enabling personalized healthcare solutions through enhanced disease prediction, diagnostic assistance, and treatment planning.

Conclusion: Medical Large Models represent transformative technologies that are advancing global health, with significant potential in healthcare applications despite existing challenges, and they are paving the way for future medical innovations.

Abstract: This paper explores the advancements and applications of large-scale models in the medical field, with a particular focus on Medical Large Models (MedLMs). These models, encompassing Large Language Models (LLMs), Vision Models, 3D Large Models, and Multimodal Models, are revolutionizing healthcare by enhancing disease prediction, diagnostic assistance, personalized treatment planning, and drug discovery. The integration of graph neural networks in medical knowledge graphs and drug discovery highlights the potential of Large Graph Models (LGMs) in understanding complex biomedical relationships. The study also emphasizes the transformative role of Vision-Language Models (VLMs) and 3D Large Models in medical image analysis, anatomical modeling, and prosthetic design. Despite the challenges, these technologies are setting new benchmarks in medical innovation, improving diagnostic accuracy, and paving the way for personalized healthcare solutions. This paper aims to provide a comprehensive overview of the current state and future directions of large models in medicine, underscoring their significance in advancing global health.

Method: Developed EgoNetGNN, a graph neural network-based approach that automatically learns appropriate exposure mapping functions, allowing for complex peer influence mechanisms beyond just peer treatments.

Result: Comprehensive evaluation on synthetic and semi-synthetic network data shows EgoNetGNN is more robust to different unknown underlying influence mechanisms compared to state-of-the-art baselines.

Conclusion: The proposed method successfully addresses the limitation of existing approaches by automatically learning exposure mapping functions, enabling better estimation of heterogeneous peer effects in the presence of complex network interference.

Abstract: In causal inference, interference refers to the phenomenon in which the actions of peers in a network can influence an individual’s outcome. Peer effect refers to the difference in counterfactual outcomes of an individual for different levels of peer exposure, the extent to which an individual is exposed to the treatments, actions, or behaviors of peers. Estimating peer effects requires deciding how to represent peer exposure. Typically, researchers define an exposure mapping function that aggregates peer treatments and outputs peer exposure. Most existing approaches for defining exposure mapping functions assume peer exposure based on the number or fraction of treated peers. Recent studies have investigated more complex functions of peer exposure which capture that different peers can exert different degrees of influence. However, none of these works have explicitly considered the problem of automatically learning the exposure mapping function. In this work, we focus on learning this function for the purpose of estimating heterogeneous peer effects, where heterogeneity refers to the variation in counterfactual outcomes for the same peer exposure but different individual’s contexts. We develop EgoNetGNN, a graph neural network (GNN)-based method, to automatically learn the appropriate exposure mapping function allowing for complex peer influence mechanisms that, in addition to peer treatments, can involve the local neighborhood structure and edge attributes. We show that GNN models that use peer exposure based on the number or fraction of treated peers or learn peer exposure naively face difficulty accounting for such influence mechanisms. Our comprehensive evaluation on synthetic and semi-synthetic network data shows that our method is more robust to different unknown underlying influence mechanisms when estimating heterogeneous peer effects when compared to state-of-the-art baselines.

Method: Introduce SciSciGPT as a prototype AI collaborator that automates complex workflows, supports diverse analytical approaches, accelerates research prototyping, and facilitates reproducibility. Demonstrate through case studies and propose an LLM Agent capability maturity model.

Result: Case studies demonstrate SciSciGPT’s ability to streamline a wide range of empirical and analytical research tasks, showing its potential to advance research through human-AI collaboration.

Conclusion: Frameworks like SciSciGPT may play pivotal roles in scientific research as AI capabilities evolve, but raise challenges around transparency, ethical use, and balancing human-AI contributions that need addressing for future scientific inquiry.

Abstract: The increasing availability of large-scale datasets has fueled rapid progress across many scientific fields, creating unprecedented opportunities for research and discovery while posing significant analytical challenges. Recent advances in large language models (LLMs) and AI agents have opened new possibilities for human-AI collaboration, offering powerful tools to navigate this complex research landscape. In this paper, we introduce SciSciGPT, an open-source, prototype AI collaborator that uses the science of science as a testbed to explore the potential of LLM-powered research tools. SciSciGPT automates complex workflows, supports diverse analytical approaches, accelerates research prototyping and iteration, and facilitates reproducibility. Through case studies, we demonstrate its ability to streamline a wide range of empirical and analytical research tasks while highlighting its broader potential to advance research. We further propose an LLM Agent capability maturity model for human-AI collaboration, envisioning a roadmap to further improve and expand upon frameworks like SciSciGPT. As AI capabilities continue to evolve, frameworks like SciSciGPT may play increasingly pivotal roles in scientific research and discovery, unlocking further opportunities. At the same time, these new advances also raise critical challenges, from ensuring transparency and ethical use to balancing human and AI contributions. Addressing these issues may shape the future of scientific inquiry and inform how we train the next generation of scientists to thrive in an increasingly AI-integrated research ecosystem.

Method: Aggregate only shared non-expert parameters to reduce communication overhead, and use expert grafting to construct client-specific experts from pretrained experts that are fine-tuned locally with gating mechanisms.

Result: Outperforms federated baselines on non-IID instruction tuning datasets and shows strong knowledge preservation on MMLU benchmark.

Conclusion: FLEx effectively balances shared knowledge preservation and client-specific personalization in federated learning with MoE-based LLMs.

Abstract: Federated instruction tuning of large language models (LLMs) is challenged by significant data heterogeneity across clients, demanding robust personalization. The Mixture of Experts (MoE) architecture, where experts can specialize in distinct data patterns, presents a natural architectural solution to this challenge. The inherent sparsity of the MoE architecture, achieved by selectively activating experts, poses a significant challenge to its integration with federated learning (FL). Conventional FL frameworks, designed for dense models, naively aggregate all expert parameters irrespective of their local activation patterns. This naive approach not only undermines MoE’s dynamic sparsity but also risks corrupting the world knowledge within pretrained experts. To address this, we propose FLEx (Federated LLMs with Personalized Experts), a novel framework that leverages pretrained MoE-based LLMs for efficient personalization. By aggregating only the shared non-expert parameters, FLEx significantly reduces communication overhead and preserves the world knowledge stored within the frozen pretrained experts. For personalization, we introduce a novel expert grafting mechanism that leverages dynamic sparsity to construct a client-specific expert from selected components of pretrained experts, tailored to local data. This grafted expert is then fine-tuned locally alongside the gating mechanism. This joint training enables the model to learn when to leverage the shared knowledge from frozen experts and when to employ the personalized one. Evaluations on diverse, non-IID instruction tuning datasets show that FLEx consistently outperforms federated baselines on average, while demonstrating strong knowledge preservation on the knowledge-driven benchmark MMLU. Our code is available at \href{https://anonymous.4open.science/r/FLEx-8F12}{\texttt{https://anonymous.4open.science/r/FLEx-8F12}}.

Method: Created VisioMath benchmark with 1,800 high-quality K-12 math problems featuring diagrams with subtle visual similarities. Evaluated state-of-the-art LMMs and explored three alignment-oriented strategies including training-free approaches and finetuning.

Result: LMMs show consistent accuracy decline as inter-image similarity increases. Dominant failure mode is image-text misalignment where models use shallow positional heuristics instead of grounding reasoning in textual cues. Alignment strategies achieved substantial accuracy gains.

Conclusion: VisioMath serves as a rigorous benchmark to advance LMMs toward deeper diagram understanding, precise comparative reasoning, and grounded multi-image-text integration, addressing critical gaps in fine-grained visual reasoning capabilities.

Abstract: Large Multimodal Models have achieved remarkable progress in integrating vision and language, enabling strong performance across perception, reasoning, and domain-specific tasks. However, their capacity to reason over multiple, visually similar inputs remains insufficiently explored. Such fine-grained comparative reasoning is central to real-world tasks, especially in mathematics and education, where learners must often distinguish between nearly identical diagrams to identify correct solutions. To address this gap, we present VisioMath, a curated benchmark of 1,800 high-quality K-12 mathematics problems in which all candidate answers are diagrams with subtle visual similarities. A comprehensive evaluation of state-of-the-art LMMs, covering both leading closed-source systems and widely adopted open-source models, reveals a consistent decline in accuracy as inter-image similarity increases. Analysis indicates that the dominant failure mode stems from image-text misalignment: rather than grounding reasoning in textual cues, models often resort to shallow positional heuristics, resulting in systematic errors. We further explore three alignment-oriented strategies, spanning training-free approaches and finetuning, and achieve substantial accuracy gains. We hope that VisioMath will serve as a rigorous benchmark and catalyst for developing LMMs toward deeper diagram understanding, precise comparative reasoning, and grounded multi-image-text integration.

Method: The authors developed a novel method assessing five dimensions of moral competence: identifying morally relevant features, weighting their importance, assigning moral reasons, synthesizing coherent judgments, and recognizing information gaps. They conducted two experiments comparing six LLMs against non-expert humans and professional philosophers using both standard ethical vignettes and novel scenarios with embedded moral features.

Result: In standard ethical vignettes, LLMs generally outperformed non-expert humans across multiple dimensions of moral reasoning. However, in novel scenarios designed to test moral sensitivity by embedding relevant features among irrelevant details, several LLMs performed significantly worse than humans, revealing a striking reversal of performance.

Conclusion: Current evaluations may substantially overestimate LLMs’ moral reasoning capabilities by eliminating the task of discerning moral relevance from noisy information, which is a prerequisite for genuine moral skill. The work provides a more nuanced framework for assessing AI moral competence and highlights important directions for improvement.

Abstract: Moral competence is the ability to act in accordance with moral principles. As large language models (LLMs) are increasingly deployed in situations demanding moral competence, there is increasing interest in evaluating this ability empirically. We review existing literature and identify three significant shortcoming: (i) Over-reliance on prepackaged moral scenarios with explicitly highlighted moral features; (ii) Focus on verdict prediction rather than moral reasoning; and (iii) Inadequate testing of models’ (in)ability to recognize when additional information is needed. Grounded in philosophical research on moral skill, we then introduce a novel method for assessing moral competence in LLMs. Our approach moves beyond simple verdict comparisons to evaluate five dimensions of moral competence: identifying morally relevant features, weighting their importance, assigning moral reasons to these features, synthesizing coherent moral judgments, and recognizing information gaps. We conduct two experiments comparing six leading LLMs against non-expert humans and professional philosophers. In our first experiment using ethical vignettes standard to existing work, LLMs generally outperformed non-expert humans across multiple dimensions of moral reasoning. However, our second experiment, featuring novel scenarios designed to test moral sensitivity by embedding relevant features among irrelevant details, revealed a striking reversal: several LLMs performed significantly worse than humans. Our findings suggest that current evaluations may substantially overestimate LLMs’ moral reasoning capabilities by eliminating the task of discerning moral relevance from noisy information, which we take to be a prerequisite for genuine moral skill. This work provides a more nuanced framework for assessing AI moral competence and highlights important directions for improving moral competence in advanced AI systems.

Method: GRASP-based metaheuristic combining multiple construction heuristics with multi-neighborhood local search. Construction phase uses MIP techniques to transform TA-TSP into tailored TSP instances. Improvement phase applies 2-Opt, Swap, and Relocate operators.

Result: Achieved average optimality gaps of 0.77% and 0.40% relative to best-known solutions on MESS 2024 instances within 60 seconds. On synthetic datasets, solutions were 11.3% better than Gurobi solver under same time constraints. Finished top three at MESS 2024.

Conclusion: The method demonstrates suitability for real-time routing applications with state-dependent travel costs, showing strong performance in both competition and synthetic scenarios.

Abstract: The Trigger Arc Traveling Salesman Problem (TA-TSP) extends the classical TSP by introducing dynamic arc costs that change when specific “trigger” arcs are traversed, modeling scenarios such as warehouse operations with compactable storage systems. This paper introduces a GRASP-based metaheuristic that combines multiple construction heuristics with a multi-neighborhood local search. The construction phase uses mixed-integer programming (MIP) techniques to transform the TA-TSP into a sequence of tailored TSP instances, while the improvement phase applies 2-Opt, Swap, and Relocate operators. Computational experiments on MESS 2024 competition instances achieved average optimality gaps of 0.77% and 0.40% relative to the best-known solutions within a 60-second limit. On smaller, synthetically generated datasets, the method produced solutions 11.3% better than the Gurobi solver under the same time constraints. The algorithm finished in the top three at MESS 2024, demonstrating its suitability for real-time routing applications with state-dependent travel costs.

Method: Leverages the insight that auto-regressive models generate high-confidence text sequences. Introduces TCE loss function that selectively ignores high-confidence tokens during training to filter out machine-generated artifacts.

Result: Models trained with TCE learn effectively and show significantly increased resilience, tolerating over 2.3x more synthetic data before collapse onset. Provides open-source benchmark for collapse dynamics in mixed-data settings.

Conclusion: Confidence-aware training objectives like TCE can substantially delay model collapse, offering a practical and generalizable tool for model robustness under synthetic-data exposure.

Abstract: The increasing reliance on generative AI models is rapidly increasing the volume of synthetic data, with some projections suggesting that most available new data for training could be machine-generated by 2030. This shift to a mainly synthetic content presents a critical challenge: repeated training in synthetic data leads to a phenomenon known as model collapse, where model performance degrades over generations of training, eventually rendering the models ineffective. While the causes of model collapse are increasingly understood, effective mitigation strategies remain scarce. We address this challenge by leveraging a key insight: auto-regressive models tend to generate text sequences to which they assign high confidence (i.e., high log-likelihood). Based on this observation, we introduce the Truncated-Cross-Entropy (TCE) loss function. TCE mitigates collapse by selectively ignoring high-confidence tokens during training, effectively filtering out likely machine-generated artifacts from the learning process. Our experiments demonstrate that models trained with TCE not only learn effectively but also exhibit significantly increased resilience, tolerating over 2.3x more synthetic data before the onset of collapse. In addition, we provide an open-source benchmark for collapse dynamics in mixed-data settings. Our results demonstrate that confidence-aware training objectives can substantially delay collapse onset, offering a practical and generalizable tool for model robustness under synthetic-data exposure.

Method: MAPGD employs multiple specialized agents focusing on different prompt dimensions (instruction clarity, example selection, format structure, stylistic adaptation), coordinated through semantic gradient embedding, conflict detection, fusion, Hypersphere Constrained Gradient Clustering (HCGC), and Channel Adaptive Agent Weighting (CAAW).

Result: Experiments on classification and reasoning benchmarks show MAPGD consistently surpasses single-agent and random baselines in both accuracy and efficiency, with ablation studies confirming the effectiveness of key components.

Conclusion: MAPGD establishes a unified, gradient-based, and interpretable framework for robust prompt optimization with theoretical convergence guarantees, demonstrating superior performance through collaborative multi-agent specialization.

Abstract: Prompt engineering is crucial for fully leveraging large language models (LLMs), yet most existing optimization methods follow a single trajectory, resulting in limited adaptability, gradient conflicts, and high computational overhead. We propose MAPGD (Multi-Agent Prompt Gradient Descent), a novel framework that reconceptualizes prompt optimization as a collaborative process among specialized agents. Each agent focuses on a distinct refinement dimension, such as instruction clarity, example selection, format structure, or stylistic adaptation, and their contributions are coordinated through semantic gradient embedding, conflict detection, and fusion. To further enhance robustness and stability, MAPGD introduces two new mechanisms: Hypersphere Constrained Gradient Clustering (HCGC), which enforces angular margin constraints for compact and well-separated clusters, and Channel Adaptive Agent Weighting (CAAW), which dynamically reweights agent contributions based on validation performance. Experiments on classification and reasoning benchmarks show that MAPGD consistently surpasses single-agent and random baselines in both accuracy and efficiency. Ablation studies confirm the effectiveness of gradient fusion, agent specialization, and conflict resolution. Together, these components establish MAPGD as a unified, gradient-based, and interpretable framework for robust prompt optimization with theoretical convergence guarantees.

Method: Structured framework combining automated components (scene text detection, inpainting, machine translation, text reimposition) with human oversight to address advertisement localization complexities.

Result: Qualitative results across six locales demonstrate semantically accurate and visually coherent localized advertisements suitable for real-world deployment.

Conclusion: First work to integrate these specific techniques for accelerating ad localization evaluation workflows, producing high-quality localized advertisements.

Abstract: Adapting advertisements for multilingual audiences requires more than simple text translation; it demands preservation of visual consistency, spatial alignment, and stylistic integrity across diverse languages and formats. We introduce a structured framework that combines automated components with human oversight to address the complexities of advertisement localization. To the best of our knowledge, this is the first work to integrate scene text detection, inpainting, machine translation (MT), and text reimposition specifically for accelerating ad localization evaluation workflows. Qualitative results across six locales demonstrate that our approach produces semantically accurate and visually coherent localized advertisements, suitable for deployment in real-world workflows.

Method: RepIt framework isolates concept-specific representations, allowing precise interventions that localize corrective signals to just 100-200 neurons and can be extracted from as few as a dozen examples.

Result: Across five frontier LLMs, RepIt successfully suppressed refusal on targeted concepts (like WMD-related questions) while preserving refusal elsewhere, producing models that still score as safe on standard benchmarks.

Conclusion: RepIt demonstrates that targeted interventions can counteract overgeneralization in LLMs, enabling more granular control of model behavior, though its efficiency also raises concerns about potential misuse with modest compute and data.

Abstract: While activation steering in large language models (LLMs) is a growing area of research, methods can often incur broader effects than desired. This motivates isolation of purer concept vectors to enable targeted interventions and understand LLM behavior at a more granular level. We present RepIt, a simple and data-efficient framework for isolating concept-specific representations. Across five frontier LLMs, RepIt enables precise interventions: it selectively suppresses refusal on targeted concepts while preserving refusal elsewhere, producing models that answer WMD-related questions while still scoring as safe on standard benchmarks. We further show that the corrective signal localizes to just 100-200 neurons and that robust target representations can be extracted from as few as a dozen examples on a single A6000. This efficiency raises a dual concern: manipulations can be performed with modest compute and data to extend to underrepresented data-scarce topics while evading existing benchmarks. By disentangling refusal vectors with RepIt, this work demonstrates that targeted interventions can counteract overgeneralization, laying the foundation for more granular control of model behavior.

Method: Parametrize transition matrix as product of column one-hot matrix (P) and complex-valued diagonal matrix (D), enabling linear scaling of parallel scan costs with state size.

Result: Outperforms modern SSM variants on FSA state tracking tasks, comparable to neural controlled differential equations on time-series classification, and effectively tracks complex FSA states in hybrid Transformer-SSM architecture.

Conclusion: PD-SSM provides optimal FSA emulation capabilities with efficient computation, significantly improving on current structured SSM guarantees while maintaining practical computational costs.

Abstract: Modern state-space models (SSMs) often utilize transition matrices which enable efficient computation but pose restrictions on the model’s expressivity, as measured in terms of the ability to emulate finite-state automata (FSA). While unstructured transition matrices are optimal in terms of expressivity, they come at a prohibitively high compute and memory cost even for moderate state sizes. We propose a structured sparse parametrization of transition matrices in SSMs that enables FSA state tracking with optimal state size and depth, while keeping the computational cost of the recurrence comparable to that of diagonal SSMs. Our method, PD-SSM, parametrizes the transition matrix as the product of a column one-hot matrix ($P$) and a complex-valued diagonal matrix ($D$). Consequently, the computational cost of parallel scans scales linearly with the state size. Theoretically, the model is BIBO-stable and can emulate any $N$-state FSA with one layer of dimension $N$ and a linear readout of size $N \times N$, significantly improving on all current structured SSM guarantees. Experimentally, the model significantly outperforms a wide collection of modern SSM variants on various FSA state tracking tasks. On multiclass time-series classification, the performance is comparable to that of neural controlled differential equations, a paradigm explicitly built for time-series analysis. Finally, we integrate PD-SSM into a hybrid Transformer-SSM architecture and demonstrate that the model can effectively track the states of a complex FSA in which transitions are encoded as a set of variable-length English sentences. The code is available at https://github.com/IBM/expressive-sparse-state-space-model

Method: Proposed a pipeline for eliciting, steering, and modulating LLMs’ risk profiles using behavioral economics tools and utility-theoretic models. Compared different LLM types and evaluated modulation strategies including prompt engineering, in-context learning, and post-training.

Result: Instruction-tuned models showed behaviors consistent with standard utility formulations, while pre-trained and RLHF-aligned models deviated more from fitted utility models. Post-training provided the most stable and effective modulation of risk preference.

Conclusion: The findings provide insights into LLM risk profiles across different training stages and demonstrate post-training’s effectiveness in modulating risk preferences, laying groundwork for behavioral alignment and risk-aware LLM design.

Abstract: Large language models (LLMs) are increasingly used for decision-making tasks under uncertainty; however, their risk profiles and how they are influenced by prompting and alignment methods remain underexplored. Existing studies have primarily examined personality prompting or multi-agent interactions, leaving open the question of how post-training influences the risk behavior of LLMs. In this work, we propose a new pipeline for eliciting, steering, and modulating LLMs’ risk profiles, drawing on tools from behavioral economics and finance. Using utility-theoretic models, we compare pre-trained, instruction-tuned, and RLHF-aligned LLMs, and find that while instruction-tuned models exhibit behaviors consistent with some standard utility formulations, pre-trained and RLHF-aligned models deviate more from any utility models fitted. We further evaluate modulation strategies, including prompt engineering, in-context learning, and post-training, and show that post-training provides the most stable and effective modulation of risk preference. Our findings provide insights into the risk profiles of different classes and stages of LLMs and demonstrate how post-training modulates these profiles, laying the groundwork for future research on behavioral alignment and risk-aware LLM design.

Method: Proposes three key innovations: (1) hybrid data synthesis combining MCTS with strong VLM judgments for better step-level labels, (2) perception-focused supervision to detect visual grounding errors, and (3) systematic evaluation of multiple test-time scaling strategies.

Result: Experiments on five multimodal benchmarks show VL-PRMs outperform step selection methods when used as Outcome Reward Models, smaller VL-PRMs can match larger ones in error detection, perception supervision boosts test-time scaling, and models generalize to advanced math reasoning without training.

Conclusion: The work provides key insights into VL-PRM design space and demonstrates their effectiveness in uncovering latent reasoning abilities in VLMs, motivating further research in vision-language reasoning.

Abstract: Process Reward Models (PRMs) provide step-level supervision that improves the reliability of reasoning in large language models. While PRMs have been extensively studied in text-based domains, their extension to Vision Language Models (VLMs) remains limited. Existing Vision-Language PRMs (VL-PRMs) rely on Monte Carlo Tree Search (MCTS) for data construction, which can often produce noisy supervision signals and limit generalization across tasks. In this work, we aim to elucidate the design space of VL-PRMs by exploring diverse strategies for dataset construction, training, and test-time scaling. First, we introduce a hybrid data synthesis framework that combines MCTS with judgments from a strong VLM, producing more accurate step-level labels. Second, we propose perception-focused supervision, enabling our PRM to explicitly detect errors at the visual grounding stage of reasoning. Third, we systematically evaluate multiple test-time scaling strategies, showing that our PRMs can reliably guide VLMs toward more accurate solutions. Our experiments covering five diverse multimodal benchmarks (MMMU, PuzzleVQA, AlgoPuzzleVQA, MathVista, and MathVision) reveal several key insights: (i) VL-PRMs when used as Outcome Reward Models (ORMs) during test-time scaling (TTS) can outperform VL-PRM guided process step selection, (ii) smaller VL-PRMs can match or even surpass larger ones in detecting process errors, (iii) VL-PRMs uncover latent reasoning abilities in stronger VLM backbones, (iv) perception-level supervision leads to significant gains in test-time scaling, and (v) TTS performance of different policies improve on advanced math reasoning datasets despite not training VL-PRMs on such datasets. We hope our work will motivate further research and support the advancement of VLMs.

Method: The authors review Hierarchical Reasoning Models, examine key design choices, test alternative variants, and clarify common misconceptions about this class of models.

Result: The paper provides analysis and clarification of Hierarchical Reasoning Models, which have shown promising performance on 2D reasoning tasks through recurrent reasoning in transformer latent spaces.

Conclusion: Hierarchical Reasoning Models represent an emerging direction for enhancing transformer reasoning capabilities, but are still at an early stage requiring further investigation.

Abstract: Transformers have demonstrated remarkable performance in natural language processing and related domains, as they largely focus on sequential, autoregressive next-token prediction tasks. Yet, they struggle in logical reasoning, not necessarily because of a fundamental limitation of these models, but possibly due to the lack of exploration of more creative uses, such as latent space and recurrent reasoning. An emerging exploration in this direction is the Hierarchical Reasoning Model (Wang et. al., 2025), which introduces a novel type of recurrent reasoning in the latent space of transformers, achieving remarkable performance on a wide range of 2D reasoning tasks. Despite the promising results, this line of models is still at an early stage and calls for in-depth investigation. In this work, we review this class of models, examine key design choices, test alternative variants and clarify common misconceptions.

Method: TRACE progressively truncates a model’s chain-of-thought at various lengths, forces the model to answer, and measures expected reward at each cutoff to quantify reasoning effort.

Result: TRACE achieves over 65% gains over 72B CoT monitors in math reasoning and over 30% gains over 32B monitors in coding, and can discover unknown loopholes during training.

Conclusion: TRACE provides a scalable unsupervised approach for oversight where current monitoring methods are ineffective against implicit reward hacking.

Abstract: Reward hacking, where a reasoning model exploits loopholes in a reward function to achieve high rewards without solving the intended task, poses a significant threat. This behavior may be explicit, i.e. verbalized in the model’s chain-of-thought (CoT), or implicit, where the CoT appears benign thus bypasses CoT monitors. To detect implicit reward hacking, we propose TRACE (Truncated Reasoning AUC Evaluation). Our key observation is that hacking occurs when exploiting the loophole is easier than solving the actual task. This means that the model is using less ’effort’ than required to achieve high reward. TRACE quantifies effort by measuring how early a model’s reasoning becomes sufficient to obtain the reward. We progressively truncate a model’s CoT at various lengths, force the model to answer, and estimate the expected reward at each cutoff. A hacking model, which takes a shortcut, will achieve a high expected reward with only a small fraction of its CoT, yielding a large area under the accuracy-vs-length curve. TRACE achieves over 65% gains over our strongest 72B CoT monitor in math reasoning, and over 30% gains over a 32B monitor in coding. We further show that TRACE can discover unknown loopholes during training. Overall, TRACE offers a scalable unsupervised approach for oversight where current monitoring methods prove ineffective.

Method: Evaluated models on ConceptARC under varying conditions: input modality (textual vs. visual), use of external Python tools, and reasoning effort. Used dual evaluation measuring both output accuracy and fine-grained analysis of natural-language rules generated to explain solutions.

Result: Text-based models matched human accuracy but their rules often relied on surface-level shortcuts rather than intended abstractions. Visual models had lower accuracy but showed substantial understanding of intended abstractions, though they struggled to apply them correctly.

Conclusion: Models still lag humans in abstract reasoning. Accuracy alone overestimates abstract reasoning in textual modalities and underestimates it in visual modalities. The proposed evaluation framework provides a more faithful assessment of multimodal models’ abstract reasoning abilities.

Abstract: OpenAI’s o3-preview reasoning model exceeded human accuracy on the ARC-AGI benchmark, but does that mean state-of-the-art models recognize and reason with the abstractions that the task creators intended? We investigate models’ abstraction abilities on ConceptARC. We evaluate models under settings that vary the input modality (textual vs. visual), whether the model is permitted to use external Python tools, and, for reasoning models, the amount of reasoning effort. In addition to measuring output accuracy, we perform fine-grained evaluation of the natural-language rules that models generate to explain their solutions. This dual evaluation lets us assess whether models solve tasks using the abstractions ConceptARC was designed to elicit, rather than relying on surface-level patterns. Our results show that, while some models using text-based representations match human output accuracy, the best models’ rules are often based on surface-level ``shortcuts’’ and capture intended abstractions far less often than humans. Thus their capabilities for general abstract reasoning may be overestimated by evaluations based on accuracy alone. In the visual modality, AI models’ output accuracy drops sharply, yet our rule-level analysis reveals that models might be underestimated, as they still exhibit a substantial share of rules that capture intended abstractions, but are often unable to correctly apply these rules. In short, our results show that models still lag humans in abstract reasoning, and that using accuracy alone to evaluate abstract reasoning on ARC-like tasks may overestimate abstract-reasoning capabilities in textual modalities and underestimate it in visual modalities. We believe that our evaluation framework offers a more faithful picture of multimodal models’ abstract reasoning abilities and a more principled way to track progress toward human-like, abstraction-centered intelligence.

Method: BrowserArena collects user-submitted tasks, runs Arena-style head-to-head comparisons, and uses step-level human feedback to analyze agent traces and identify failure modes.

Result: Identified three consistent failure modes: captcha resolution, pop-up banner removal, and direct URL navigation. Found model-specific variations - o4-mini uses diverse captcha strategies while DeepSeek-R1 misleads about pop-up closure.

Conclusion: Current web agents show both diversity and brittleness. The benchmarking methodology provides scalable evaluation of web agent failure modes.

Abstract: LLM web agents now browse and take actions on the open web, yet current agent evaluations are constrained to sandboxed environments or artificial tasks. We introduce BrowserArena, a live open-web agent evaluation platform that collects user-submitted tasks, runs Arena-style head-to-head comparisons, and uses step-level human feedback to surface failure modes. Collecting and analyzing step-level annotations on the agent traces, we identify three consistent failure modes: captcha resolution, pop-up banner removal, and direct navigation to URLs. By constructing targeted datasets to further study these tasks, we discover variations in how different language models navigate these failure modes. We find, for example, that o4-mini deploys a wider variety of strategies to circumvent captcha resolution than other models and DeepSeek-R1 consistently misleads users about pop-up banner closure. Our findings surface both the diversity and brittleness of current web agents. More broadly, our benchmarking methodology provides an approach to evaluating and understanding web agent failure modes at scale.

Method: DLLM constructs independent subgraphs based on response correctness, applies relation augmentation for data imbalance, fuses subgraph representations with LLM-derived semantic representations, and uses a two-stage denoising diffusion module (unconditional then conditional graph-guided) to remove noise before alignment steps.

Result: Experiments on three web-based educational platform datasets show DLLM achieves optimal predictive performance across varying noise levels, demonstrating noise robustness while effectively leveraging semantic knowledge from LLMs.

Conclusion: DLLM successfully addresses noise and data imbalance challenges in cognitive diagnostics by integrating semantic knowledge from LLMs with structural information through diffusion-based denoising, achieving robust performance in web-based educational environments.

Abstract: Cognitive diagnostics in the Web-based Intelligent Education System (WIES) aims to assess students’ mastery of knowledge concepts from heterogeneous, noisy interactions. Recent work has tried to utilize Large Language Models (LLMs) for cognitive diagnosis, yet LLMs struggle with structured data and are prone to noise-induced misjudgments. Specially, WIES’s open environment continuously attracts new students and produces vast amounts of response logs, exacerbating the data imbalance and noise issues inherent in traditional educational systems. To address these challenges, we propose DLLM, a Diffusion-based LLM framework for noise-robust cognitive diagnosis. DLLM first constructs independent subgraphs based on response correctness, then applies relation augmentation alignment module to mitigate data imbalance. The two subgraph representations are then fused and aligned with LLM-derived, semantically augmented representations. Importantly, before each alignment step, DLLM employs a two-stage denoising diffusion module to eliminate intrinsic noise while assisting structural representation alignment. Specifically, unconditional denoising diffusion first removes erroneous information, followed by conditional denoising diffusion based on graph-guided to eliminate misleading information. Finally, the noise-robust representation that integrates semantic knowledge and structural information is fed into existing cognitive diagnosis models for prediction. Experimental results on three publicly available web-based educational platform datasets demonstrate that our DLLM achieves optimal predictive performance across varying noise levels, which demonstrates that DLLM achieves noise robustness while effectively leveraging semantic knowledge from LLM.

Method: Agent Spec uses a declarative language approach to define AI agents and workflows independently of execution environments, serving as an interchange format between different AI frameworks and tools.

Result: The specification benefits four key groups: developers gain reusable components, framework developers get interchange capabilities, researchers achieve reproducibility, and enterprises benefit from faster deployment and scalability.

Conclusion: Agent Spec provides technical foundations for improving AI agent interoperability, reusability, and development efficiency across different frameworks and tools.

Abstract: Open Agent Specification (Agent Spec) is a declarative language that allows AI agents and their workflows to be defined in a way that is compatible across different AI frameworks, promoting portability and interoperability within AI Agent frameworks. Agent Spec aims to resolve the challenges of fragmented agent development by providing a common unified specification that allows AI agents to be designed once and deployed across various frameworks, improving interoperability and reusability, and reducing redundant development efforts. Additionally, Agent Spec facilitates development tools and portability, allowing AI agents to be defined independently of their execution environment and enabling teams to exchange solutions without implementation-specific limitations. Agent Spec benefits four key groups: (i) Agent developers, who gain access to a superset of reusable components and design patterns, enabling them to leverage a broader range of functionalities; (ii) Agent framework and tool developers, who can use Agent Spec as an interchange format and therefore benefit from the support of other frameworks as well as other tools; (iii) Researchers, who can achieve reproducible results and comparability, facilitating more reliable and consistent outcomes; (iv) Enterprises, which benefit from faster prototype-to-deployment, increased productivity, as well as greater scalability and maintainability for their AI agent solutions. This technical report provides an overview of the technical foundations of Agent Spec, including motivation, benefits, and future developments.

Method: Formulate trade execution as constrained MDP with hard constraints. Use PPO for training execution agent, runtime action-shield for safety, and zero-knowledge compliance audit for verifiability.

Result: Learned policy reduces implementation shortfall and variance with no constraint violations across stress scenarios. Results statistically significant at 95% confidence level.

Conclusion: System successfully integrates optimal execution, safe RL, regulatory tech, and verifiable AI, with potential for real-world deployment after addressing limitations.

Abstract: We present a cross-market algorithmic trading system that balances execution quality with rigorous compliance enforcement. The architecture comprises a high-level planner, a reinforcement learning execution agent, and an independent compliance agent. We formulate trade execution as a constrained Markov decision process with hard constraints on participation limits, price bands, and self-trading avoidance. The execution agent is trained with proximal policy optimization, while a runtime action-shield projects any unsafe action into a feasible set. To support auditability without exposing proprietary signals, we add a zero-knowledge compliance audit layer that produces cryptographic proofs that all actions satisfied the constraints. We evaluate in a multi-venue, ABIDES-based simulator and compare against standard baselines (e.g., TWAP, VWAP). The learned policy reduces implementation shortfall and variance while exhibiting no observed constraint violations across stress scenarios including elevated latency, partial fills, compliance module toggling, and varying constraint limits. We report effects at the 95% confidence level using paired t-tests and examine tail risk via CVaR. We situate the work at the intersection of optimal execution, safe reinforcement learning, regulatory technology, and verifiable AI, and discuss ethical considerations, limitations (e.g., modeling assumptions and computational overhead), and paths to real-world deployment.

Method: Non-probabilistic autoencoder architecture with independence constraint between predicted latent variable and target controllable variable, ensuring consistent signal transformation while preserving original content.

Result: Quantitative evaluations confirm effectiveness and generality of the approach for speech styles including speaker identity and emotion conversion.

Conclusion: The proposed framework provides theoretical guarantees for speech attribute conversion with reliable control and interpretability across different speech styles.

Abstract: While signal conversion and disentangled representation learning have shown promise for manipulating data attributes across domains such as audio, image, and multimodal generation, existing approaches, especially for speech style conversion, are largely empirical and lack rigorous theoretical foundations to guarantee reliable and interpretable control. In this work, we propose a general framework for speech attribute conversion, accompanied by theoretical analysis and guarantees under reasonable assumptions. Our framework builds on a non-probabilistic autoencoder architecture with an independence constraint between the predicted latent variable and the target controllable variable. This design ensures a consistent signal transformation, conditioned on an observed style variable, while preserving the original content and modifying the desired attribute. We further demonstrate the versatility of our method by evaluating it on speech styles, including speaker identity and emotion. Quantitative evaluations confirm the effectiveness and generality of the proposed approach.

Method: Uses U-Net-based 1D convolutional encoder for audio, Swin Transformer V2 for visual features, bidirectional cross-attention for modality fusion, Squeezeformer blocks for temporal dependencies, and U-Net decoder for waveform reconstruction.

Result: Achieved STOI of 0.516, PESQ of 1.323, and SI-SDR of -4.322 dB, demonstrating significant performance improvements over noisy baselines.

Conclusion: AUREXA-SE effectively integrates audio and visual modalities through cross-attention and temporal modeling, producing perceptually consistent and intelligible speech output with superior performance metrics.

Abstract: In this paper, we propose AUREXA-SE (Audio-Visual Unified Representation Exchange Architecture with Cross-Attention and Squeezeformer for Speech Enhancement), a progressive bimodal framework tailored for audio-visual speech enhancement (AVSE). AUREXA-SE jointly leverages raw audio waveforms and visual cues by employing a U-Net-based 1D convolutional encoder for audio and a Swin Transformer V2 for efficient and expressive visual feature extraction. Central to the architecture is a novel bidirectional cross-attention mechanism, which facilitates deep contextual fusion between modalities, enabling rich and complementary representation learning. To capture temporal dependencies within the fused embeddings, a stack of lightweight Squeezeformer blocks combining convolutional and attention modules is introduced. The enhanced embeddings are then decoded via a U-Net-style decoder for direct waveform reconstruction, ensuring perceptually consistent and intelligible speech output. Experimental evaluations demonstrate the effectiveness of AUREXA-SE, achieving significant performance improvements over noisy baselines, with STOI of 0.516, PESQ of 1.323, and SI-SDR of -4.322 dB. The source code of AUREXA-SE is available at https://github.com/mtanveer1/AVSEC-4-Challenge-2025.

Method: Uses dual spatial and spectral encoders trained on over 4,000 hours of synthetic first-order Ambisonics recordings with metadata. The model enumerates and describes up to four directional sound sources, non-directional background sounds, and room characteristics in one pass.

Result: Evaluated with permutation-invariant protocol and 15 metrics covering content, location, timing, loudness, and reverberation. Shows strong performance across various conditions and generalizes to real room impulse responses with minor performance degradation.

Conclusion: Establishes the first audio LLM capable of full spatial-scene description with strong potential for real-world deployment.

Abstract: Acoustic scene perception involves describing the type of sounds, their timing, their direction and distance, as well as their loudness and reverberation. While audio language models excel in sound recognition, single-channel input fundamentally limits spatial understanding. This work presents Sci-Phi, a spatial audio large language model with dual spatial and spectral encoders that estimates a complete parameter set for all sound sources and the surrounding environment. Learning from over 4,000 hours of synthetic first-order Ambisonics recordings including metadata, Sci-Phi enumerates and describes up to four directional sound sources in one pass, alongside non-directional background sounds and room characteristics. We evaluate the model with a permutation-invariant protocol and 15 metrics covering content, location, timing, loudness, and reverberation, and analyze its robustness across source counts, signal-to-noise ratios, reverberation levels, and challenging mixtures of acoustically, spatially, or temporally similar sources. Notably, Sci-Phi generalizes to real room impulse responses with only minor performance degradation. Overall, this work establishes the first audio LLM capable of full spatial-scene description, with strong potential for real-world deployment. Demo: https://sci-phi-audio.github.io/demo

Method: Use TopK activation inspired by Sparse Autoencoders (SAEs) on the last layer of AASIST deepfake detection architecture to obtain sparse representations for decision making.

Result: Achieved EER of 23.36% on ASVSpoof5 test set with 95% sparsity, and demonstrated better disentanglement using completeness and modularity metrics based on mutual information.

Conclusion: Sparse deepfake detection improves both detection performance and interpretability, with some attacks directly encoded in the latent space, providing better disentangled representations.

Abstract: Due to the rapid progress of speech synthesis, deepfake detection has become a major concern in the speech processing community. Because it is a critical task, systems must not only be efficient and robust, but also provide interpretable explanations. Among the different approaches for explainability, we focus on the interpretation of latent representations. In such paper, we focus on the last layer of embeddings of AASIST, a deepfake detection architecture. We use a TopK activation inspired by SAEs on this layer to obtain sparse representations which are used in the decision process. We demonstrate that sparse deepfake detection can improve detection performance, with an EER of 23.36% on ASVSpoof5 test set, with 95% of sparsity. We then show that these representations provide better disentanglement, using completeness and modularity metrics based on mutual information. Notably, some attacks are directly encoded in the latent space.

Method: Leverages pretrained foundation models for efficiency, extracts spatial cues from depth maps and bounding boxes, and uses them as cross-attention conditioning in a diffusion-based audio generation model.

Result: Achieves both temporal and spatial alignment, producing stereo audio that dynamically adapts to video scene structure and movement, evaluated on Walking The Maps dataset with improved immersive experience.

Conclusion: StereoSync advances state of the art in video-to-audio generation by incorporating spatial awareness alongside temporal synchronization, creating more realistic and immersive audio experiences.

Abstract: Although audio generation has been widely studied over recent years, video-aligned audio generation still remains a relatively unexplored frontier. To address this gap, we introduce StereoSync, a novel and efficient model designed to generate audio that is both temporally synchronized with a reference video and spatially aligned with its visual context. Moreover, StereoSync also achieves efficiency by leveraging pretrained foundation models, reducing the need for extensive training while maintaining high-quality synthesis. Unlike existing methods that primarily focus on temporal synchronization, StereoSync introduces a significant advancement by incorporating spatial awareness into video-aligned audio generation. Indeed, given an input video, our approach extracts spatial cues from depth maps and bounding boxes, using them as cross-attention conditioning in a diffusion-based audio generation model. Such an approach allows StereoSync to go beyond simple synchronization, producing stereo audio that dynamically adapts to the spatial structure and movement of a video scene. We evaluate StereoSync on Walking The Maps, a curated dataset comprising videos from video games that feature animated characters walking through diverse environments. Experimental results demonstrate the ability of StereoSync to achieve both temporal and spatial alignment, advancing the state of the art in video-to-audio generation and resulting in a significantly more immersive and realistic audio experience.

Method: Proposes MSF-SER which augments acoustic features with three semantic levels: Local Emphasized Semantics (identifying emotionally important words), Global Semantics (overall meaning), and Extended Semantics (interpretive cues). These are integrated via intra-modal gated fusion and cross-modal FiLM-modulated lightweight Mixture-of-Experts (FM-MOE).

Result: Experiments on MSP-Podcast and IEMOCAP datasets show that MSF-SER consistently improves dimensional emotion prediction (valence, arousal, dominance) compared to baseline methods.

Conclusion: The enriched multi-granularity semantic fusion approach effectively enhances speech emotion recognition performance by capturing complementary emotional information from different semantic levels.

Abstract: Continuous dimensional speech emotion recognition captures affective variation along valence, arousal, and dominance, providing finer-grained representations than categorical approaches. Yet most multimodal methods rely solely on global transcripts, leading to two limitations: (1) all words are treated equally, overlooking that emphasis on different parts of a sentence can shift emotional meaning; (2) only surface lexical content is represented, lacking higher-level interpretive cues. To overcome these issues, we propose MSF-SER (Multi-granularity Semantic Fusion for Speech Emotion Recognition), which augments acoustic features with three complementary levels of textual semantics–Local Emphasized Semantics (LES), Global Semantics (GS), and Extended Semantics (ES). These are integrated via an intra-modal gated fusion and a cross-modal FiLM-modulated lightweight Mixture-of-Experts (FM-MOE). Experiments on MSP-Podcast and IEMOCAP show that MSF-SER consistently improves dimensional prediction, demonstrating the effectiveness of enriched semantic fusion for SER.

Method: Uses GRAM to align embeddings across video, text, and audio modalities, combined with a diffusion-based audio synthesis model conditioned on GRAM-aligned embeddings and waveform envelopes.

Result: Evaluation on Greatest Hits dataset shows enhanced semantic alignment between generated audio and video content, advancing state of the art in video-to-audio synthesis.

Conclusion: FoleyGRAM demonstrates that aligning multimodal encoders using GRAM improves semantic control and alignment in video-to-audio generation, representing an advancement in the field.

Abstract: In this work, we present FoleyGRAM, a novel approach to video-to-audio generation that emphasizes semantic conditioning through the use of aligned multimodal encoders. Building on prior advancements in video-to-audio generation, FoleyGRAM leverages the Gramian Representation Alignment Measure (GRAM) to align embeddings across video, text, and audio modalities, enabling precise semantic control over the audio generation process. The core of FoleyGRAM is a diffusion-based audio synthesis model conditioned on GRAM-aligned embeddings and waveform envelopes, ensuring both semantic richness and temporal alignment with the corresponding input video. We evaluate FoleyGRAM on the Greatest Hits dataset, a standard benchmark for video-to-audio models. Our experiments demonstrate that aligning multimodal encoders using GRAM enhances the system’s ability to semantically align generated audio with video content, advancing the state of the art in video-to-audio synthesis.

Method: Three-step framework: 1) approximate stem separation to extract guitar part, 2) strum detection using pre-trained MERT foundation model, 3) pattern decoding with expert-curated vocabulary to represent strum sequences.

Result: The method can transcribe guitar rhythmic patterns in polyphonic music with high accuracy, producing human-readable representations with automatically detected bar lines and time signature markers.

Conclusion: The proposed framework successfully addresses rhythmic pattern transcription for guitar, with ablation studies, error analysis, and evaluation metrics demonstrating its effectiveness.

Abstract: Whereas chord transcription has received considerable attention during the past couple of decades, far less work has been devoted to transcribing and encoding the rhythmic patterns that occur in a song. The topic is especially relevant for instruments such as the rhythm guitar, which is typically played by strumming rhythmic patterns that repeat and vary over time. However, in many cases one cannot objectively define a single “right” rhythmic pattern for a given song section. To create a dataset with well-defined ground-truth labels, we asked expert musicians to transcribe the rhythmic patterns in 410 popular songs and record cover versions where the guitar tracks followed those transcriptions. To transcribe the strums and their corresponding rhythmic patterns, we propose a three-step framework. Firstly, we perform approximate stem separation to extract the guitar part from the polyphonic mixture. Secondly, we detect individual strums within the separated guitar audio, using a pre-trained foundation model (MERT) as a backbone. Finally, we carry out a pattern-decoding process in which the transcribed sequence of guitar strums is represented by patterns drawn from an expert-curated vocabulary. We show that it is possible to transcribe the rhythmic patterns of the guitar track in polyphonic music with quite high accuracy, producing a representation that is human-readable and includes automatically detected bar lines and time signature markers. We perform ablation studies and error analysis and propose a set of evaluation metrics to assess the accuracy and readability of the predicted rhythmic pattern sequence.

Method: Factorizes songs into segments and generates each through selective attention to related segments, wrapped in a web application for iterative co-creation

Result: Achieves higher quality and efficiency compared to prior work in symbolic full-song generation

Conclusion: SFS model enables effective human-AI interaction for music co-creation with customizable structures and flexible ordering

Abstract: We propose the Segmented Full-Song Model (SFS) for symbolic full-song generation. The model accepts a user-provided song structure and an optional short seed segment that anchors the main idea around which the song is developed. By factorizing a song into segments and generating each one through selective attention to related segments, the model achieves higher quality and efficiency compared to prior work. To demonstrate its suitability for human-AI interaction, we further wrap SFS into a web application that enables users to iteratively co-create music on a piano roll with customizable structures and flexible ordering.

Method: Combines supervised fine-tuning with reinforcement learning using task-specific rewards for emotion category, intensity, and emphasis; unifies global intensity control in VAD space with local emphasis regulation.

Result: Improves emotion accuracy, intensity differentiation, and emphasis clarity while preserving synthesis quality comparable to strong LLM-based baselines.

Conclusion: EMORL-TTS successfully enables fine-grained emotional control in LLM-based TTS systems through reinforcement learning, demonstrating effective modulation of emotion intensity through emphasis placement.

Abstract: Recent LLM-based TTS systems achieve strong quality and zero-shot ability, but lack fine-grained emotional control due to their reliance on discrete speech tokens. Existing approaches either limit emotions to categorical labels or cannot generalize to LLM-based architectures. We propose EMORL-TTS (Fine-grained Emotion-controllable TTS with Reinforcement Learning), a framework that unifies global intensity control in the VAD space with local emphasis regulation. Our method combines supervised fine-tuning with reinforcement learning guided by task-specific rewards for emotion category, intensity, and emphasis. Moreover, we further investigate how emphasis placement modulates fine-grained emotion intensity. Experiments show that EMORL-TTS improves emotion accuracy, intensity differentiation, and emphasis clarity, while preserving synthesis quality comparable to strong LLM-based baselines.

Method: Uses Latent Affective Representation Alignment (LARA) to align internal hidden states with external music understanding models, plus an emotion control module based on continuous valence-arousal space that disentangles emotions from text content.

Result: Achieves continuous, fine-grained emotion control and significantly outperforms baselines in both emotion adherence and music quality.

Conclusion: LARA-Gen successfully enables continuous emotional control in music generation while maintaining high music quality.

Abstract: Recent advances in text-to-music models have enabled coherent music generation from text prompts, yet fine-grained emotional control remains unresolved. We introduce LARA-Gen, a framework for continuous emotion control that aligns the internal hidden states with an external music understanding model through Latent Affective Representation Alignment (LARA), enabling effective training. In addition, we design an emotion control module based on a continuous valence-arousal space, disentangling emotional attributes from textual content and bypassing the bottlenecks of text-based prompting. Furthermore, we establish a benchmark with a curated test set and a robust Emotion Predictor, facilitating objective evaluation of emotional controllability in music generation. Extensive experiments demonstrate that LARA-Gen achieves continuous, fine-grained control of emotion and significantly outperforms baselines in both emotion adherence and music quality. Generated samples are available at https://nieeim.github.io/LARA-Gen/.

Method: Incorporates Easy Consistency Tuning (ECT) strategy by progressively tightening consistency constraints on pre-trained diffusion models, and designs a multi-scale gate module (MSGate) to enhance feature fusion at different scales.

Result: Experimental results on LJSpeech dataset show ECTSpeech achieves audio quality comparable to state-of-the-art methods under single-step sampling while substantially reducing training cost and complexity.

Conclusion: ECTSpeech provides a simple and effective framework for one-step speech synthesis that maintains high audio quality while significantly reducing both training complexity and inference time.

Abstract: Diffusion models have demonstrated remarkable performance in speech synthesis, but typically require multi-step sampling, resulting in low inference efficiency. Recent studies address this issue by distilling diffusion models into consistency models, enabling efficient one-step generation. However, these approaches introduce additional training costs and rely heavily on the performance of pre-trained teacher models. In this paper, we propose ECTSpeech, a simple and effective one-step speech synthesis framework that, for the first time, incorporates the Easy Consistency Tuning (ECT) strategy into speech synthesis. By progressively tightening consistency constraints on a pre-trained diffusion model, ECTSpeech achieves high-quality one-step generation while significantly reducing training complexity. In addition, we design a multi-scale gate module (MSGate) to enhance the denoiser’s ability to fuse features at different scales. Experimental results on the LJSpeech dataset demonstrate that ECTSpeech achieves audio quality comparable to state-of-the-art methods under single-step sampling, while substantially reducing the model’s training cost and complexity.

Method: Adapts High-Resolution Networks (HRNet) for SER by converting audio to spectrograms and using HRNet’s architecture to maintain high-resolution representations from start to finish, extracting high-level features.

Result: Achieved accuracies of 92.45% on RAVDESS, 80.06% on IEMOCAP, and 92.77% on EMOVO, outperforming leading models.

Conclusion: EmoHRNet sets a new benchmark in the speech emotion recognition domain by effectively leveraging HRNet architecture to capture emotional cues from speech signals.

Abstract: Speech emotion recognition (SER) is pivotal for enhancing human-machine interactions. This paper introduces “EmoHRNet”, a novel adaptation of High-Resolution Networks (HRNet) tailored for SER. The HRNet structure is designed to maintain high-resolution representations from the initial to the final layers. By transforming audio samples into spectrograms, EmoHRNet leverages the HRNet architecture to extract high-level features. EmoHRNet’s unique architecture maintains high-resolution representations throughout, capturing both granular and overarching emotional cues from speech signals. The model outperforms leading models, achieving accuracies of 92.45% on RAVDESS, 80.06% on IEMOCAP, and 92.77% on EMOVO. Thus, we show that EmoHRNet sets a new benchmark in the SER domain.

Method: Uses modulation extraction with constrained control signal parameterizations and differentiable digital signal processing (DDSP) to discover modulations in audio.

Result: Effective on highly modulated synthetic and real audio, applicable to different DDSP synth architectures, with trade-off between interpretability and sound-matching accuracy.

Conclusion: The approach successfully extracts interpretable modulation signals from audio and is made available as code, audio samples, and VST plugin.

Abstract: Modulations are a critical part of sound design and music production, enabling the creation of complex and evolving audio. Modern synthesizers provide envelopes, low frequency oscillators (LFOs), and more parameter automation tools that allow users to modulate the output with ease. However, determining the modulation signals used to create a sound is difficult, and existing sound-matching / parameter estimation systems are often uninterpretable black boxes or predict high-dimensional framewise parameter values without considering the shape, structure, and routing of the underlying modulation curves. We propose a neural sound-matching approach that leverages modulation extraction, constrained control signal parameterizations, and differentiable digital signal processing (DDSP) to discover the modulations present in a sound. We demonstrate the effectiveness of our approach on highly modulated synthetic and real audio samples, its applicability to different DDSP synth architectures, and investigate the trade-off it incurs between interpretability and sound-matching accuracy. We make our code and audio samples available and provide the trained DDSP synths in a VST plugin.

Method: Deployed Mamba-based speech enhancement models (SEMamba) with different configurations (basic, advanced, causal, non-causal) and used both signal-level distance and metric-oriented loss functions.

Result: SEMamba achieved competitive PESQ of 3.55 on VoiceBank-DEMAND dataset, and a new SOTA PESQ of 3.69 when combined with PCS. It also showed ~12% FLOPs reduction compared to transformer-based equivalents and performed well as ASR pre-processing.

Conclusion: Mamba-based models provide an effective attention-free alternative for speech enhancement with competitive performance and computational efficiency compared to transformer-based approaches.

Abstract: This work aims to investigate the use of a recently proposed, attention-free, scalable state-space model (SSM), Mamba, for the speech enhancement (SE) task. In particular, we employ Mamba to deploy different regression-based SE models (SEMamba) with different configurations, namely basic, advanced, causal, and non-causal. Furthermore, loss functions either based on signal-level distances or metric-oriented are considered. Experimental evidence shows that SEMamba attains a competitive PESQ of 3.55 on the VoiceBank-DEMAND dataset with the advanced, non-causal configuration. A new state-of-the-art PESQ of 3.69 is also reported when SEMamba is combined with Perceptual Contrast Stretching (PCS). Compared against Transformed-based equivalent SE solutions, a noticeable FLOPs reduction up to ~12% is observed with the advanced non-causal configurations. Finally, SEMamba can be used as a pre-processing step before automatic speech recognition (ASR), showing competitive performance against recent SE solutions.

Method: Proposes DERDM-SE with dual-stream encoding to utilize both deterministic and noisy features. Uses a deterministic model combining coarse- and fine-grained processing, and investigates two conditioning approaches: deterministic-only and deterministic-noisy.

Result: Experimental results on CHiME4 show the proposed models achieve better SE evaluation scores and more stable performance compared to other diffusion-based SE models. Deterministic-enhanced conditions improve hearing experiences on real data.

Conclusion: The proposed DERDM-SE effectively leverages deterministic models to enhance diffusion-based SE, with fine-grained deterministic models showing potential for objective metrics and UNet-based models providing stable diffusion performance.

Abstract: Diffusion-based speech enhancement (SE) models need to incorporate correct prior knowledge as reliable conditions to generate accurate predictions. However, providing reliable conditions using noisy features is challenging. One solution is to use features enhanced by deterministic methods as conditions. However, the information distortion and loss caused by deterministic methods might affect the diffusion process. In this paper, we first investigate the effects of using different deterministic SE models as conditions for diffusion. We validate two conditions depending on whether the noisy feature was used as part of the condition: one using only the deterministic feature (deterministic-only), and the other using both deterministic and noisy features (deterministic-noisy). Preliminary investigation found that using deterministic enhanced conditions improves hearing experiences on real data, while the choice between using deterministic-only or deterministic-noisy conditions depends on the deterministic models. Based on these findings, we propose a dual-streaming encoding Repair-Diffusion Model for SE (DERDM-SE) to more effectively utilize both conditions. Moreover, we found that fine-grained deterministic models have greater potential in objective evaluation metrics, while UNet-based deterministic models provide more stable diffusion performance. Therefore, in the DERDM-SE, we propose a deterministic model that combines coarse- and fine-grained processing. Experimental results on CHiME4 show that the proposed models effectively leverage deterministic models to achieve better SE evaluation scores, along with more stable performance compared to other diffusion-based SE models.

Method: Applied unlearning-based attribution to a text-to-music diffusion model trained on large-scale dataset, performed grid search over hyperparameters, quantitatively evaluated consistency, and compared with non-counterfactual approaches.

Result: Unlearning-based approaches can be effectively adapted to music generative models, enabling large-scale training data attribution in this domain.

Conclusion: The work introduces large-scale training data attribution to music generation domain and paves the way for more ethical and accountable AI systems for music creation.

Abstract: This paper explores the use of unlearning methods for training data attribution (TDA) in music generative models trained on large-scale datasets. TDA aims to identify which specific training data points contributed the most to the generation of a particular output from a specific model. This is crucial in the context of AI-generated music, where proper recognition and credit for original artists are generally overlooked. By enabling white-box attribution, our work supports a fairer system for acknowledging artistic contributions and addresses pressing concerns related to AI ethics and copyright. We apply unlearning-based attribution to a text-to-music diffusion model trained on a large-scale dataset and investigate its feasibility and behavior in this setting. To validate the method, we perform a grid search over different hyperparameter configurations and quantitatively evaluate the consistency of the unlearning approach. We then compare attribution patterns from unlearning with non-counterfactual approaches. Our findings suggest that unlearning-based approaches can be effectively adapted to music generative models, introducing large-scale TDA to this domain and paving the way for more ethical and accountable AI systems for music creation.

Method: Proposes Scattering Transformer that leverages wavelet scattering networks with contextual dependencies in a transformer-like architecture without any backpropagation or training.

Result: Achieves Weighted Accuracy (WAR) of 0.786 and Unweighted Average Recall (UAR) of 0.697 on CirCor DigiScope dataset, performing competitively with state-of-the-art methods.

Conclusion: Scattering Transformer is established as a viable and promising alternative for automatic cardiac auscultation in resource-constrained setups.

Abstract: In an attempt to address the need for skilled clinicians in heart sound interpretation, recent research efforts on automating cardiac auscultation have explored deep learning approaches. The majority of these approaches have been based on supervised learning that is always challenged in occasions where training data is limited. More recently, there has been a growing interest in potentials of pre-trained self-supervised audio foundation models for biomedical end tasks. Despite exhibiting promising results, these foundational models are typically computationally intensive. Within the context of automatic cardiac auscultation, this study explores a lightweight alternative to these general-purpose audio foundation models by introducing the Scattering Transformer, a novel, training-free transformer architecture for heart murmur detection. The proposed method leverages standard wavelet scattering networks by introducing contextual dependencies in a transformer-like architecture without any backpropagation. We evaluate our approach on the public CirCor DigiScope dataset, directly comparing it against leading general-purpose foundational models. The Scattering Transformer achieves a Weighted Accuracy(WAR) of 0.786 and an Unweighted Average Recall(UAR) of 0.697, demonstrating performance highly competitive with contemporary state of the art methods. This study establishes the Scattering Transformer as a viable and promising alternative in resource-constrained setups.

Method: Created a comprehensive benchmark with 90 hours of audio (21,000 samples) from 21 real sources and 17 TTS models across 3 tasks, evaluating detection models in blind testing scenarios.

Result: The challenge provides initial insights into the strengths and limitations of current synthetic audio detection approaches, establishing a foundation for advancing detection research.

Conclusion: SAFE offers a standardized evaluation framework to drive progress in synthetic audio forensics by systematically testing detection capabilities against increasingly sophisticated synthetic speech.

Abstract: The increasing realism of synthetic speech generated by advanced text-to-speech (TTS) models, coupled with post-processing and laundering techniques, presents a significant challenge for audio forensic detection. In this paper, we introduce the SAFE (Synthetic Audio Forensics Evaluation) Challenge, a fully blind evaluation framework designed to benchmark detection models across progressively harder scenarios: raw synthetic speech, processed audio (e.g., compression, resampling), and laundered audio intended to evade forensic analysis. The SAFE challenge consisted of a total of 90 hours of audio and 21,000 audio samples split across 21 different real sources and 17 different TTS models and 3 tasks. We present the challenge, evaluation design and tasks, dataset details, and initial insights into the strengths and limitations of current approaches, offering a foundation for advancing synthetic audio detection research. More information is available at \href{https://stresearch.github.io/SAFE/}{https://stresearch.github.io/SAFE/}.

Method: The framework integrates type-2 fuzzy sets, granular computing, and clustering to handle uncertainty in noisy data, generate linguistic rules for intrinsic explainability, and incorporate fairness measures using silhouette scores and entropy.

Result: The method achieves 4% improvement in silhouette score (0.365 vs. 0.349) over type-1 fuzzy clustering, better fairness (entropy 0.918), 0.65 average rule coverage, and linear runtime efficiency (~0.005 seconds for sampled big data).

Conclusion: The proposed framework outperforms baseline methods like DBSCAN and Agglomerative Clustering in interpretability, fairness, and efficiency, making it suitable for big data environments requiring transparent and ethical ML solutions.

Abstract: The growing complexity of machine learning (ML) models in big data analytics, especially in domains such as environmental monitoring, highlights the critical need for interpretability and explainability to promote trust, ethical considerations, and regulatory adherence (e.g., GDPR). Traditional “black-box” models obstruct transparency, whereas post-hoc explainable AI (XAI) techniques like LIME and SHAP frequently compromise accuracy or fail to deliver inherent insights. This paper presents a novel framework that combines type-2 fuzzy sets, granular computing, and clustering to boost explainability and fairness in big data environments. When applied to the UCI Air Quality dataset, the framework effectively manages uncertainty in noisy sensor data, produces linguistic rules, and assesses fairness using silhouette scores and entropy. Key contributions encompass: (1) A type-2 fuzzy clustering approach that enhances cohesion by about 4% compared to type-1 methods (silhouette 0.365 vs. 0.349) and improves fairness (entropy 0.918); (2) Incorporation of fairness measures to mitigate biases in unsupervised scenarios; (3) A rule-based component for intrinsic XAI, achieving an average coverage of 0.65; (4) Scalable assessments showing linear runtime (roughly 0.005 seconds for sampled big data sizes). Experimental outcomes reveal superior performance relative to baselines such as DBSCAN and Agglomerative Clustering in terms of interpretability, fairness, and efficiency. Notably, the proposed method achieves a 4% improvement in silhouette score over type-1 fuzzy clustering and outperforms baselines in fairness (entropy reduction by up to 1%) and efficiency.

Method: Employ a transformer model for prediction and introduce a metric based on Partial Information Decomposition (PID) to measure each asset’s influence on decision-making.

Result: Two key observations: the model disregards data volatility entirely and is biased toward data with lower-frequency price movements.

Conclusion: Transformer models in finance exhibit significant biases, ignoring volatility and favoring lower-frequency data, highlighting the need for careful auditing in financial applications.

Abstract: Transformer models have become increasingly popular in financial applications, yet their potential risk making and biases remain under-explored. The purpose of this work is to audit the reliance of the model on volatile data for decision-making, and quantify how the frequency of price movements affects the model’s prediction confidence. We employ a transformer model for prediction, and introduce a metric based on Partial Information Decomposition (PID) to measure the influence of each asset on the model’s decision making. Our analysis reveals two key observations: first, the model disregards data volatility entirely, and second, it is biased toward data with lower-frequency price movements.

Method: Used Deep Q-Networks (DQN) in a custom OpenAI Gym environment modeling brute-force attacks and reactive defenses, with features like background traffic noise, honeypot traps, IP-based evasion, and multi-level rate-limiting.

Result: Defender observability and trap effectiveness created substantial barriers to attacks. Defender consistently maintained strategic advantage across 50,000+ training episodes, especially with complex defensive strategies like adaptive IP blocking.

Conclusion: The zero-sum formulation with realistic operational constraints provides a suitable environment for studying autonomous defense systems, attacker-defender co-evolution, and transfer learning to real-world network security scenarios.

Abstract: This paper presents a controlled study of adversarial reinforcement learning in network security through a custom OpenAI Gym environment that models brute-force attacks and reactive defenses on multi-port services. The environment captures realistic security trade-offs including background traffic noise, progressive exploitation mechanics, IP-based evasion tactics, honeypot traps, and multi-level rate-limiting defenses. Competing attacker and defender agents are trained using Deep Q-Networks (DQN) within a zero-sum reward framework, where successful exploits yield large terminal rewards while incremental actions incur small costs. Through systematic evaluation across multiple configurations (varying trap detection probabilities, exploitation difficulty thresholds, and training regimens), the results demonstrate that defender observability and trap effectiveness create substantial barriers to successful attacks. The experiments reveal that reward shaping and careful training scheduling are critical for learning stability in this adversarial setting. The defender consistently maintains strategic advantage across 50,000+ training episodes, with performance gains amplifying when exposed to complex defensive strategies including adaptive IP blocking and port-specific controls. Complete implementation details, reproducible hyperparameter configurations, and architectural guidelines are provided to support future research in adversarial RL for cybersecurity. The zero-sum formulation and realistic operational constraints make this environment suitable for studying autonomous defense systems, attacker-defender co-evolution, and transfer learning to real-world network security scenarios.

Method: A Conditional Variational Autoencoder (CVAE) framework that learns probabilistic mapping between design parameters and performance, enabling generation of diverse candidates conditioned on specific performance objectives.

Result: The CVAE generated 256 novel designs with 94.1% validity rate, and 77.2% of valid designs achieved superior performance compared to the SBO baseline’s single optimal design.

Conclusion: Generative inverse design discovers higher-quality solutions and provides diverse candidate portfolios, fundamentally enhancing engineering design through multi-criteria decision-making.

Abstract: Inverse design, which seeks to find optimal parameters for a target output, is a central challenge in engineering. Surrogate-based optimization (SBO) has become a standard approach, yet it is fundamentally structured to converge to a single-point solution, thereby limiting design space exploration and ignoring potentially valuable alternative topologies. This paper presents a paradigm shift from single-point optimization to generative inverse design. We introduce a framework based on a Conditional Variational Autoencoder (CVAE) that learns a probabilistic mapping between a system’s design parameters and its performance, enabling the generation of a diverse portfolio of high-performing candidates conditioned on a specific performance objective. We apply this methodology to the complex, non-linear problem of minimizing airfoil self-noise, using a high-performing SBO method from a prior benchmark study as a rigorous baseline. The CVAE framework successfully generated 256 novel designs with a 94.1% validity rate. A subsequent surrogate-based evaluation revealed that 77.2% of these valid designs achieved superior performance compared to the single optimal design found by the SBO baseline. This work demonstrates that the generative approach not only discovers higher-quality solutions but also provides a rich portfolio of diverse candidates, fundamentally enhancing the engineering design process by enabling multi-criteria decision-making.

Method: Systematic literature review following PRISMA guidelines with structured identification, screening, eligibility, and inclusion process of 118 peer-reviewed studies and institutional reports.

Result: Supervised learning (decision trees, logistic regression, SVM) dominates due to interpretability; unsupervised anomaly detection addresses novel fraud in imbalanced data; deep learning (RNN, CNN) models sequential data and complex fraud; hybrid models show superior adaptability and accuracy.

Conclusion: Hybrid models combining supervised, unsupervised, and deep learning strategies represent the most promising convergent solutions for digital banking fraud detection, though interpretability and real-time deployment challenges persist.

Abstract: This systematic literature review examines the role of machine learning in fraud detection within digital banking, synthesizing evidence from 118 peer-reviewed studies and institutional reports. Following the PRISMA guidelines, the review applied a structured identification, screening, eligibility, and inclusion process to ensure methodological rigor and transparency. The findings reveal that supervised learning methods, such as decision trees, logistic regression, and support vector machines, remain the dominant paradigm due to their interpretability and established performance, while unsupervised anomaly detection approaches are increasingly adopted to address novel fraud patterns in highly imbalanced datasets. Deep learning architectures, particularly recurrent and convolutional neural networks, have emerged as transformative tools capable of modeling sequential transaction data and detecting complex fraud typologies, though challenges of interpretability and real-time deployment persist. Hybrid models that combine supervised, unsupervised, and deep learning strategies demonstrate superior adaptability and detection accuracy, highlighting their potential as convergent solutions.

Method: Developed the first discretization of QIF neuron model for deep SNNs, derived analytical formulation for surrogate gradient windows from discretization parameters to minimize gradient mismatch and ensure training stability.

Result: Outperformed state-of-the-art LIF-based methods on CIFAR-10, CIFAR-100, ImageNet, and CIFAR-10 DVS datasets, demonstrating superior performance with richer dynamics.

Conclusion: The discretized QIF neuron model provides a compelling alternative to LIF neurons for deep SNNs, successfully combining richer nonlinear dynamics with practical scalability and training stability.

Abstract: Spiking Neural Networks (SNNs) have emerged as energy-efficient alternatives to traditional artificial neural networks, leveraging asynchronous and biologically inspired neuron dynamics. Among existing neuron models, the Leaky Integrate-and-Fire (LIF) neuron has become widely adopted in deep SNNs due to its simplicity and computational efficiency. However, this efficiency comes at the expense of expressiveness, as LIF dynamics are constrained to linear decay at each timestep. In contrast, more complex models, such as the Quadratic Integrate-and-Fire (QIF) neuron, exhibit richer, nonlinear dynamics but have seen limited adoption due to their training instability. On that note, we propose the first discretization of the QIF neuron model tailored for high-performance deep spiking neural networks and provide an in-depth analysis of its dynamics. To ensure training stability, we derive an analytical formulation for surrogate gradient windows directly from our discretizations’ parameter set, minimizing gradient mismatch. We evaluate our method on CIFAR-10, CIFAR-100, ImageNet, and CIFAR-10 DVS, demonstrating its ability to outperform state-of-the-art LIF-based methods. These results establish our discretization of the QIF neuron as a compelling alternative to LIF neurons for deep SNNs, combining richer dynamics with practical scalability.

Method: Developed a Multi-head Attention Deep & Cross Network (MADCN) framework combining feature interaction modeling and attention mechanisms, with SHAP for interpretable analysis, tested on panel data from 275 Chinese cities.

Result: MADCN achieved MSE of 406,151.063, MAE of 612.304, and R-squared of 0.991, outperforming traditional ML and DL baselines. SHAP analysis identified population, city size, urbanization rate, and GDP as most influential, with NQPF, digital economy, and AI showing moderate but meaningful effects.

Conclusion: Advancing NQPF, strengthening digital economy, and accelerating AI development can significantly reduce urban carbon emissions. Policymakers should integrate technological innovation into carbon reduction strategies through intelligent infrastructure and sector digitalization.

Abstract: New quality productive forces (NQPF), digital economy advancement, and artificial intelligence (AI) technologies are becoming crucial for promoting sustainable urban development. This study proposes a Multi-head Attention Deep & Cross Network (MADCN) framework, combining feature interaction modeling and attention mechanisms, to predict urban carbon emissions and investigate the impacts of technological factors. The framework incorporates an interpretable learning phase using SHapley Additive exPlanations (SHAP) to assess the contributions of different features. A panel dataset covering 275 Chinese cities is utilized to test the MADCN model. Experimental results demonstrate that the MADCN model achieves superior predictive performance compared to traditional machine learning and deep learning baselines, with a Mean Squared Error (MSE) of 406,151.063, a Mean Absolute Error (MAE) of 612.304, and an R-squared value of 0.991 on the test set. SHAP analysis highlights that population, city size, urbanization rate, and GDP are among the most influential factors on carbon emissions, while NQPF, digital economy index, and AI technology level also show meaningful but relatively moderate effects. Advancing NQPF, strengthening the digital economy, and accelerating AI technology development can significantly contribute to reducing urban carbon emissions. Policymakers should prioritize integrating technological innovation into carbon reduction strategies, particularly by promoting intelligent infrastructure and enhancing digitalization across sectors, to effectively achieve dual-carbon goals.

Method: Proposes snippet similarity-weighted masked input reconstruction framework using contrastive learning to capture high-level similarities among fragmented charging snippets, followed by similarity-weighted masked reconstruction to learn both granular patterns and cross-snippet relationships.

Result: Model consistently outperforms state-of-the-art baselines, achieving 31.9% lower test error than best-performing benchmark, even under challenging domain-shifted settings affected by manufacturer and age-induced distribution shifts.

Conclusion: The proposed self-supervised pre-training framework effectively learns rich representations from privacy-friendly, fragmented charging data, enabling robust battery capacity estimation that generalizes well under real-world distribution shifts.

Abstract: Accurate battery capacity estimation is key to alleviating consumer concerns about battery performance and reliability of electric vehicles (EVs). However, practical data limitations imposed by stringent privacy regulations and labeled data shortages hamper the development of generalizable capacity estimation models that remain robust to real-world data distribution shifts. While self-supervised learning can leverage unlabeled data, existing techniques are not particularly designed to learn effectively from challenging field data – let alone from privacy-friendly data, which are often less feature-rich and noisier. In this work, we propose a first-of-its-kind capacity estimation model based on self-supervised pre-training, developed on a large-scale dataset of privacy-friendly charging data snippets from real-world EV operations. Our pre-training framework, snippet similarity-weighted masked input reconstruction, is designed to learn rich, generalizable representations even from less feature-rich and fragmented privacy-friendly data. Our key innovation lies in harnessing contrastive learning to first capture high-level similarities among fragmented snippets that otherwise lack meaningful context. With our snippet-wise contrastive learning and subsequent similarity-weighted masked reconstruction, we are able to learn rich representations of both granular charging patterns within individual snippets and high-level associative relationships across different snippets. Bolstered by this rich representation learning, our model consistently outperforms state-of-the-art baselines, achieving 31.9% lower test error than the best-performing benchmark, even under challenging domain-shifted settings affected by both manufacturer and age-induced distribution shifts.

Method: Uses algebraic identities discovered via machine learning and combinatorial search to optimize matrix multiplications where operands or outputs are upper/lower-triangular, including masked attention operations.

Result: Achieves noticeable speedups over default PyTorch implementations and Triton compiled kernels for causal attention operations on GPUs.

Conclusion: FCA provides an efficient algorithm for exact causal attention computation with reduced operations and improved GPU performance.

Abstract: We present Fast Causal Attention (FCA), an algorithm that computes exact Causal Attention using 10% fewer operations. FCA accelerates a special class of matrix multiplications where either one operand or the output matrix is upper- or lower-triangular. This includes all operations in forward and backward pass of Causal Attention, such as masked product $\mathrm{Mask}(QK^{T})$. For these matrix multiplications on GPU, FCA reaches noticeable accelerations over the default PyTorch implementations and Triton compiled kernels. FCA is built upon algebraic identities discovered via machine learning and combinatorial search.

Method: PatternKV mines representative pattern vectors online, aligns each KV vector to its nearest pattern, and quantizes only the residual. This approach reshapes the KV distribution by flattening the quantization target and narrowing its range, leveraging the stable structure of K cache that evolves gradually with context and the latent semantic regularities in V cache.

Result: PatternKV delivers consistent 2-bit gains across long-context and test-time scaling settings on multiple backbones, with only 0.08% average 4-bit drop relative to FP16. It improves test-time scaling accuracy by 10% on average, raises throughput by 1.4x, and supports 1.25x larger batches.

Conclusion: The pattern-aligned residual quantization scheme effectively addresses KV cache bottlenecks by leveraging structural insights about K and V caches, enabling efficient low-bit quantization with minimal accuracy degradation while improving inference performance and scalability.

Abstract: KV cache in autoregressive LLMs eliminates redundant recomputation but has emerged as the dominant memory and bandwidth bottleneck during inference, notably with long contexts and test-time scaling. KV quantization is a key lever for reducing cache cost, but accuracy drops sharply as the native KV distribution lacks flatness and thus maintains a wide quantization range. Prior work focuses on isolating outliers, which caps their error but fails to flatten the overall distribution, leaving performance fragile under low-bit settings. In this work, we show that the K cache maintains a stable structure that evolves gradually with context, while the V cache carries latent semantic regularities. Building on these insights, we propose PatternKV, a pattern-aligned residual quantization scheme. It mines representative pattern vectors online, aligns each KV vector to its nearest pattern, and quantizes only the residual. This reshaping of the KV distribution flattens the quantization target and narrows its range, thereby improving the fidelity of low-bit KV quantization. Across long-context and test-time scaling settings on multiple backbones, PatternKV delivers consistent 2-bit gains, with a 0.08% average 4-bit drop relative to FP16, improves test-time scaling accuracy by 10% on average, and raises throughput by 1.4x while supporting 1.25x larger batches.

Method: The authors analyze the moment generating function (MGF) of quantities of interest (norm-squared of gradient or optimality gap) and the MGF of the consensus gap between users’ models, providing novel results on variance-reduction effects and fine-grained MGF bounds.

Result: DSGD converges in high-probability under the same conditions as MSE convergence, achieves order-optimal rates for both non-convex and strongly convex costs, and demonstrates linear speed-up in the number of users.

Conclusion: The analysis bridges the gap between high-probability and MSE convergence guarantees for decentralized optimization, showing that DSGD maintains strong performance in high-probability sense while matching existing MSE guarantees.

Abstract: Convergence in high-probability (HP) has been receiving increasing interest, due to its attractive properties, such as exponentially decaying tail bounds and strong guarantees for each individual run of an algorithm. While HP guarantees are extensively studied in centralized settings, much less is understood in the decentralized, networked setup. Existing HP studies in decentralized settings impose strong assumptions, like uniformly bounded gradients, or asymptotically vanishing noise, resulting in a significant gap between assumptions used to establish convergence in the HP and the mean-squared error (MSE) sense, even for vanilla Decentralized Stochastic Gradient Descent ($\mathtt{DSGD}$) algorithm. This is contrary to centralized settings, where it is known that $\mathtt{SGD}$ converges in HP under the same conditions on the cost function as needed to guarantee MSE convergence. Motivated by this observation, we revisit HP guarantees for $\mathtt{DSGD}$ in the presence of light-tailed noise. We show that $\mathtt{DSGD}$ converges in HP under the same conditions on the cost as in the MSE sense, removing uniformly bounded gradients and other restrictive assumptions, while simultaneously achieving order-optimal rates for both non-convex and strongly convex costs. Moreover, our improved analysis yields linear speed-up in the number of users, demonstrating that $\mathtt{DSGD}$ maintains strong performance in the HP sense and matches existing MSE guarantees. Our improved results stem from a careful analysis of the MGF of quantities of interest (norm-squared of gradient or optimality gap) and the MGF of the consensus gap between users’ models. To achieve linear speed-up, we provide a novel result on the variance-reduction effect of decentralized methods in the HP sense and more fine-grained bounds on the MGF for strongly convex costs, which are both of independent interest.

Method: Propose logistic-gated operators (LGO) - differentiable gates with learnable location and steepness, embedded as typed primitives and mapped back to physical units for audit. Tested hard-gate and soft variants on ICU and NHANES health datasets.

Result: Hard-gate variant recovered clinically plausible cut-points: 71% of thresholds within 10% of guideline anchors, 100% within 20%, using fewer gates than soft variant (ICU: 4.0 vs 10.0; NHANES: 5.0 vs 12.5) while maintaining competitive accuracy. Gates pruned on smooth tasks.

Conclusion: LGO enables compact symbolic equations with explicit, unit-aware thresholds that transform interpretability from post-hoc explanation into modeling constraint, providing practical calculus for regime switching and governance-ready deployment.

Abstract: Symbolic regression promises readable equations but struggles to encode unit-aware thresholds and conditional logic. We propose logistic-gated operators (LGO) – differentiable gates with learnable location and steepness – embedded as typed primitives and mapped back to physical units for audit. Across two primary health datasets (ICU, NHANES), the hard-gate variant recovers clinically plausible cut-points: 71% (5/7) of assessed thresholds fall within 10% of guideline anchors and 100% within 20%, while using far fewer gates than the soft variant (ICU median 4.0 vs 10.0; NHANES 5.0 vs 12.5), and remaining within the competitive accuracy envelope of strong SR baselines. On predominantly smooth tasks, gates are pruned, preserving parsimony. The result is compact symbolic equations with explicit, unit-aware thresholds that can be audited against clinical anchors – turning interpretability from a post-hoc explanation into a modeling constraint and equipping symbolic regression with a practical calculus for regime switching and governance-ready deployment.

Method: Proposes OptiFLIDS which applies pruning techniques during local training to reduce model complexity and energy consumption, and incorporates a customized aggregation method to handle pruned models with non-IID data distributions.

Result: Experiments on three IoT IDS datasets (TON_IoT, X-IIoTID, IDSIoT2024) show that OptiFLIDS maintains strong detection performance while improving energy efficiency.

Conclusion: OptiFLIDS is well-suited for deployment in real-world IoT environments as it addresses key FL challenges while maintaining security performance.

Abstract: In critical IoT environments, such as smart homes and industrial systems, effective Intrusion Detection Systems (IDS) are essential for ensuring security. However, developing robust IDS solutions remains a significant challenge. Traditional machine learning-based IDS models typically require large datasets, but data sharing is often limited due to privacy and security concerns. Federated Learning (FL) presents a promising alternative by enabling collaborative model training without sharing raw data. Despite its advantages, FL still faces key challenges, such as data heterogeneity (non-IID data) and high energy and computation costs, particularly for resource constrained IoT devices. To address these issues, this paper proposes OptiFLIDS, a novel approach that applies pruning techniques during local training to reduce model complexity and energy consumption. It also incorporates a customized aggregation method to better handle pruned models that differ due to non-IID data distributions. Experiments conducted on three recent IoT IDS datasets, TON_IoT, X-IIoTID, and IDSIoT2024, demonstrate that OptiFLIDS maintains strong detection performance while improving energy efficiency, making it well-suited for deployment in real-world IoT environments.

Method: Use diffusion models that learn from multiple views containing different linear transformations of unknown sources, without requiring explicit source models.

Result: The method succeeds even when no source is individually observed and observations are noisy, incomplete, and vary in resolution. It enables sampling from source priors, evaluating candidate source probabilities, and drawing from joint posterior distributions.

Conclusion: The diffusion-based approach effectively solves source separation problems across synthetic and real-world galaxy observations, providing a flexible framework for disentangling unknown sources from complex data.

Abstract: A common challenge in the natural sciences is to disentangle distinct, unknown sources from observations. Examples of this source separation task include deblending galaxies in a crowded field, distinguishing the activity of individual neurons from overlapping signals, and separating seismic events from an ambient background. Traditional analyses often rely on simplified source models that fail to accurately reproduce the data. Recent advances have shown that diffusion models can directly learn complex prior distributions from noisy, incomplete data. In this work, we show that diffusion models can solve the source separation problem without explicit assumptions about the source. Our method relies only on multiple views, or the property that different sets of observations contain different linear transformations of the unknown sources. We show that our method succeeds even when no source is individually observed and the observations are noisy, incomplete, and vary in resolution. The learned diffusion models enable us to sample from the source priors, evaluate the probability of candidate sources, and draw from the joint posterior of the source distribution given an observation. We demonstrate the effectiveness of our method on a range of synthetic problems as well as real-world galaxy observations.

Method: Used 13-parameter permutation invariant Gaussian matrix models to represent weight matrix distributions, analyzed using representation theory and graph theory, and measured distribution changes using Wasserstein distance.

Result: Permutation invariant Gaussian models effectively capture correlated Gaussianity in weight matrices beyond initialization, with interpretable parameters that characterize small departures from Gaussianity.

Conclusion: The framework provides an interpretable way to model neural network weight distributions, identifying conditions where departures from Gaussianity occur and enabling development of more general yet interpretable models.

Abstract: Ensembles of neural network weight matrices are studied through the training process for the MNIST classification problem, testing the efficacy of matrix models for representing their distributions, under assumptions of Gaussianity and permutation-symmetry. The general 13-parameter permutation invariant Gaussian matrix models are found to be effective models for the correlated Gaussianity in the weight matrices, beyond the range of applicability of the simple Gaussian with independent identically distributed matrix variables, and notably well beyond the initialisation step. The representation theoretic model parameters, and the graph-theoretic characterisation of the permutation invariant matrix observables give an interpretable framework for the best-fit model and for small departures from Gaussianity. Additionally, the Wasserstein distance is calculated for this class of models and used to quantify the movement of the distributions over training. Throughout the work, the effects of varied initialisation regimes, regularisation, layer depth, and layer width are tested for this formalism, identifying limits where particular departures from Gaussianity are enhanced and how more general, yet still highly-interpretable, models can be developed.

Method: Created a dataset of 50 CMT problems through expert collaboration, covering analytical and computational approaches. Developed machine-grading with symbolic handling of non-commuting operators via normal ordering, and evaluated 17 LLMs by programmatically checking solutions against expert ground truth.

Result: Frontier models struggle significantly - GPT5 solved only 30% of problems, average across 17 models was 11.4±2.1%. 18 problems were unsolved by any model, 26 solved by at most one model. Unsolved problems span Quantum Monte Carlo, Variational Monte Carlo, and DMRG. Answers sometimes violated fundamental symmetries or had unphysical scaling.

Conclusion: There’s a significant gap in physical reasoning skills of current LLMs for advanced research problems. The benchmark will guide development toward capable AI research assistants and tutors in condensed matter theory.

Abstract: Large language models (LLMs) have shown remarkable progress in coding and math problem-solving, but evaluation on advanced research-level problems in hard sciences remains scarce. To fill this gap, we present CMT-Benchmark, a dataset of 50 problems covering condensed matter theory (CMT) at the level of an expert researcher. Topics span analytical and computational approaches in quantum many-body, and classical statistical mechanics. The dataset was designed and verified by a panel of expert researchers from around the world. We built the dataset through a collaborative environment that challenges the panel to write and refine problems they would want a research assistant to solve, including Hartree-Fock, exact diagonalization, quantum/variational Monte Carlo, density matrix renormalization group (DMRG), quantum/classical statistical mechanics, and model building. We evaluate LLMs by programmatically checking solutions against expert-supplied ground truth. We developed machine-grading, including symbolic handling of non-commuting operators via normal ordering. They generalize across tasks too. Our evaluations show that frontier models struggle with all of the problems in the dataset, highlighting a gap in the physical reasoning skills of current LLMs. Notably, experts identified strategies for creating increasingly difficult problems by interacting with the LLMs and exploiting common failure modes. The best model, GPT5, solves 30% of the problems; average across 17 models (GPT, Gemini, Claude, DeepSeek, Llama) is 11.4$\pm$2.1%. Moreover, 18 problems are solved by none of the 17 models, and 26 by at most one. These unsolved problems span Quantum Monte Carlo, Variational Monte Carlo, and DMRG. Answers sometimes violate fundamental symmetries or have unphysical scaling dimensions. We believe this benchmark will guide development toward capable AI research assistants and tutors.

Method: Two algorithms based on accelerated primal-dual (APD): 1) naive extension using evolving parameter estimates, and 2) learning-aware variant that explicitly accounts for parameter dynamics with adjusted momentum updates and backtracking strategy for adaptive step-size selection.

Result: Both methods achieve O(log K/K) convergence rate. The learning-aware approach attains tighter O(1) constant and benefits from adaptive step-size. Extended framework for multiple optimal solutions achieves O(1/√K) rate. Superior empirical performance demonstrated on misspecified portfolio optimization.

Conclusion: The proposed learning-aware accelerated primal-dual methods effectively handle misspecified saddle point problems with unknown parameters, achieving strong theoretical guarantees and practical performance improvements over state-of-the-art approaches.

Abstract: We study a class of misspecified saddle point (SP) problems, where the optimization objective depends on an unknown parameter that must be learned concurrently from data. Unlike existing studies that assume parameters are fully known or pre-estimated, our framework integrates optimization and learning into a unified formulation, enabling a more flexible problem class. To address this setting, we propose two algorithms based on the accelerated primal-dual (APD) by Hamedani & Aybat 2021. In particular, we first analyze the naive extension of the APD method by directly substituting the evolving parameter estimates into the primal-dual updates; then, we design a new learning-aware variant of the APD method that explicitly accounts for parameter dynamics by adjusting the momentum updates. Both methods achieve a provable convergence rate of $\mathcal{O}(\log K / K)$, while the learning-aware approach attains a tighter $\mathcal{O}(1)$ constant and further benefits from an adaptive step-size selection enabled by a backtracking strategy. Furthermore, we extend the framework to problems where the learning problem admits multiple optimal solutions, showing that our modified algorithm for a structured setting achieves an $\mathcal{O}(1/\sqrt{K})$ rate. To demonstrate practical impact, we evaluate our methods on a misspecified portfolio optimization problem and show superior empirical performance compared to state-of-the-art algorithms.

Method: Developed generalized SDP framework with decomposition into sequential sub-problems, and ECLipsE-Gen-Local algorithms incorporating local input information with closed-form solutions.

Result: Achieved substantial speedups over benchmarks with significantly tighter Lipschitz bounds than global approaches, providing strict upper bounds approaching exact Jacobian values.

Conclusion: The approach offers scalable, tight Lipschitz estimation that closely aligns with network robustness, enabling practical certification of neural network robustness to input perturbations.

Abstract: The Lipschitz constant is a key measure for certifying the robustness of neural networks to input perturbations. However, computing the exact constant is NP-hard, and standard approaches to estimate the Lipschitz constant involve solving a large matrix semidefinite program (SDP) that scales poorly with network size. Further, there is a potential to efficiently leverage local information on the input region to provide tighter Lipschitz estimates. We address this problem here by proposing a compositional framework that yields tight yet scalable Lipschitz estimates for deep feedforward neural networks. Specifically, we begin by developing a generalized SDP framework that is highly flexible, accommodating heterogeneous activation function slope, and allowing Lipschitz estimates with respect to arbitrary input-output pairs and arbitrary choices of sub-networks of consecutive layers. We then decompose this generalized SDP into a sequence of small sub-problems, with computational complexity that scales linearly with respect to the network depth. We also develop a variant that achieves near-instantaneous computation through closed-form solutions to each sub-problem. All our algorithms are accompanied by theoretical guarantees on feasibility and validity. Next, we develop a series of algorithms, termed as ECLipsE-Gen-Local, that effectively incorporate local information on the input. Our experiments demonstrate that our algorithms achieve substantial speedups over a multitude of benchmarks while producing significantly tighter Lipschitz bounds than global approaches. Moreover, we show that our algorithms provide strict upper bounds for the Lipschitz constant with values approaching the exact Jacobian from autodiff when the input region is small enough. Finally, we demonstrate the practical utility of our approach by showing that our Lipschitz estimates closely align with network robustness.

Method: Conducted ablation studies comparing encoder-only and decoder-only models, then introduced two novel approaches: Parallel Flipping and Sequence Doubling to mimic bidirectionality in autoregressive decoder-only models.

Result: Decoder-only models performed far worse than encoder-only models when using existing approaches unmodified, and scaling didn’t help. However, both new methods improved decoder-only model performance across all tasks and adaptation methods, closing the gap to encoder-only performance.

Conclusion: The findings enable broader use of decoder-only models in cross-modal adaptation for scientific machine learning by addressing their limitations through novel architectural modifications.

Abstract: Large language models have shown great success on natural language tasks in recent years, but they have also shown great promise when adapted to new modalities, e.g., for scientific machine learning tasks. Even though decoder-only models are more popular within NLP and scale exceedingly well at generating natural language, most proposed approaches for cross-modal adaptation focus on encoder-only models, raising the question of how model architecture affects these approaches. In this paper, we therefore perform a series of ablation studies to answer this question, systematically comparing encoder-only and decoder-only models on cross-modal adaptation for time-dependent simulation tasks based on partial differential equations (PDEs). We find that decoder-only models are far worse than encoder-only models, when existing approaches are applied unmodified. In contrast to several other domains, scaling decoder-only models also does not help. To harness the potential of decoder-only models in this context, we introduce two novel approaches, Parallel Flipping and Sequence Doubling, attempting to mimic bidirectionality in autoregressive models. Both our methods improve overall performance using decoder-only models for all tasks and all cross-model adaptation methods, closing the gap to encoder-only model performance. We hope that our findings broaden the spectrum of models used on cross-modal adaptation tasks to further scientific ML.

Method: Action Gradient (AG) methodology that directly adjusts actions using the gradient of Q-value with respect to action, facilitating efficient integration with token prediction techniques.

Result: Significantly enhances performance of DT-based algorithms, with some results achieving state-of-the-art levels.

Conclusion: AG provides a stable and effective solution for Decision Transformer’s key challenges, enabling improved performance through direct action optimization using Q-value gradients.

Abstract: Decision Transformer (DT), which integrates reinforcement learning (RL) with the transformer model, introduces a novel approach to offline RL. Unlike classical algorithms that take maximizing cumulative discounted rewards as objective, DT instead maximizes the likelihood of actions. This paradigm shift, however, presents two key challenges: stitching trajectories and extrapolation of action. Existing methods, such as substituting specific tokens with predictive values and integrating the Policy Gradient (PG) method, address these challenges individually but fail to improve performance stably when combined due to inherent instability. To address this, we propose Action Gradient (AG), an innovative methodology that directly adjusts actions to fulfill a function analogous to that of PG, while also facilitating efficient integration with token prediction techniques. AG utilizes the gradient of the Q-value with respect to the action to optimize the action. The empirical results demonstrate that our method can significantly enhance the performance of DT-based algorithms, with some results achieving state-of-the-art levels.

Method: Compute frustration levels of signed graphs associated with pretrained deep convolutional neural networks and compare them to null models. Analyze the networks using an Ising spin glass model framework to measure disorder.

Result: All pretrained networks showed frustration levels lower than null models, indicating reduced disorder and near-monotonic behavior. Networks behave more similarly to monotone functions than expected.

Conclusion: Deep convolutional neural networks exhibit more ordered behavior than expected, with low frustration indicating proximity to structural balance, revealing a novel form of implicit regularization in these networks.

Abstract: For the signed graph associated to a deep neural network, one can compute the frustration level, i.e., test how close or distant the graph is to structural balance. For all the pretrained deep convolutional neural networks we consider, we find that the frustration is always less than expected from null models. From a statistical physics point of view, and in particular in reference to an Ising spin glass model, the reduced frustration indicates that the amount of disorder encoded in the network is less than in the null models. From a functional point of view, low frustration (i.e., proximity to structural balance) means that the function representing the network behaves near-monotonically, i.e., more similarly to a monotone function than in the null models. Evidence of near-monotonic behavior along the partial order determined by frustration is observed for all networks we consider. This confirms that the class of deep convolutional neural networks tends to have a more ordered behavior than expected from null models, and suggests a novel form of implicit regularization.

Method: Enhanced differentially private optimization algorithms with adaptable gradient clipping and other engineering improvements to create DP-Adam-AC, then fine-tuned two localizable LLM designs (Qwen2.5-0.5B and Bitnet-b1.58-2B) using synthetic datasets.

Result: Demonstrated promising improvements in loss through experimentation with two synthetic datasets, showing the effectiveness of the proposed differentially private optimization approach.

Conclusion: The work successfully addresses LLM security concerns by enabling secure local deployment through differentially private fine-tuning of localizable models, protecting both against network attacks and training data reproduction vulnerabilities.

Abstract: Large language models (LLMs) such as ChatGPT have evolved into powerful and ubiquitous tools. Fine-tuning on small datasets allows LLMs to acquire specialized skills for specific tasks efficiently. Although LLMs provide great utility in both general and task-specific use cases, they are limited by two security-related concerns. First, traditional LLM hardware requirements make them infeasible to run locally on consumer-grade devices. A remote network connection with the LLM provider’s server is usually required, making the system vulnerable to network attacks. Second, fine-tuning an LLM for a sensitive task may involve sensitive data. Non-private fine-tuning algorithms produce models vulnerable to training data reproduction attacks. Our work addresses these security concerns by enhancing differentially private optimization algorithms and applying them to fine-tune localizable language models. We introduce adaptable gradient clipping along with other engineering enhancements to the standard DP-Adam optimizer to create DP-Adam-AC. We use our optimizer to fine-tune examples of two localizable LLM designs, small language model (Qwen2.5-0.5B) and 1.58 bit quantization (Bitnet-b1.58-2B). We demonstrate promising improvements in loss through experimentation with two synthetic datasets.

Method: Analyzed cosine similarity distributions from document embeddings, proposed hierarchical clustering model for gamma mixtures, and developed EM algorithm for fitting shifted gamma mixtures.

Result: Found that cosine similarity distributions are well captured by gamma distributions (shifted/truncated to [-1,1]) and gamma mixtures, with hierarchical clustering naturally producing this structure.

Conclusion: Gamma mixtures provide effective models for sentence embedding similarity distributions, with practical fitting algorithms available for analysis.

Abstract: We study the cosine similarity of sentence transformer embeddings and observe that they are well modeled by gamma mixtures. From a fixed corpus, we measure similarities between all document embeddings and a reference query embedding. Empirically we find that these distributions are often well captured by a gamma distribution shifted and truncated to [-1,1], and in many cases, by a gamma mixture. We propose a heuristic model in which a hierarchical clustering of topics naturally leads to a gamma-mixture structure in the similarity scores. Finally, we outline an expectation-maximization algorithm for fitting shifted gamma mixtures, which provides a practical tool for modeling similarity distributions.

Method: Two regularization strategies: (1) Weighted adversarial mutual regularization with decomposed KL-divergence loss allowing flexible weight control between main and auxiliary objectives; (2) Adversarial generalization regularization that adds clean target distribution to the training objective.

Result: Extensive experiments show the proposed methods significantly improve adversarial robustness compared to existing regularization-based approaches.

Conclusion: The proposed regularization strategies effectively address the limitations of MSE in adversarial training and substantially enhance model robustness against adversarial attacks.

Abstract: Adversarial training is the most effective defense against adversarial attacks. The effectiveness of the adversarial attacks has been on the design of its loss function and regularization term. The most widely used loss function in adversarial training is cross-entropy and mean squared error (MSE) as its regularization objective. However, MSE enforces overly uniform optimization between two output distributions during training, which limits its robustness in adversarial training scenarios. To address this issue, we revisit the idea of mutual learning (originally designed for knowledge distillation) and propose two novel regularization strategies tailored for adversarial training: (i) weighted adversarial mutual regularization and (ii) adversarial generalization regularization. In the former, we formulate a decomposed adversarial mutual Kullback-Leibler divergence (KL-divergence) loss, which allows flexible control over the optimization process by assigning unequal weights to the main and auxiliary objectives. In the latter, we introduce an additional clean target distribution into the adversarial training objective, improving generalization and enhancing model robustness. Extensive experiments demonstrate that our proposed methods significantly improve adversarial robustness compared to existing regularization-based approaches.

Method: Introduces Tensor-on-Tensor Regression Neural Network (TRNN) that unifies tensor-based regression with neural networks, maintaining tensor structure while enabling nonlinear modeling.

Result: The paper presents a unified framework that combines the geometric preservation of tensor methods with the nonlinear expressiveness of neural networks.

Conclusion: TRNN provides a solution for regression on high-dimensional tensor data that preserves spatial structure while capturing complex nonlinear relationships that dominate industrial processes.

Abstract: Modern sensing and metrology systems now stream terabytes of heterogeneous, high-dimensional (HD) data profiles, images, and dense point clouds, whose natural representation is multi-way tensors. Understanding such data requires regression models that preserve tensor geometry, yet remain expressive enough to capture the pronounced nonlinear interactions that dominate many industrial and mechanical processes. Existing tensor-based regressors meet the first requirement but remain essentially linear. Conversely, conventional neural networks offer nonlinearity only after flattening, thereby discarding spatial structure and incurring prohibitive parameter counts. This paper introduces a Tensor-on-Tensor Regression Neural Network (TRNN) that unifies these two paradigms.

Method: Two-step approach: first train reward model to estimate preference margins, then use these margins to apply continuous adaptive weights to DPO loss for each training sample, amplifying hard pairs and dampening easy pairs.

Result: MADPO significantly outperforms baselines, achieving +33.3% on High Quality data and +10.5% on Low Quality data over next-best method in sentiment generation tasks.

Conclusion: MADPO provides a more robust and principled approach to preference alignment with stable optimization landscape and granular control over learning signals.

Abstract: Direct Preference Optimization (DPO) has emerged as a simple and effective method for aligning large language models. However, its reliance on a fixed temperature parameter leads to suboptimal training on diverse preference data, causing overfitting on easy examples and under-learning from informative ones. Recent methods have emerged to counter this. While IPO addresses general overfitting, its uniform regularization can be overly conservative. The more targeted approach of $\beta$-DPO suffers from its own limitations: its batch-level adaptation applies a single, compromised temperature to mixed-margin pairs, its linear update rule can produce unstable negative $\beta$ values, and its filtering mechanism discards potentially useful training signals. In this work, we introduce Margin-Adaptive Direct Preference Optimization (MADPO), a method that provides a stable, data-preserving, and instance-level solution. MADPO employs a practical two-step approach: it first trains a reward model to estimate preference margins and then uses these margins to apply a continuous, adaptive weight to the DPO loss for each individual training sample. This re-weighting scheme creates an effective target margin that is amplified for hard pairs and dampened for easy pairs, allowing for granular control over the learning signal. We provide a comprehensive theoretical analysis, proving that MADPO has a well-behaved optimization landscape and is robust to reward model estimation errors. We validate our theory with experiments on a sentiment generation task, where MADPO consistently and significantly outperforms strong baselines across datasets of varying quality. It achieves performance gains of up to +33.3% on High Quality data and +10.5% on Low Quality data over the next-best method. Our results establish MADPO as a more robust and principled approach to preference alignment.

Method: Integrates Efficient Channel Attention (ECA) mechanisms with Dilated Convolutions (DC) and applies physics-informed loss by enforcing force-free and divergence-free conditions during training.

Result: Outperforms state-of-the-art neural operators in accuracy and shows strong consistency with physical characteristics of NLFFF data across various solar active regions.

Conclusion: PIANO provides an effective physics-informed deep learning approach for direct 3D magnetic field reconstruction from 2D boundary conditions in solar physics.

Abstract: We propose Physics-informed Attention-enhanced Fourier Neural Operator (PIANO) to solve the Nonlinear Force-Free Field (NLFFF) problem in solar physics. Unlike conventional approaches that rely on iterative numerical methods, our proposed PIANO directly learns the 3D magnetic field structure from 2D boundary conditions. Specifically, PIANO integrates Efficient Channel Attention (ECA) mechanisms with Dilated Convolutions (DC), which enhances the model’s ability to capture multimodal input by prioritizing critical channels relevant to the magnetic field’s variations. Furthermore, we apply physics-informed loss by enforcing the force-free and divergence-free conditions in the training process so that our prediction is consistent with underlying physics with high accuracy. Experimental results on the ISEE NLFFF dataset show that our PIANO not only outperforms state-of-the-art neural operators in terms of accuracy but also shows strong consistency with the physical characteristics of NLFFF data across magnetic fields reconstructed from various solar active regions. The GitHub of this project is available https://github.com/Autumnstar-cjh/PIANO

Method: Proposes MT-DAO optimizer family that employs multiple slow- and fast-moving first momenta to track gradient dynamics across different time scales, with theoretical convergence guarantees.

Result: Eliminates performance gap with DDP, outperforms infrequent-communication baselines in perplexity, reduces wall-clock time by 6-27% on Ethernet, and reaches target perplexity in 24% fewer steps and 35% less time at 720M scale.

Conclusion: MT-DAO enables effective cross-datacenter training and training over wide geographic areas by solving the time-scale mismatch problem in infrequent communication strategies.

Abstract: Training large models with distributed data parallelism (DDP) requires frequent communication of gradients across workers, which can saturate bandwidth. Infrequent communication strategies (e.g., Local SGD) reduce this overhead but, when applied to adaptive optimizers, often suffer a performance gap relative to fully synchronous DDP. We trace this gap to a time-scale mismatch: the optimizer’s fast-moving momentum, tuned for frequent updates, decays too quickly to smooth gradients over long intervals, leading to noise-dominated optimization. To address this, we propose MT-DAO, a family of optimizers that employs multiple slow- and fast-moving first momenta or the gradient to track update dynamics across different time scales, for which we provide the first convergence guarantees. Empirically, for language-model pre-training, this eliminates the performance gap with DDP, outperforming infrequent-communication baselines in perplexity and reducing iso-token wall-clock time by 6-27% on Ethernet interconnects. At the 720M scale, MT-DAO reaches a target perplexity in 24% fewer steps and 35% less time than the single-momentum DDP baseline. MT-DAO enables effective cross-datacenter training and training over wide geographic areas.

Method: Combines Hadamard rotation to reduce quantization error in values with lightweight linear correction adapters that compensate for errors from quantized keys.

Result: Consistently matches or surpasses baselines across LLaMA, Qwen2.5, and Qwen3 models with higher KV-cache compression, and achieves up to 2.55x faster inference via custom attention kernel.

Conclusion: KVLinC effectively addresses KV cache quantization errors, enabling efficient long-context LLM inference with maintained quality and improved performance.

Abstract: Quantizing the key-value (KV) cache is a promising strategy for improving the inference efficiency of large language models (LLMs). However, aggressive quantization to very low precision (e.g., 2 bits) introduces significant errors in the stored key and value tensors, which propagate through the dot-product attention mechanism and ultimately degrade generation quality. To address this, we propose KVLinC, a framework to mitigate attention errors introduced by KV cache quantization in the extreme low-precision regime. KVLinC combines a Hadamard rotation, which reduces quantization error in values, with lightweight linear correction adapters that explicitly compensate for errors introduced by quantized keys. Across extensive evaluations on the LLaMA, Qwen2.5, and Qwen3 model families, KVLinC consistently matches or surpasses strong baselines while achieving higher KV-cache compression. Furthermore, we implement a custom attention kernel that results in upto 2.55x faster inference compared to Flash Attention baseline, enabling efficient long-context LLM inference.

Method: Redesigned PINNSformer by removing the encoder (found unnecessary for spatiotemporal correlations with self-attention) and integrated Fourier feature embeddings to explicitly mitigate spectral bias and enable adaptive multiscale frequency encoding.

Result: S-Pformer outperforms encoder-decoder PINNSformer architectures across all benchmarks, achieving or outperforming MLP performance while significantly reducing parameter count.

Conclusion: The proposed S-Pformer provides a more efficient and effective Transformer-based PINN architecture by eliminating encoder redundancy and addressing spectral bias through Fourier feature embeddings.

Abstract: Physics-Informed Neural Networks (PINNs) are a useful framework for approximating partial differential equation solutions using deep learning methods. In this paper, we propose a principled redesign of the PINNsformer, a Transformer-based PINN architecture. We present the Spectral PINNSformer (S-Pformer), a refinement of encoder-decoder PINNSformers that addresses two key issues; 1. the redundancy (i.e. increased parameter count) of the encoder, and 2. the mitigation of spectral bias. We find that the encoder is unnecessary for capturing spatiotemporal correlations when relying solely on self-attention, thereby reducing parameter count. Further, we integrate Fourier feature embeddings to explicitly mitigate spectral bias, enabling adaptive encoding of multiscale behaviors in the frequency domain. Our model outperforms encoder-decoder PINNSformer architectures across all benchmarks, achieving or outperforming MLP performance while reducing parameter count significantly.

Method: Uses shallow neural network with randomized hidden weights and biases (random feature method) for KL divergence estimation.

Result: The algorithm achieves KL divergence estimation error of O(m^{-1/2}+T^{-1/3}) with high probability, where m is number of neurons and T is number of steps/samples.

Conclusion: Provides a constructive neural network approach for KL divergence estimation with provable error bounds, addressing limitations of non-constructive theoretical analyses.

Abstract: Estimating the Kullback-Leibler (KL) divergence between random variables is a fundamental problem in statistical analysis. For continuous random variables, traditional information-theoretic estimators scale poorly with dimension and/or sample size. To mitigate this challenge, a variety of methods have been proposed to estimate KL divergences and related quantities, such as mutual information, using neural networks. The existing theoretical analyses show that neural network parameters achieving low error exist. However, since they rely on non-constructive neural network approximation theorems, they do not guarantee that the existing algorithms actually achieve low error. In this paper, we propose a KL divergence estimation algorithm using a shallow neural network with randomized hidden weights and biases (i.e. a random feature method). We show that with high probability, the algorithm achieves a KL divergence estimation error of $O(m^{-1/2}+T^{-1/3})$, where $m$ is the number of neurons and $T$ is both the number of steps of the algorithm and the number of samples.

Method: Proposes a data-efficient neural architecture using transfer learning and model fusion, pretraining specialized neural regressors on distinct conditions and integrating them into a unified global model with skip connections for stability.

Result: Experimental validation shows significant improvements in generalization across material variability and geometric diversity, achieving superior performance compared to models trained from scratch while reducing need for large simulation datasets.

Conclusion: The approach establishes a scalable ML-based solution for intelligent thermal control in manufacturing and demonstrates how data-efficient generalization strategies can extend to other industrial applications with limited data.

Abstract: Accurate and efficient temperature prediction is critical for optimizing the preheating process of PET preforms in industrial microwave systems prior to blow molding. We propose a novel deep learning framework for generalized temperature prediction. Unlike traditional models that require extensive retraining for each material or design variation, our method introduces a data-efficient neural architecture that leverages transfer learning and model fusion to generalize across unseen scenarios. By pretraining specialized neural regressor on distinct conditions such as recycled PET heat capacities or varying preform geometries and integrating their representations into a unified global model, we create a system capable of learning shared thermal dynamics across heterogeneous inputs. The architecture incorporates skip connections to enhance stability and prediction accuracy. Our approach reduces the need for large simulation datasets while achieving superior performance compared to models trained from scratch. Experimental validation on two case studies material variability and geometric diversity demonstrates significant improvements in generalization, establishing a scalable ML-based solution for intelligent thermal control in manufacturing environments. Moreover, the approach highlights how data-efficient generalization strategies can extend to other industrial applications involving complex physical modeling with limited data.

Method: Used deep learning sequence-to-sequence models based on LSTM networks with 40 well-connected SPEs from 1997-2017, employing 4-fold stratified cross-validation and testing various configurations including proton-only vs. proton+X-ray inputs, original vs. trend-smoothed data, and autoregressive vs. one-shot forecasting.

Result: One-shot forecasting consistently yielded lower error than autoregressive prediction; proton-only models outperformed proton+X-ray models on original data, but this gap narrowed with trend-smoothed data; trend-smoothing significantly enhanced proton+X-ray model performance; best-performing model was trained on original data despite trend-smoothed data performing better on average.

Conclusion: Architectural choices can sometimes outweigh the benefits of data preprocessing, and careful model configuration is essential for optimal SPE proton flux forecasting performance.

Abstract: Solar Proton Events (SPEs) cause significant radiation hazards to satellites, astronauts, and technological systems. Accurate forecasting of their proton flux time profiles is crucial for early warnings and mitigation. This paper explores deep learning sequence-to-sequence (seq2seq) models based on Long Short-Term Memory networks to predict 24-hour proton flux profiles following SPE onsets. We used a dataset of 40 well-connected SPEs (1997-2017) observed by NOAA GOES, each associated with a >=M-class western-hemisphere solar flare and undisturbed proton flux profiles. Using 4-fold stratified cross-validation, we evaluate seq2seq model configurations (varying hidden units and embedding dimensions) under multiple forecasting scenarios: (i) proton-only input vs. combined proton+X-ray input, (ii) original flux data vs. trend-smoothed data, and (iii) autoregressive vs. one-shot forecasting. Our major results are as follows: First, one-shot forecasting consistently yields lower error than autoregressive prediction, avoiding the error accumulation seen in iterative approaches. Second, on the original data, proton-only models outperform proton+X-ray models. However, with trend-smoothed data, this gap narrows or reverses in proton+X-ray models. Third, trend-smoothing significantly enhances the performance of proton+X-ray models by mitigating fluctuations in the X-ray channel. Fourth, while models trained on trendsmoothed data perform best on average, the best-performing model was trained on original data, suggesting that architectural choices can sometimes outweigh the benefits of data preprocessing.

Method: Propose NoiseCurve technique that uses model curvature estimated from public unlabeled data to improve cross-iteration noise correlation quality in DP-SGD.

Result: Experiments show NoiseCurve provides consistent and significant accuracy improvements over DP-MF correlation scheme across various datasets, models, and privacy parameters.

Conclusion: NoiseCurve effectively enhances privacy-preserving training by leveraging model curvature to improve noise correlation, bridging the accuracy gap in differentially private deep learning.

Abstract: Differentially private stochastic gradient descent (DP-SGD) offers the promise of training deep learning models while mitigating many privacy risks. However, there is currently a large accuracy gap between DP-SGD and normal SGD training. This has resulted in different lines of research investigating orthogonal ways of improving privacy-preserving training. One such line of work, known as DP-MF, correlates the privacy noise across different iterations of stochastic gradient descent – allowing later iterations to cancel out some of the noise added to earlier iterations. In this paper, we study how to improve this noise correlation. We propose a technique called NoiseCurve that uses model curvature, estimated from public unlabeled data, to improve the quality of this cross-iteration noise correlation. Our experiments on various datasets, models, and privacy parameters show that the noise correlations computed by NoiseCurve offer consistent and significant improvements in accuracy over the correlation scheme used by DP-MF.

Method: Partitions LLM into drafter and verifier, converts verification decisions into supervision signals for online drafter updates using KL→RL schedule (online distillation followed by reward-masked cross-entropy with policy gradient).

Result: Achieves 2.16x wall-time speedup on Spec-Bench, comparable to EAGLE-2 but with orders of magnitude less training data, and outperforms KL-only online distillation.

Conclusion: Training-aware self-speculation can deliver state-of-the-art, lossless speedups with minimal training overhead.

Abstract: Autoregressive (AR) decoding is a major latency bottleneck for large language models. Speculative decoding (SD) accelerates AR by letting a drafter propose multi-token blocks that a verifier accepts or rejects. However, many SD systems require heavy offline training or extra components. These choices raise data/compute cost and can yield brittle drafters under distribution drift. We introduce \emph{Draft, Verify, & Improve (DVI)}, a training-aware self-speculative framework that combines inference with continual online learning. We partition an LLM into a drafter and a verifier, and during generation, verifier accept/reject decisions are converted into supervision signals and used to update the drafter head. A simple \emph{KL$\rightarrow$RL} schedule bootstraps calibration via online distillation and then adds reward-masked cross-entropy with a on-policy policy-gradient term, preserving lossless, single model deployment. On Spec-Bench, DVI achieves a $2.16\times$ wall-time speedup, on par with SoTA approaches like EAGLE-2, while orders of magnitude less data for training, and ablations show that DVI outperforms KL-only online distillation. DVI demonstrates that \emph{training-aware} self-speculation can deliver state-of-the-art, lossless speedups with minimal training overhead.

Method: Review of three PIML frameworks: 1) Physics-informed neural networks (PINNs) that embed governing equations into deep learning models, 2) Neural ordinary differential equations (NODEs) for continuous-time modeling of dynamic systems, and 3) Neural operators for learning mappings between function spaces in multiscale biological domains.

Result: Successful applications demonstrated across various biomedical areas including biosolid and biofluid mechanics, mechanobiology, medical imaging, physiological systems, pharmacokinetics, cell signaling, and multiscale biological simulations.

Conclusion: PIML is emerging as a transformative paradigm for biomedical modeling, with future directions including addressing uncertainty quantification, generalization challenges, and integration with large language models.

Abstract: Physics-informed machine learning (PIML) is emerging as a potentially transformative paradigm for modeling complex biomedical systems by integrating parameterized physical laws with data-driven methods. Here, we review three main classes of PIML frameworks: physics-informed neural networks (PINNs), neural ordinary differential equations (NODEs), and neural operators (NOs), highlighting their growing role in biomedical science and engineering. We begin with PINNs, which embed governing equations into deep learning models and have been successfully applied to biosolid and biofluid mechanics, mechanobiology, and medical imaging among other areas. We then review NODEs, which offer continuous-time modeling, especially suited to dynamic physiological systems, pharmacokinetics, and cell signaling. Finally, we discuss deep NOs as powerful tools for learning mappings between function spaces, enabling efficient simulations across multiscale and spatially heterogeneous biological domains. Throughout, we emphasize applications where physical interpretability, data scarcity, or system complexity make conventional black-box learning insufficient. We conclude by identifying open challenges and future directions for advancing PIML in biomedical science and engineering, including issues of uncertainty quantification, generalization, and integration of PIML and large language models.

Method: Proposes ARLAS framework using adversarial RL as a two-player zero-sum game, co-training attacker and defender LLMs with population-based learning to defend against all previous attacker checkpoints.

Result: ARLAS-trained agents achieve significantly lower attack success rates while improving task success rates on BrowserGym and AgentDojo benchmarks, generating diverse and challenging attacks.

Conclusion: The adversarial process creates robust agents that are more resilient to prompt injections compared to base models, demonstrating the effectiveness of autonomous attack generation for defense training.

Abstract: Large Language Model (LLM) agents can leverage tools such as Google Search to complete complex tasks. However, this tool usage introduces the risk of indirect prompt injections, where malicious instructions hidden in tool outputs can manipulate the agent, posing security risks like data leakage. Current defense strategies typically rely on fine-tuning LLM agents on datasets of known attacks. However, the generation of these datasets relies on manually crafted attack patterns, which limits their diversity and leaves agents vulnerable to novel prompt injections. To address this limitation, we propose Adversarial Reinforcement Learning for Agent Safety (ARLAS), a novel framework that leverages adversarial reinforcement learning (RL) by formulating the problem as a two-player zero-sum game. ARLAS co-trains two LLMs: an attacker that learns to autonomously generate diverse prompt injections and an agent that learns to defend against them while completing its assigned tasks. To ensure robustness against a wide range of attacks and to prevent cyclic learning, we employ a population-based learning framework that trains the agent to defend against all previous attacker checkpoints. Evaluated on BrowserGym and AgentDojo, agents fine-tuned with ARLAS achieve a significantly lower attack success rate than the original model while also improving their task success rate. Our analysis further confirms that the adversarial process generates a diverse and challenging set of attacks, leading to a more robust agent compared to the base model.

Method: Proposed two Thompson-style algorithms: MTSRL (learns prior mean with known covariance) and MTSRL⁺ (learns both mean and covariance with prior widening). Used linear Q-function representation with Gaussian meta-prior and developed prior-alignment technique for theoretical guarantees.

Result: Achieved meta-regret bounds of Õ(H⁴S³/²√ANK) for known covariance and Õ(H⁴S³/²√AN³K) for learned covariance. Both outperform prior-independent methods after sufficient tasks (K ≳ Õ(H²) and K ≳ Õ(N²H²) respectively). Simulations show tracking meta-oracle performance after brief exploration.

Conclusion: First meta-regret guarantees for Thompson-style RL with learned Q-priors. Practical algorithms (MTSRL/MTSRL⁺) significantly outperform prior-independent methods and provide effective recipes for experiment-rich settings.

Abstract: We study meta-reinforcement learning in finite-horizon MDPs where related tasks share similar structures in their optimal action-value functions. Specifically, we posit a linear representation $Q^_h(s,a)=\Phi_h(s,a),\theta^{(k)}_h$ and place a Gaussian meta-prior $ \mathcal{N}(\theta^_h,\Sigma^*_h)$ over the task-specific parameters $\theta^{(k)}_h$. Building on randomized value functions, we propose two Thompson-style algorithms: (i) MTSRL, which learns only the prior mean and performs posterior sampling with the learned mean and known covariance; and (ii) $\text{MTSRL}^{+}$, which additionally estimates the covariance and employs prior widening to control finite-sample estimation error. Further, we develop a prior-alignment technique that couples the posterior under the learned prior with a meta-oracle that knows the true prior, yielding meta-regret guarantees: we match prior-independent Thompson sampling in the small-task regime and strictly improve with more tasks once the prior is learned. Concretely, for known covariance we obtain $\tilde{O}(H^{4}S^{3/2}\sqrt{ANK})$ meta-regret, and with learned covariance $\tilde{O}(H^{4}S^{3/2}\sqrt{AN^3K})$; both recover a better behavior than prior-independent after $K \gtrsim \tilde{O}(H^2)$ and $K \gtrsim \tilde{O}(N^2H^2)$, respectively. Simulations on a stateful recommendation environment (with feature and prior misspecification) show that after brief exploration, MTSRL/MTSRL(^+) track the meta-oracle and substantially outperform prior-independent RL and bandit-only meta-baselines. Our results give the first meta-regret guarantees for Thompson-style RL with learned Q-priors, and provide practical recipes (warm-start via RLSVI, OLS aggregation, covariance widening) for experiment-rich settings.

Method: Extends DeepGR4J using quantile regression-based ensemble learning framework to quantify uncertainty, enables multi-step streamflow predictions with uncertainty bounds, and evaluates using CAMELS-Aus dataset.

Result: Quantile DeepGR4J improves predictive accuracy and uncertainty interval quality compared to baseline deep learning models, and demonstrates suitability as an early warning system for flood risk evaluation.

Conclusion: The proposed framework successfully combines conceptual hydrological models with deep learning and quantile regression for improved streamflow prediction with uncertainty quantification and flood risk assessment capabilities.

Abstract: Conceptual rainfall-runoff models aid hydrologists and climate scientists in modelling streamflow to inform water management practices. Recent advances in deep learning have unravelled the potential for combining hydrological models with deep learning models for better interpretability and improved predictive performance. In our previous work, we introduced DeepGR4J, which enhanced the GR4J conceptual rainfall-runoff model using a deep learning model to serve as a surrogate for the routing component. DeepGR4J had an improved rainfall-runoff prediction accuracy, particularly in arid catchments. Quantile regression models have been extensively used for quantifying uncertainty while aiding extreme value forecasting. In this paper, we extend DeepGR4J using a quantile regression-based ensemble learning framework to quantify uncertainty in streamflow prediction. We also leverage the uncertainty bounds to identify extreme flow events potentially leading to flooding. We further extend the model to multi-step streamflow predictions for uncertainty bounds. We design experiments for a detailed evaluation of the proposed framework using the CAMELS-Aus dataset. The results show that our proposed Quantile DeepGR4J framework improves the predictive accuracy and uncertainty interval quality (interval score) compared to baseline deep learning models. Furthermore, we carry out flood risk evaluation using Quantile DeepGR4J, and the results demonstrate its suitability as an early warning system.

Method: Implement Adaptive Mixed bit Activation Quantization (AMAQ) that progressively compresses activations and gradients from 6-8 bits to 3-4 bits using bit regularization to allocate bit budgets based on channel-wise and layer-wise importance.

Result: AMAQ outperforms fixed-precision approaches with 2.5% higher generation accuracy and 1.3% better classification accuracy for models like LLaMA3 8B and Qwen2.5 7B under same bit budgets. It also enhances training stability and reduces ultra-low bit representation collapse.

Conclusion: AMAQ effectively integrates into practical multi-machine collaborative training setups, offering superior inference accuracy with modest communication overhead, making it a practical solution for collaborative training with minimal communication cost.

Abstract: Large Language Models (LLMs) are scaling rapidly, creating significant challenges for collaborative server client distributed training, particularly in terms of communication efficiency and computational overheads. To address these challenges, we implement Parameter-efficient Split Learning, which effectively balances efficiency and performance for collaborative training on low-resource devices. To reduce communication overhead in collaborative training, we introduce Adaptive Mixed bit Activation Quantization (AMAQ), a strategy that progressively compresses activations and gradients from high precision (6 to 8 bits) to low precision (3 to 4 bits). AMAQ achieves this by effectively allocating bit budgets across channels based on feature wise and layer wise importance using bit regularization. Under the same bit budgets, AMAQ outperforms fixed-precision approaches, delivering about 2.5% higher generation accuracy and about 1.3% better classification accuracy for models like LLaMA3 8B and Qwen2.5 7B. In addition, it significantly enhances training stability and reducing ultra-low bit representation collapse during the training. Experiments demonstrate that AMAQ integrates effectively into practical multi-machine collaborative training setups, offering superior inference accuracy with only a modest communication overhead for bits adaptation during training. This trade off makes AMAQ a practical and effective solution for collaborative training with minimal communication cost.

Method: ATOM adopts a quasi-equivariant design without explicit molecular graphs and employs temporal attention for parallel decoding of multiple future states. It’s pretrained on TG80, a large diverse MD dataset with 2.5M femtoseconds of trajectories across 80 compounds.

Result: ATOM achieves SOTA performance on established benchmarks (MD17, RMD17, MD22) and shows exceptional zero-shot generalization to unseen molecules across varying time horizons after multitask pretraining.

Conclusion: ATOM represents a significant step toward accurate, efficient, and transferable molecular dynamics models.

Abstract: Molecular dynamics (MD) simulations underpin modern computational drug dis- covery, materials science, and biochemistry. Recent machine learning models provide high-fidelity MD predictions without the need to repeatedly solve quantum mechanical forces, enabling significant speedups over conventional pipelines. Yet many such methods typically enforce strict equivariance and rely on sequential rollouts, thus limiting their flexibility and simulation efficiency. They are also com- monly single-task, trained on individual molecules and fixed timeframes, which restricts generalization to unseen compounds and extended timesteps. To address these issues, we propose Atomistic Transformer Operator for Molecules (ATOM), a pretrained transformer neural operator for multitask molecular dynamics. ATOM adopts a quasi-equivariant design that requires no explicit molecular graph and employs a temporal attention mechanism, allowing for the accurate parallel decod- ing of multiple future states. To support operator pretraining across chemicals and timescales, we curate TG80, a large, diverse, and numerically stable MD dataset with over 2.5 million femtoseconds of trajectories across 80 compounds. ATOM achieves state-of-the-art performance on established single-task benchmarks, such as MD17, RMD17 and MD22. After multitask pretraining on TG80, ATOM shows exceptional zero-shot generalization to unseen molecules across varying time hori- zons. We believe ATOM represents a significant step toward accurate, efficient, and transferable molecular dynamics models

Method: Proposes Method of Infinite Descent which reformulates training as solving first-order optimality conditions directly. Uses analytical resummation of Taylor expansions to derive exact algebraic equations for update steps. Introduces AION architecture designed for algebraic closure required by the method.

Result: In a simple test problem, AION reaches the optimum in a single descent step. Demonstrates how analytic structure enables exact, non-iterative convergence. The method applies to any appropriately closed architecture, defining a new ‘Infinity Class’ of models.

Conclusion: The work presents a pathway toward non-iterative learning through semi-analytic optimization. The proposed optimiser-model pair exemplifies how analytic structure can enable exact convergence without traditional iterative updates.

Abstract: Training - the optimisation of complex models - is traditionally performed through small, local, iterative updates [D. E. Rumelhart, G. E. Hinton, R. J. Williams, Nature 323, 533-536 (1986)]. Approximating solutions through truncated gradients is a paradigm dating back to Cauchy [A.-L. Cauchy, Comptes Rendus Math'ematique 25, 536-538 (1847)] and Newton [I. Newton, The Method of Fluxions and Infinite Series (Henry Woodfall, London, 1736)]. This work introduces the Method of Infinite Descent, a semi-analytic optimisation paradigm that reformulates training as the direct solution to the first-order optimality condition. By analytical resummation of its Taylor expansion, this method yields an exact, algebraic equation for the update step. Realisation of the infinite Taylor tower’s cascading resummation is formally derived, and an exploitative algorithm for the direct solve step is proposed. This principle is demonstrated with the herein-introduced AION (Analytic, Infinitely-Optimisable Network) architecture. AION is a model designed expressly to satisfy the algebraic closure required by Infinite Descent. In a simple test problem, AION reaches the optimum in a single descent step. Together, this optimiser-model pair exemplify how analytic structure enables exact, non-iterative convergence. Infinite Descent extends beyond this example, applying to any appropriately closed architecture. This suggests a new class of semi-analytically optimisable models: the \emph{Infinity Class}; sufficient conditions for class membership are discussed. This offers a pathway toward non-iterative learning.

Method: NorMuon maintains second-order momentum statistics for each neuron and applies row-wise normalization after orthogonalization, with an efficient distributed implementation under FSDP2 framework.

Result: NorMuon outperforms both Adam and Muon, achieving 21.74% better training efficiency than Adam and 11.31% improvement over Muon on 1.1B parameter pretraining, with comparable memory footprint to Muon.

Conclusion: Orthogonalization and adaptive learning rates are complementary approaches, opening new avenues for optimizer design in large-scale deep learning.

Abstract: The choice of optimizer significantly impacts the training efficiency and computational costs of large language models (LLMs). Recently, the Muon optimizer has demonstrated promising results by orthogonalizing parameter updates, improving optimization geometry through better conditioning. Despite Muon’s emergence as a candidate successor to Adam, the potential for jointly leveraging their strengths has not been systematically explored. In this work, we bridge this gap by proposing NorMuon (Neuron-wise Normalized Muon), an optimizer that synergistically combines orthogonalization with neuron-level adaptive learning rates. Our analysis reveals that while Muon effectively reduces condition numbers, the resulting updates exhibit highly non-uniform neuron norms, causing certain neurons to dominate the optimization process. NorMuon addresses this imbalance by maintaining second-order momentum statistics for each neuron and applying row-wise normalization after orthogonalization, ensuring balanced parameter utilization while preserving Muon’s conditioning benefits. To enable practical deployment at scale, we develop an efficient distributed implementation under the FSDP2 framework that strategically distributes orthogonalization computations across devices. Experiments across multiple model scales demonstrate that NorMuon consistently outperforms both Adam and Muon, achieving 21.74% better training efficiency than Adam and 11.31% improvement over Muon on 1.1 B pretraining setting, while maintaining a comparable memory footprint to Muon. Our findings suggest that orthogonalization and adaptive learning rates are complementary rather than competing approaches, opening new avenues for optimizer design in large-scale deep learning.

Method: Conditional diffusion-based Structured State Space Model (SSSD-ECG) with two innovations: (1) MIDT-ECG training paradigm with time-frequency domain supervision for physiological structural realism, (2) multi-modal demographic conditioning for patient-specific synthesis.

Result: Substantial improvements in morphological coherence, privacy preservation (4-8% better than baseline), 74% reduction in interlead correlation error, enhanced signal-to-noise ratio and personalization. In low-data regimes, classifiers trained with synthetic data perform comparably to those trained on real data.

Conclusion: ECG synthesizers with time-frequency structural regularization can serve as personalized, high-fidelity, privacy-preserving surrogates when real data are scarce, advancing responsible use of generative AI in healthcare.

Abstract: The development of machine learning for cardiac care is severely hampered by privacy restrictions on sharing real patient electrocardiogram (ECG) data. Although generative AI offers a promising solution, the real-world use of existing model-synthesized ECGs is limited by persistent gaps in trustworthiness and clinical utility. In this work, we address two major shortcomings of current generative ECG methods: insufficient morphological fidelity and the inability to generate personalized, patient-specific physiological signals. To address these gaps, we build on a conditional diffusion-based Structured State Space Model (SSSD-ECG) with two principled innovations: (1) MIDT-ECG (Mel-Spectrogram Informed Diffusion Training), a novel training paradigm with time-frequency domain supervision to enforce physiological structural realism, and (2) multi-modal demographic conditioning to enable patient-specific synthesis. We comprehensively evaluate our approach on the PTB-XL dataset, assessing the synthesized ECG signals on fidelity, clinical coherence, privacy preservation, and downstream task utility. MIDT-ECG achieves substantial gains: it improves morphological coherence, preserves strong privacy guarantees with all metrics evaluated exceeding the baseline by 4-8%, and notably reduces the interlead correlation error by an average of 74%, while demographic conditioning enhances signal-to-noise ratio and personalization. In critical low-data regimes, a classifier trained on datasets supplemented with our synthetic ECGs achieves performance comparable to a classifier trained solely on real data. Together, we demonstrate that ECG synthesizers, trained with the proposed time-frequency structural regularization scheme, can serve as personalized, high-fidelity, privacy-preserving surrogates when real data are scarce, advancing the responsible use of generative AI in healthcare.

Method: The authors analyze EGNN computations through a circuit-complexity lens, examining layers acting on node features, atomic coordinates, and lattice matrices. They prove that under polynomial precision with specific constraints (O(n) embedding width, O(1) layers, etc.), EGNNs admit simulation by uniform TC^0 threshold-circuit families.

Result: The analysis situates EGNNs within TC^0, providing a concrete ceiling on solvable decision and prediction problems under realistic resource constraints. This clarifies which architectural modifications are needed to transcend this computational regime.

Conclusion: The work offers a complexity-theoretic foundation for symmetry-aware graph learning on crystalline systems, complementing Weisfeiler-Lehman style results that don’t directly transfer to periodic crystals, and provides insights into the computational limits of EGNN architectures.

Abstract: Graph neural networks (GNNs) have become a core paradigm for learning on relational data. In materials science, equivariant GNNs (EGNNs) have emerged as a compelling backbone for crystalline-structure prediction, owing to their ability to respect Euclidean symmetries and periodic boundary conditions. Despite strong empirical performance, their expressive power in periodic, symmetry-constrained settings remains poorly understood. This work characterizes the intrinsic computational and expressive limits of EGNNs for crystalline-structure prediction through a circuit-complexity lens. We analyze the computations carried out by EGNN layers acting on node features, atomic coordinates, and lattice matrices, and prove that, under polynomial precision, embedding width $d=O(n)$ for $n$ nodes, $O(1)$ layers, and $O(1)$-depth, $O(n)$-width MLP instantiations of the message/update/readout maps, these models admit a simulation by a uniform $\mathsf{TC}^0$ threshold-circuit family of polynomial size (with an explicit constant-depth bound). Situating EGNNs within $\mathsf{TC}^0$ provides a concrete ceiling on the decision and prediction problems solvable by such architectures under realistic resource constraints and clarifies which architectural modifications (e.g., increased depth, richer geometric primitives, or wider layers) are required to transcend this regime. The analysis complements Weisfeiler-Lehman style results that do not directly transfer to periodic crystals, and offers a complexity-theoretic foundation for symmetry-aware graph learning on crystalline systems.

Method: Used EEG data from 52 healthy adults exposed to laser-evoked pain at three intensities. Transformed 4-second epochs into 537-feature vectors including spectral power, band ratios, Hjorth parameters, entropy, coherence, wavelet energies, and peak-frequency metrics. Evaluated 9 traditional ML models with leave-one-participant-out cross-validation.

Result: SVM with RBF kernel achieved best offline performance (88.9% accuracy, 1.02ms inference). XGBoost real-time model maintained 94.2% accuracy with 4ms latency. Feature importance aligned with pain physiology: contralateral alpha suppression, midline theta/alpha enhancement, and frontal gamma bursts.

Conclusion: EEG-based pain monitoring is technically feasible for clinical settings and provides a pathway toward clinical validation, offering objective pain assessment for patients who cannot self-report.

Abstract: Current pain assessment within hospitals often relies on self-reporting or non-specific EKG vital signs. This system leaves critically ill, sedated, and cognitively impaired patients vulnerable to undertreated pain and opioid overuse. Electroencephalography (EEG) offers a noninvasive method of measuring brain activity. This technology could potentially be applied as an assistive tool to highlight nociceptive processing in order to mitigate this issue. In this study, we compared machine learning models for classifying high-pain versus low/no-pain EEG epochs using data from fifty-two healthy adults exposed to laser-evoked pain at three intensities (low, medium, high). Each four-second epoch was transformed into a 537-feature vector spanning spectral power, band ratios, Hjorth parameters, entropy measures, coherence, wavelet energies, and peak-frequency metrics. Nine traditional machine learning models were evaluated with leave-one-participant-out cross-validation. A support vector machine with radial basis function kernel achieved the best offline performance with 88.9% accuracy and sub-millisecond inference time (1.02 ms). Our Feature importance analysis was consistent with current canonical pain physiology, showing contralateral alpha suppression, midline theta/alpha enhancement, and frontal gamma bursts. The real-time XGBoost model maintained an end-to-end latency of about 4 ms and 94.2% accuracy, demonstrating that an EEG-based pain monitor is technically feasible within a clinical setting and provides a pathway towards clinical validation.

Method: Jointly learn gradient and Hessian with GP surrogates, use one-step lookahead bound on Newton-step error, optimize in low-dimensional subspaces (random embeddings or learned sparse subspaces) to reduce computational cost from O(d²) to O(m²).

Result: NeST-BO achieves faster convergence and lower regret across high-dimensional synthetic and real-world problems, including cases with thousands of variables and unknown active subspaces, outperforming state-of-the-art local and high-dimensional BO baselines.

Conclusion: NeST-BO provides an effective approach for high-dimensional Bayesian optimization by targeting Newton steps with scalable subspace methods, inheriting inexact-Newton convergence properties while maintaining computational efficiency.

Abstract: Bayesian optimization (BO) is effective for expensive black-box problems but remains challenging in high dimensions. We propose NeST-BO, a local BO method that targets the Newton step by jointly learning gradient and Hessian information with Gaussian process surrogates, and selecting evaluations via a one-step lookahead bound on Newton-step error. We show that this bound (and hence the step error) contracts with batch size, so NeST-BO directly inherits inexact-Newton convergence: global progress under mild stability assumptions and quadratic local rates once steps are sufficiently accurate. To scale, we optimize the acquisition in low-dimensional subspaces (e.g., random embeddings or learned sparse subspaces), reducing the dominant cost of learning curvature from $O(d^2)$ to $O(m^2)$ with $m \ll d$ while preserving step targeting. Across high-dimensional synthetic and real-world problems, including cases with thousands of variables and unknown active subspaces, NeST-BO consistently yields faster convergence and lower regret than state-of-the-art local and high-dimensional BO baselines.

Method: Proposed RLHF-COV and DPO-COV algorithms that simultaneously mitigate all three issues in both offline and online settings. DPO-COV is simple to implement without reward estimation and is theoretically equivalent to RLHF-COV.

Result: Theoretical analysis shows length-regularized generalization error rates for DPO-COV trained on corrupted data match best-known rates for clean data without length regularization. Experiments demonstrate effectiveness in both offline and online settings.

Conclusion: The proposed COV algorithms effectively address multiple alignment issues simultaneously with theoretical guarantees, simple implementation, and proven equivalence between RLHF and DPO approaches.

Abstract: Reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) are important techniques to align large language models (LLM) with human preference. However, the quality of RLHF and DPO training is seriously compromised by \textit{\textbf{C}orrupted} preference, reward \textit{\textbf{O}veroptimization}, and bias towards \textit{\textbf{V}erbosity}. To our knowledge, most existing works tackle only one of these important issues, and the few other works require much computation to estimate multiple reward models and lack theoretical guarantee of generalization ability. In this work, we propose RLHF-\textbf{COV} and DPO-\textbf{COV} algorithms that can simultaneously mitigate these three issues, in both offline and online settings. This ability is theoretically demonstrated by obtaining length-regularized generalization error rates for our DPO-COV algorithms trained on corrupted data, which match the best-known rates for simpler cases with clean data and without length regularization. Moreover, our DPO-COV algorithm is simple to implement without reward estimation, and is proved to be equivalent to our RLHF-COV algorithm, which directly implies the equivalence between the vanilla RLHF and DPO algorithms. Experiments demonstrate the effectiveness of our DPO-COV algorithms under both offline and online settings.

Method: GTRANS integrates neighborhood smoothing with Gromov-Wasserstein optimal transport to align structural patterns between graphs. It includes an adaptive debiasing mechanism via residual smoothing to identify and correct target-specific deviations, preventing negative transfer.

Result: Theoretical guarantees show stability of the estimated alignment matrix. Extensive experiments on synthetic and real data demonstrate improved accuracy in target graph estimation, leading to enhanced performance in downstream applications like graph classification and link prediction.

Conclusion: GTRANS provides an effective transfer learning solution for graphon estimation in small networks, successfully leveraging structural information from larger related graphs while preventing negative transfer through adaptive debiasing mechanisms.

Abstract: Graphon models provide a flexible nonparametric framework for estimating latent connectivity probabilities in networks, enabling a range of downstream applications such as link prediction and data augmentation. However, accurate graphon estimation typically requires a large graph, whereas in practice, one often only observes a small-sized network. One approach to addressing this issue is to adopt a transfer learning framework, which aims to improve estimation in a small target graph by leveraging structural information from a larger, related source graph. In this paper, we propose a novel method, namely GTRANS, a transfer learning framework that integrates neighborhood smoothing and Gromov-Wasserstein optimal transport to align and transfer structural patterns between graphs. To prevent negative transfer, GTRANS includes an adaptive debiasing mechanism that identifies and corrects for target-specific deviations via residual smoothing. We provide theoretical guarantees on the stability of the estimated alignment matrix and demonstrate the effectiveness of GTRANS in improving the accuracy of target graph estimation through extensive synthetic and real data experiments. These improvements translate directly to enhanced performance in downstream applications, such as the graph classification task and the link prediction task.

Method: ARMOR factorizes each weight matrix into a 2:4 sparse core wrapped by two low-overhead block diagonal matrices that act as error correctors. These components are selected using a block coordinate descent algorithm that minimizes layer-wise proxy loss, with theoretical convergence guarantees.

Result: Experiments on Llama and Qwen model families show ARMOR consistently and significantly outperforms state-of-the-art 2:4 pruning methods across various downstream tasks and perplexity evaluations, while maintaining inference speedups and memory usage reductions.

Conclusion: ARMOR establishes a more effective trade-off between model compression and task accuracy compared to conventional 2:4 pruning techniques, offering superior performance without sacrificing the hardware acceleration benefits of 2:4 sparsity.

Abstract: Large language models (LLMs) present significant deployment challenges due to their immense computational and memory requirements. While semi-structured pruning, particularly 2:4 sparsity, offers a path to practical hardware acceleration, existing methods often incur substantial performance degradation. To bridge this gap, we introduce ARMOR: (Adaptive Representation with Matrix-factORization), a novel one-shot post-training pruning algorithm. Instead of directly pruning weights, ARMOR factorizes each weight matrix into a 2:4 sparse core wrapped by two low-overhead, block diagonal matrices. These wrappers act as efficient pre and post-transformation error correctors, offering greater flexibility to preserve model quality compared to conventional 2:4 pruning techniques. The sparse core and block diagonal wrappers are chosen through a block coordinate descent algorithm that minimizes a layer-wise proxy loss. We theoretically prove this optimization is guaranteed to converge to a solution with a proxy loss less than or equal to state-of-the-art pruning algorithms. Experiments on Llama (Touvron et al., 2023; Dubey et al., 2024) and Qwen (Yang et al., 2025) model families demonstrate that ARMOR consistently and significantly outperforms state-of-the-art 2:4 pruning methods across a wide range of downstream tasks and perplexity evaluations. ARMOR achieves this superior performance while retaining the inference speedups and substantial memory usage reductions of 2:4 pruning, establishing a more effective trade-off between model compression and task accuracy

Method: LATTA uses two key mechanisms: (1) noisy weight perturbation inspired by SGLD to explore local parameter space and escape poor local minima, and (2) a stable weight anchor to prevent divergence from robust source pre-training, requiring no architectural changes or expensive Monte Carlo passes.

Result: LATTA significantly outperforms Tent, CoTTA, and EATA, setting new state-of-the-art for self-supervised TTA by improving average accuracy on CIFAR-10-C by over 2% while reducing performance variance.

Conclusion: LATTA provides an effective and stable test-time adaptation approach that combines exploration through noisy updates with stability through weight anchoring, achieving superior performance without architectural modifications.

Abstract: Test-time adaptation (TTA) aims to adapt a pretrained model to distribution shifts using only unlabeled test data. While promising, existing methods like Tent suffer from instability and can catastrophically forget the source knowledge, especially with small batch sizes or challenging corruptions. We argue that this arises from overly deterministic updates on a complex loss surface. In this paper, we introduce Langevin-Anchored Test-Time Adaptation (LATTA), a novel approach that regularizes adaptation through two key mechanisms: (1) a noisy weight perturbation inspired by Stochastic Gradient Langevin Dynamics (SGLD) to explore the local parameter space and escape poor local minima, and (2) a stable weight anchor that prevents the model from diverging from its robust source pre-training. This combination allows LATTA to adapt effectively without sacrificing stability. Unlike prior Bayesian TTA methods, LATTA requires no architectural changes or expensive Monte Carlo passes. We conduct extensive experiments on standard benchmarks, including Rotated-MNIST and the more challenging CIFAR-10-C. Our results demonstrate that LATTA significantly outperforms existing methods, including Tent, CoTTA, and EATA, setting a new state of the art for self-supervised TTA by improving average accuracy on CIFAR-10-C by over 2% while simultaneously reducing performance variance.

Method: Integrates permutation-invariant embedding with policy-guided search, enhanced with privacy-preserving knowledge fusion and sample-aware weighting for federated learning environments.

Result: Extensive experiments validate effectiveness, robustness, efficiency, and strong generalization ability in federated learning scenarios.

Conclusion: The framework successfully addresses key challenges in federated feature selection while maintaining privacy and handling data heterogeneity.

Abstract: Feature selection eliminates redundancy among features to improve downstream task performance while reducing computational overhead. Existing methods often struggle to capture intricate feature interactions and adapt across diverse application scenarios. Recent advances employ generative intelligence to alleviate these drawbacks. However, these methods remain constrained by permutation sensitivity in embedding and reliance on convexity assumptions in gradient-based search. To address these limitations, our initial work introduces a novel framework that integrates permutation-invariant embedding with policy-guided search. Although effective, it still left opportunities to adapt to realistic distributed scenarios. In practice, data across local clients is highly imbalanced, heterogeneous and constrained by strict privacy regulations, limiting direct sharing. These challenges highlight the need for a framework that can integrate feature selection knowledge across clients without exposing sensitive information. In this extended journal version, we advance the framework from two perspectives: 1) developing a privacy-preserving knowledge fusion strategy to derive a unified representation space without sharing sensitive raw data. 2) incorporating a sample-aware weighting strategy to address distributional imbalance among heterogeneous local clients. Extensive experiments validate the effectiveness, robustness, and efficiency of our framework. The results further demonstrate its strong generalization ability in federated learning scenarios. The code and data are publicly available: https://anonymous.4open.science/r/FedCAPS-08BF.

Method: The authors analyze a simplified tractable model that magnifies attention scaling effects. They study phase transitions governed by scaling factor β_n and identify the critical scaling threshold.

Result: The analysis reveals a phase transition: insufficient scaling causes all tokens to collapse to a single direction, while excessive scaling reduces attention to identity. The critical scaling is β_n ∝ log n, which maintains sparse, content-adaptive attention.

Conclusion: The paper provides rigorous justification for attention scaling in models like YaRN and Qwen, explaining why logarithmic scaling preserves meaningful token interactions while preventing rank-collapse at large context lengths.

Abstract: As large language models scale to longer contexts, attention layers suffer from a fundamental pathology: attention scores collapse toward uniformity as context length $n$ increases, causing tokens to cluster excessively, a phenomenon known as rank-collapse. While $\textit{attention scaling}$ effectively addresses this deficiency by rescaling attention scores with a polylogarithmic factor $\beta_n$, theoretical justification for this approach remains lacking. We analyze a simplified yet tractable model that magnifies the effect of attention scaling. In this model, attention exhibits a phase transition governed by the scaling factor $\beta_n$: insufficient scaling collapses all tokens to a single direction, while excessive scaling reduces attention to identity, thereby eliminating meaningful interactions between tokens. Our main result identifies the critical scaling $\beta_n \asymp \log n$ and provides a rigorous justification for attention scaling in YaRN and Qwen, clarifying why logarithmic scaling maintains sparse, content-adaptive attention at large context lengths.

Method: Uses Generative Dynamic Graph Model (GDGM) with neural ODEs and GRUs to model temporal dynamics, plus pseudo-label generation and heterogeneous aggregation to enhance conspiracy spoofing detection.

Result: Outperforms state-of-the-art models in detection accuracy and has been successfully deployed in one of the largest global trading markets.

Conclusion: The proposed GDGM framework effectively captures complex dynamic trading behaviors and relationships, demonstrating practical applicability and superior performance in real-world financial spoofing detection.

Abstract: Spoofing detection in financial trading is crucial, especially for identifying complex behaviors such as conspiracy spoofing. Traditional machine-learning approaches primarily focus on isolated node features, often overlooking the broader context of interconnected nodes. Graph-based techniques, particularly Graph Neural Networks (GNNs), have advanced the field by leveraging relational information effectively. However, in real-world spoofing detection datasets, trading behaviors exhibit dynamic, irregular patterns. Existing spoofing detection methods, though effective in some scenarios, struggle to capture the complexity of dynamic and diverse, evolving inter-node relationships. To address these challenges, we propose a novel framework called the Generative Dynamic Graph Model (GDGM), which models dynamic trading behaviors and the relationships among nodes to learn representations for conspiracy spoofing detection. Specifically, our approach incorporates the generative dynamic latent space to capture the temporal patterns and evolving market conditions. Raw trading data is first converted into time-stamped sequences. Then we model trading behaviors using the neural ordinary differential equations and gated recurrent units, to generate the representation incorporating temporal dynamics of spoofing patterns. Furthermore, pseudo-label generation and heterogeneous aggregation techniques are employed to gather relevant information and enhance the detection performance for conspiratorial spoofing behaviors. Experiments conducted on spoofing detection datasets demonstrate that our approach outperforms state-of-the-art models in detection accuracy. Additionally, our spoofing detection system has been successfully deployed in one of the largest global trading markets, further validating the practical applicability and performance of the proposed method.

Method: Proposed Temporal Graph Autoencoder (TGAE) with attention-based graph encoder to encode temporal and structural characteristics on sampled ego-graphs, and an ego-graph decoder for efficient generation.

Result: Experimental evaluation shows TGAE outperforms state-of-the-art temporal graph generators in both simulation quality and efficiency on real-life and synthesized temporal graphs.

Conclusion: TGAE provides an efficient learning-based solution for temporal graph simulation that achieves good trade-off between quality and efficiency, addressing limitations of existing temporal graph generators.

Abstract: Graph simulation has recently received a surge of attention in graph processing and analytics. In real-life applications, e.g. social science, biology, and chemistry, many graphs are composed of a series of evolving graphs (i.e., temporal graphs). While most of the existing graph generators focus on static graphs, the temporal information of the graphs is ignored. In this paper, we focus on simulating temporal graphs, which aim to reproduce the structural and temporal properties of the observed real-life temporal graphs. In this paper, we first give an overview of the existing temporal graph generators, including recently emerged learning-based approaches. Most of these learning-based methods suffer from one of the limitations: low efficiency in training or slow generating, especially for temporal random walk-based methods. Therefore, we propose an efficient learning-based approach to generate graph snapshots, namely temporal graph autoencoder (TGAE). Specifically, we propose an attention-based graph encoder to encode temporal and structural characteristics on sampled ego-graphs. And we proposed an ego-graph decoder that can achieve a good trade-off between simulation quality and efficiency in temporal graph generation. Finally, the experimental evaluation is conducted among our proposed TGAE and representative temporal graph generators on real-life temporal graphs and synthesized graphs. It is reported that our proposed approach outperforms the state-of-the-art temporal graph generators by means of simulation quality and efficiency.

Method: Learn a privacy encoding network with a utility generation network to create embeddings with formal differential privacy guarantees. These privatized embeddings are shared with a server that learns post-processing for higher accuracy.

Result: The approach requires only one round of privatized communication and less client computation than traditional methods. The privatized embeddings are agnostic to server model types (deep learning, random forests, XGBoost).

Conclusion: The co-design of collaborative and private learning enables efficient, privacy-preserving collaborative learning with formal guarantees and reduced computational requirements.

Abstract: Traditional collaborative learning approaches are based on sharing of model weights between clients and a server. However, there are advantages to resource efficiency through schemes based on sharing of embeddings (activations) created from the data. Several differentially private methods were developed for sharing of weights while such mechanisms do not exist so far for sharing of embeddings. We propose Ours to learn a privacy encoding network in conjunction with a small utility generation network such that the final embeddings generated from it are equipped with formal differential privacy guarantees. These privatized embeddings are then shared with a more powerful server, that learns a post-processing that results in a higher accuracy for machine learning tasks. We show that our co-design of collaborative and private learning results in requiring only one round of privatized communication and lesser compute on the client than traditional methods. The privatized embeddings that we share from the client are agnostic to the type of model (deep learning, random forests or XGBoost) used on the server in order to process these activations to complete a task.

Method: Proposes InfoRMIA, an information-theoretic formulation of membership inference, and extends it to token-level analysis to localize memorization.

Result: InfoRMIA consistently outperforms RMIA across benchmarks with improved computational efficiency, and token-level analysis achieves stronger sequence-level inference while pinpointing memorized tokens.

Conclusion: Token-level membership inference provides a more granular approach to privacy assessment in LLMs, enabling targeted mitigation strategies like exact unlearning.

Abstract: Machine learning models are known to leak sensitive information, as they inevitably memorize (parts of) their training data. More alarmingly, large language models (LLMs) are now trained on nearly all available data, which amplifies the magnitude of information leakage and raises serious privacy risks. Hence, it is more crucial than ever to quantify privacy risk before the release of LLMs. The standard method to quantify privacy is via membership inference attacks, where the state-of-the-art approach is the Robust Membership Inference Attack (RMIA). In this paper, we present InfoRMIA, a principled information-theoretic formulation of membership inference. Our method consistently outperforms RMIA across benchmarks while also offering improved computational efficiency. In the second part of the paper, we identify the limitations of treating sequence-level membership inference as the gold standard for measuring leakage. We propose a new perspective for studying membership and memorization in LLMs: token-level signals and analyses. We show that a simple token-based InfoRMIA can pinpoint which tokens are memorized within generated outputs, thereby localizing leakage from the sequence level down to individual tokens, while achieving stronger sequence-level inference power on LLMs. This new scope rethinks privacy in LLMs and can lead to more targeted mitigation, such as exact unlearning.

Method: Developed a unified, reproducible benchmarking framework (HydraGNN) that enables seamless switching among four controlled model classes: MPNN, MPNN with encoders, GPS-style hybrids, and fully fused local-global models with encoders. Evaluated on seven diverse open-source datasets across regression and classification tasks.

Result: Encoder-augmented MPNNs form a robust baseline, while fused local-global models yield the clearest benefits for properties governed by long-range interaction effects. The study quantifies the accuracy-compute trade-offs and memory overhead of attention mechanisms.

Conclusion: This work establishes the first controlled evaluation of global attention in atomistic graph learning and provides a reproducible testbed for future model development, clarifying when different architectural components provide meaningful improvements.

Abstract: Graph neural networks (GNNs) are widely used as surrogates for costly experiments and first-principles simulations to study the behavior of compounds at atomistic scale, and their architectural complexity is constantly increasing to enable the modeling of complex physics. While most recent GNNs combine more traditional message passing neural networks (MPNNs) layers to model short-range interactions with more advanced graph transformers (GTs) with global attention mechanisms to model long-range interactions, it is still unclear when global attention mechanisms provide real benefits over well-tuned MPNN layers due to inconsistent implementations, features, or hyperparameter tuning. We introduce the first unified, reproducible benchmarking framework - built on HydraGNN - that enables seamless switching among four controlled model classes: MPNN, MPNN with chemistry/topology encoders, GPS-style hybrids of MPNN with global attention, and fully fused local - global models with encoders. Using seven diverse open-source datasets for benchmarking across regression and classification tasks, we systematically isolate the contributions of message passing, global attention, and encoder-based feature augmentation. Our study shows that encoder-augmented MPNNs form a robust baseline, while fused local-global models yield the clearest benefits for properties governed by long-range interaction effects. We further quantify the accuracy - compute trade-offs of attention, reporting its overhead in memory. Together, these results establish the first controlled evaluation of global attention in atomistic graph learning and provide a reproducible testbed for future model development.

Method: Three components: (1) dual-branch invariant disentangled feature learning with season-trend decomposition, (2) lightweight proxy denoising to calibrate LLM biases, (3) bidirectional knowledge distillation between denoised and original predictions.

Result: Outperforms state-of-the-art baselines by 9.3% on average across real-world datasets.

Conclusion: TimePD effectively enables source-free domain adaptation for time series forecasting while maintaining data privacy, demonstrating the value of combining LLMs with specialized denoising techniques.

Abstract: The proliferation of mobile devices generates a massive volume of time series across various domains, where effective time series forecasting enables a variety of real-world applications. This study focuses on a new problem of source-free domain adaptation for time series forecasting. It aims to adapt a pretrained model from sufficient source time series to the sparse target time series domain without access to the source data, embracing data protection regulations. To achieve this, we propose TimePD, the first source-free time series forecasting framework with proxy denoising, where large language models (LLMs) are employed to benefit from their generalization capabilities. Specifically, TimePD consists of three key components: (1) dual-branch invariant disentangled feature learning that enforces representation- and gradient-wise invariance by means of season-trend decomposition; (2) lightweight, parameter-free proxy denoising that dynamically calibrates systematic biases of LLMs; and (3) knowledge distillation that bidirectionally aligns the denoised prediction and the original target prediction. Extensive experiments on real-world datasets offer insight into the effectiveness of the proposed TimePD, outperforming SOTA baselines by 9.3% on average.

Method: Analytically linking chaotic learning dynamics and symmetry-induced invariant subspaces to derive sufficient conditions for riddled basins in realistic deep networks.

Result: Revealed basins of attraction with infinitely fine-scale fractal structure characterized by near-zero uncertainty exponent, making outcome predictability insensitive to increased precision of initial conditions.

Conclusion: Riddling imposes fundamental limits on predictability and reproducibility of neural network training, providing unified explanation for empirical observations with implications for optimization and safe AI deployment.

Abstract: Fundamental limits to predictability are central to our understanding of many physical and computational systems. Here we show that, despite its remarkable capabilities, deep learning exhibits such fundamental limits rooted in the fractal, riddled geometry of its basins of attraction: any initialization that leads to one solution lies arbitrarily close to another that leads to a different one. We derive sufficient conditions for the emergence of riddled basins by analytically linking features widely observed in deep learning, including chaotic learning dynamics and symmetry-induced invariant subspaces, to reveal a general route to riddling in realistic deep networks. The resulting basins of attraction possess an infinitely fine-scale fractal structure characterized by an uncertainty exponent near zero, so that even large increases in the precision of initial conditions yield only marginal gains in outcome predictability. Riddling thus imposes a fundamental limit on the predictability and hence reproducibility of neural network training, providing a unified account of many empirical observations. These results reveal a general organizing principle of deep learning with important implications for optimization and the safe deployment of artificial intelligence.

Method: Represents kernel as learnable tensor over sampled input-output pairs, performs uniform random sampling once from discretized grid, uses Monte Carlo-type approach to approximate integral operator, and includes interpolation step for grid flexibility.

Result: Achieves competitive accuracy with efficient computational cost on standard 1D PDE benchmarks, with theoretical analysis showing bounded bias and variance under mild regularity assumptions.

Conclusion: MCNO provides a theoretically supported alternative to spectral methods like FNO and graph-based Monte Carlo approaches, with potential for natural extension beyond one-dimensional problems.

Abstract: The Monte Carlo-type Neural Operator (MCNO) introduces a framework for learning solution operators of one-dimensional partial differential equations (PDEs) by directly learning the kernel function and approximating the associated integral operator using a Monte Carlo-type approach. Unlike Fourier Neural Operators (FNOs), which rely on spectral representations and assume translation-invariant kernels, MCNO makes no such assumptions. The kernel is represented as a learnable tensor over sampled input-output pairs, and sampling is performed once, uniformly at random from a discretized grid. This design enables generalization across multiple grid resolutions without relying on fixed global basis functions or repeated sampling during training, while an interpolation step maps between arbitrary input and output grids to further enhance flexibility. Experiments on standard 1D PDE benchmarks show that MCNO achieves competitive accuracy with efficient computational cost. We also provide a theoretical analysis proving that the Monte Carlo estimator yields a bounded bias and variance under mild regularity assumptions. This result holds in any spatial dimension, suggesting that MCNO may extend naturally beyond one-dimensional problems. More broadly, this work explores how Monte Carlo-type integration can be incorporated into neural operator frameworks for continuous-domain PDEs, providing a theoretically supported alternative to spectral methods (such as FNO) and to graph-based Monte Carlo approaches (such as the Graph Kernel Neural Operator, GNO).

Method: NEO re-centers target data embeddings at the origin based on theoretical insights about latent space geometry. It’s hyperparameter-free and adds minimal compute overhead compared to standard inference.

Result: NEO improves ViT-Base accuracy on ImageNet-C from 55.6% to 59.2% with just one batch of 64 samples. It outperforms 7 TTA methods on ImageNet-C/R/S and 6/7 on CIFAR-10-C while using least compute. Reduces inference time by 63% and memory by 9% on edge devices.

Conclusion: NEO provides an efficient and effective TTA method that works well across multiple ViT architectures and datasets, demonstrating practical utility for real-world deployment on resource-constrained devices.

Abstract: Test-Time Adaptation (TTA) methods are often computationally expensive, require a large amount of data for effective adaptation, or are brittle to hyperparameters. Based on a theoretical foundation of the geometry of the latent space, we are able to significantly improve the alignment between source and distribution-shifted samples by re-centering target data embeddings at the origin. This insight motivates NEO – a hyperparameter-free fully TTA method, that adds no significant compute compared to vanilla inference. NEO is able to improve the classification accuracy of ViT-Base on ImageNet-C from 55.6% to 59.2% after adapting on just one batch of 64 samples. When adapting on 512 samples NEO beats all 7 TTA methods we compare against on ImageNet-C, ImageNet-R and ImageNet-S and beats 6/7 on CIFAR-10-C, while using the least amount of compute. NEO performs well on model calibration metrics and additionally is able to adapt from 1 class to improve accuracy on 999 other classes in ImageNet-C. On Raspberry Pi and Jetson Orin Nano devices, NEO reduces inference time by 63% and memory usage by 9% compared to baselines. Our results based on 3 ViT architectures and 4 datasets show that NEO can be used efficiently and effectively for TTA.

Method: Developed a unified probabilistic concept sidechannel meta-model, introduced Sidechannel Independence Score (SIS) to quantify sidechannel reliance, and proposed SIS regularization to penalize this reliance while maintaining accuracy.

Result: Empirical results show state-of-the-art CSMs trained for accuracy alone have low interpretability, but SIS regularization substantially improves interpretability, intervenability, and quality of learned interpretable predictors.

Conclusion: The work provides theoretical and practical tools for developing CSMs that can balance accuracy and interpretability in a principled manner through controlled sidechannel reliance.

Abstract: Concept Bottleneck Models (CBNMs) are deep learning models that provide interpretability by enforcing a bottleneck layer where predictions are based exclusively on human-understandable concepts. However, this constraint also restricts information flow and often results in reduced predictive accuracy. Concept Sidechannel Models (CSMs) address this limitation by introducing a sidechannel that bypasses the bottleneck and carry additional task-relevant information. While this improves accuracy, it simultaneously compromises interpretability, as predictions may rely on uninterpretable representations transmitted through sidechannels. Currently, there exists no principled technique to control this fundamental trade-off. In this paper, we close this gap. First, we present a unified probabilistic concept sidechannel meta-model that subsumes existing CSMs as special cases. Building on this framework, we introduce the Sidechannel Independence Score (SIS), a metric that quantifies a CSM’s reliance on its sidechannel by contrasting predictions made with and without sidechannel information. We propose SIS regularization, which explicitly penalizes sidechannel reliance to improve interpretability. Finally, we analyze how the expressivity of the predictor and the reliance of the sidechannel jointly shape interpretability, revealing inherent trade-offs across different CSM architectures. Empirical results show that state-of-the-art CSMs, when trained solely for accuracy, exhibit low representation interpretability, and that SIS regularization substantially improves their interpretability, intervenability, and the quality of learned interpretable task predictors. Our work provides both theoretical and practical tools for developing CSMs that balance accuracy and interpretability in a principled manner.

Method: Developed an inductive graph gradient boosting machine (G-GBM) that combines graph learning with gradient boosting, tested on simulated random graphs and applied to insurance fraud detection using open-source and proprietary datasets.

Result: G-GBM competes with popular graph neural network approaches in experiments and demonstrates effective fraud detection performance, with the added benefit of explainability through established gradient boosting interpretability methods.

Conclusion: G-GBM provides a powerful alternative to graph neural networks for insurance fraud detection, offering competitive performance while maintaining the explainability advantages of gradient boosting models.

Abstract: Graph-based methods are becoming increasingly popular in machine learning due to their ability to model complex data and relations. Insurance fraud is a prime use case, since false claims are often the result of organised criminals that stage accidents or the same persons filing erroneous claims on multiple policies. One challenge is that graph-based approaches struggle to find meaningful representations of the data because of the high class imbalance present in fraud data. Another is that insurance networks are heterogeneous and dynamic, given the changing relations among people, companies and policies. That is why gradient boosted tree approaches on tabular data still dominate the field. Therefore, we present a novel inductive graph gradient boosting machine (G-GBM) for supervised learning on heterogeneous and dynamic graphs. We show that our estimator competes with popular graph neural network approaches in an experiment using a variety of simulated random graphs. We demonstrate the power of G-GBM for insurance fraud detection using an open-source and a real-world, proprietary dataset. Given that the backbone model is a gradient boosting forest, we apply established explainability methods to gain better insights into the predictions made by G-GBM.

Method: Uses local surrogates fit on structure-preserving perturbations of graphs, aggregates surrogate attributions with their dispersion, and provides Dvoretzky-Kiefer-Wolfowitz bound for finite-sample guarantee on surrogate ensemble size.

Result: Empirical studies on controlled synthetic graphs with known ground truth demonstrate accurate and stable explanations, with ablations showing benefits of nonlinear surrogate modeling and sensitivity to perturbation design.

Conclusion: Establishes a principled, uncertainty-aware, and structure-sensitive approach to explaining quantum graph neural networks, laying groundwork for scaling to broader architectures and real-world datasets as quantum resources mature.

Abstract: Quantum graph neural networks offer a powerful paradigm for learning on graph-structured data, yet their explainability is complicated by measurement-induced stochasticity and the combinatorial nature of graph structure. In this paper, we introduce QuantumGraphLIME (QGraphLIME), a model-agnostic, post-hoc framework that treats model explanations as distributions over local surrogates fit on structure-preserving perturbations of a graph. By aggregating surrogate attributions together with their dispersion, QGraphLIME yields uncertainty-aware node and edge importance rankings for quantum graph models. The framework further provides a distribution-free, finite-sample guarantee on the size of the surrogate ensemble: a Dvoretzky-Kiefer-Wolfowitz bound ensures uniform approximation of the induced distribution of a binary class probability at target accuracy and confidence under standard independence assumptions. Empirical studies on controlled synthetic graphs with known ground truth demonstrate accurate and stable explanations, with ablations showing clear benefits of nonlinear surrogate modeling and highlighting sensitivity to perturbation design. Collectively, these results establish a principled, uncertainty-aware, and structure-sensitive approach to explaining quantum graph neural networks, and lay the groundwork for scaling to broader architectures and real-world datasets, as quantum resources mature. Code is available at https://github.com/smlab-niser/qglime.

Method: vAttention unifies top-k and random sampling approaches, leveraging their complementary strengths. It provides user-specified (ε, δ) statistical guarantees on approximation accuracy, making it the first practical sparse attention with verified accuracy guarantees.

Result: vAttention significantly improves sparse attention quality (∼4.5 percentage points for Llama-3.1-8B-Inst and Deepseek-R1-Distill-Llama-8B on RULER-HARD), matches full model quality with up to 20x sparsity across datasets, and achieves full model quality on AIME2024 at 10x sparsity with up to 32K token generations.

Conclusion: vAttention is a compelling step toward practical, reliable deployment of sparse attention at scale, delivering superior quality-efficiency trade-off and enabling fast decoding without compromising model quality in reasoning scenarios.

Abstract: State-of-the-art sparse attention methods for reducing decoding latency fall into two main categories: approximate top-$k$ (and its extension, top-$p$) and recently introduced sampling-based estimation. However, these approaches are fundamentally limited in their ability to approximate full attention: they fail to provide consistent approximations across heads and query vectors and, most critically, lack guarantees on approximation quality, limiting their practical deployment. We observe that top-$k$ and random sampling are complementary: top-$k$ performs well when attention scores are dominated by a few tokens, whereas random sampling provides better estimates when attention scores are relatively uniform. Building on this insight and leveraging the statistical guarantees of sampling, we introduce vAttention, the first practical sparse attention mechanism with user-specified $(\epsilon, \delta)$ guarantees on approximation accuracy (thus, verified). These guarantees make vAttention a compelling step toward practical, reliable deployment of sparse attention at scale. By unifying top-k and sampling, vAttention outperforms both individually, delivering a superior quality-efficiency trade-off. Our experiments show that vAttention significantly improves the quality of sparse attention (e.g., $\sim$4.5 percentage points for Llama-3.1-8B-Inst and Deepseek-R1-Distill-Llama-8B on RULER-HARD), and effectively bridges the gap between full and sparse attention (e.g., across datasets, it matches full model quality with upto 20x sparsity). We also demonstrate that it can be deployed in reasoning scenarios to achieve fast decoding without compromising model quality (e.g., vAttention achieves full model quality on AIME2024 at 10x sparsity with up to 32K token generations). Code is open-sourced at https://github.com/xAlg-ai/sparse-attention-hub.

Method: First trains a model using standard DPO on reward preference data, then uses a rearranged Lagrangian DPO objective with the reward information to fine-tune LLMs on cost preference data. Also extends to online data with exploration bonuses.

Result: Experimental results on PKU-SafeRLHF dataset demonstrate effectiveness. Approach significantly reduces memory and computational costs while providing theoretical guarantees on suboptimality and constraint violation.

Conclusion: The proposed primal-dual DPO approach effectively addresses constrained alignment in LLMs with reduced computational requirements and strong theoretical guarantees, making safe alignment more practical and scalable.

Abstract: The widespread application of Large Language Models (LLMs) imposes increasing demands on safety, such as reducing harmful content and fake information, and avoiding certain forbidden tokens due to rules and laws. While there have been several recent works studying safe alignment of LLMs, these works either require the training of reward and cost models and incur high memory and computational costs, or need prior knowledge about the optimal solution. Motivated by this fact, we study the problem of constrained alignment in LLMs, i.e., maximizing the output reward while restricting the cost due to potentially unsafe content to stay below a threshold. For this problem, we propose a novel primal-dual DPO approach, which first trains a model using standard DPO on reward preference data to provide reward information, and then adopts a rearranged Lagrangian DPO objective utilizing the provided reward information to fine-tune LLMs on cost preference data. Our approach significantly reduces memory and computational costs, and does not require extra prior knowledge. Moreover, we establish rigorous theoretical guarantees on the suboptimality and constraint violation of the output policy. We also extend our approach to an online data setting by incorporating exploration bonuses, which enables our approach to explore uncovered prompt-response space, and then provide theoretical results that get rid of the dependence on preference data coverage. Experimental results on the widely-used preference dataset PKU-SafeRLHF demonstrate the effectiveness of our approach.

Method: DiffSDA uses a modal-agnostic framework combining probabilistic modeling, latent diffusion, and efficient samplers for sequential disentanglement across time series, video, and audio data.

Result: Experiments on diverse real-world benchmarks show DiffSDA outperforms recent state-of-the-art methods in sequential disentanglement performance.

Conclusion: DiffSDA provides an effective diffusion-based solution for sequential disentanglement that works across multiple data modalities and establishes a rigorous evaluation protocol for real-world settings.

Abstract: Unsupervised representation learning, particularly sequential disentanglement, aims to separate static and dynamic factors of variation in data without relying on labels. This remains a challenging problem, as existing approaches based on variational autoencoders and generative adversarial networks often rely on multiple loss terms, complicating the optimization process. Furthermore, sequential disentanglement methods face challenges when applied to real-world data, and there is currently no established evaluation protocol for assessing their performance in such settings. Recently, diffusion models have emerged as state-of-the-art generative models, but no theoretical formalization exists for their application to sequential disentanglement. In this work, we introduce the Diffusion Sequential Disentanglement Autoencoder (DiffSDA), a novel, modal-agnostic framework effective across diverse real-world data modalities, including time series, video, and audio. DiffSDA leverages a new probabilistic modeling, latent diffusion, and efficient samplers, while incorporating a challenging evaluation protocol for rigorous testing. Our experiments on diverse real-world benchmarks demonstrate that DiffSDA outperforms recent state-of-the-art methods in sequential disentanglement.

Method: Develops Neighborhood-Adaptive Generalized Linear Graph Embedding (NGLGE) with adaptive graph learning for neighborhoods, reconstructed low-rank representation, and ℓ₂,₀ norm constraint on projection matrix to explore additional pattern information. Includes an efficient iterative solving algorithm.

Result: Comparative evaluations on diverse datasets demonstrate superior performance compared to state-of-the-art methods.

Conclusion: NGLGE effectively addresses limitations of current graph embedding methods by adaptively learning neighborhood structures and discovering latent patterns through flexible constraints.

Abstract: Graph embedding has been widely applied in areas such as network analysis, social network mining, recommendation systems, and bioinformatics. However, current graph construction methods often require the prior definition of neighborhood size, limiting the effective revelation of potential structural correlations in the data. Additionally, graph embedding methods using linear projection heavily rely on a singular pattern mining approach, resulting in relative weaknesses in adapting to different scenarios. To address these challenges, we propose a novel model, Neighborhood-Adaptive Generalized Linear Graph Embedding (NGLGE), grounded in latent pattern mining. This model introduces an adaptive graph learning method tailored to the neighborhood, effectively revealing intrinsic data correlations. Simultaneously, leveraging a reconstructed low-rank representation and imposing $\ell_{2,0}$ norm constraint on the projection matrix allows for flexible exploration of additional pattern information. Besides, an efficient iterative solving algorithm is derived for the proposed model. Comparative evaluations on datasets from diverse scenarios demonstrate the superior performance of our model compared to state-of-the-art methods.

Method: Formulate denoising as a KL-regularized Markov decision process with explicit reference policy. Optimize a regularized objective that provides policy improvement and convergence guarantees.

Result: Empirical results across four benchmarks show consistent outperformance over max-confidence scheduling. On SUDOKU, achieves 20.1% gain over random and 11.2% gain over max-confidence.

Conclusion: Learned scheduling policies generate samples that more closely match the data distribution than heuristic schedules, providing theoretical guarantees and empirical improvements.

Abstract: Masked diffusion models (MDMs) have recently emerged as a novel framework for language modeling. MDMs generate sentences by iteratively denoising masked sequences, filling in [MASK] tokens step by step. Although MDMs support any-order sampling, performance is highly sensitive to the choice of which position to unmask next. Prior work typically relies on rule-based schedules (e.g., max-confidence, max-margin), which provide ad hoc improvements. In contrast, we replace these heuristics with a learned scheduler. Specifically, we cast denoising as a KL-regularized Markov decision process (MDP) with an explicit reference policy and optimize a regularized objective that admits policy improvement and convergence guarantees under standard assumptions. We prove that the optimized policy under this framework generates samples that more closely match the data distribution than heuristic schedules. Empirically, across four benchmarks, our learned policy consistently outperforms max-confidence: for example, on SUDOKU, where unmasking order is critical, it yields a 20.1% gain over random and a 11.2% gain over max-confidence.

Method: Tested two approaches: direct communication (one-word “cheap talk” channel) and curriculum learning (pedagogical curriculum through progressively complex games) in 4-player Stag Hunt and Iterated Public Goods Game with Punishment.

Result: Communication increased cooperation from 0% to 48.3%, while curriculum learning reduced agent payoffs by 27.4% and induced “learned pessimism” when emphasizing defection-equilibrium games.

Conclusion: Simple communication protocols are more reliable than experience-based training for coordination problems, and curriculum design for social dilemmas requires careful attention to strategic lessons in game sequences.

Abstract: Eliciting cooperation in multi-agent LLM systems is critical for AI alignment. We investigate two approaches: direct communication and curriculum learning. In a 4-player Stag Hunt, a one-word “cheap talk” channel increases cooperation from 0% to 48.3%, demonstrating communication as a robust coordination mechanism. In contrast, we find that curriculum learning is highly sensitive to design choices: our pedagogical curriculum through progressively complex games reduced agent payoffs by 27.4% in an Iterated Public Goods Game with Punishment. Qualitative analysis reveals that curricula emphasizing defection-equilibrium games can induce “learned pessimism” in agents. These findings suggest that for coordination problems, simple communication protocols may be more reliable than experience-based training, and that curriculum design for social dilemmas requires careful attention to the strategic lessons embedded in game sequences.

Method: Systematic reproduction across 21 datasets and 20 baselines with hyperparameter retuning, plus a causal effect estimation framework using factual/counterfactual analyses with robustness validation.

Result: Model architecture/complexity has no causal effect on performance; heterogeneous information has positive causal effect by increasing homophily and local-global distribution discrepancy.

Conclusion: HGNNs’ performance gains come from heterogeneous information making node classes more distinguishable, not from architectural complexity.

Abstract: Graph neural networks (GNNs) have achieved remarkable success in node classification. Building on this progress, heterogeneous graph neural networks (HGNNs) integrate relation types and node and edge semantics to leverage heterogeneous information. Causal analysis for HGNNs is advancing rapidly, aiming to separate genuine causal effects from spurious correlations. However, whether HGNNs are intrinsically effective remains underexamined, and most studies implicitly assume rather than establish this effectiveness. In this work, we examine HGNNs from two perspectives: model architecture and heterogeneous information. We conduct a systematic reproduction across 21 datasets and 20 baselines, complemented by comprehensive hyperparameter retuning. To further disentangle the source of performance gains, we develop a causal effect estimation framework that constructs and evaluates candidate factors under standard assumptions through factual and counterfactual analyses, with robustness validated via minimal sufficient adjustment sets, cross-method consistency checks, and sensitivity analyses. Our results lead to two conclusions. First, model architecture and complexity have no causal effect on performance. Second, heterogeneous information exerts a positive causal effect by increasing homophily and local-global distribution discrepancy, which makes node classes more distinguishable. The implementation is publicly available at https://github.com/YXNTU/CausalHGNN.

Method: The authors compared performance of diverse MIAs in transfer learning settings, evaluating different attack methods across various experimental scenarios and datasets including PatchCamelyon.

Result: Attack efficacy decreases with increased training data for score-based MIAs. LiRA performs best in most scenarios, while IHA is more effective against models fine-tuned on PatchCamelyon in high data regimes.

Conclusion: No single MIA captures all privacy risks in transfer learning models, and practitioners need to consider multiple attack types for comprehensive privacy risk evaluation.

Abstract: With the emergence of powerful large-scale foundation models, the training paradigm is increasingly shifting from from-scratch training to transfer learning. This enables high utility training with small, domain-specific datasets typical in sensitive applications.Membership inference attacks (MIAs) provide an empirical estimate of the privacy leakage by machine learning models. Yet, prior assessments of MIAs against models fine-tuned with transfer learning rely on a small subset of possible attacks. We address this by comparing performance of diverse MIAs in transfer learning settings to help practitioners identify the most efficient attacks for privacy risk evaluation. We find that attack efficacy decreases with the increase in training data for score-based MIAs. We find that there is no one MIA which captures all privacy risks in models trained with transfer learning. While the Likelihood Ratio Attack (LiRA) demonstrates superior performance across most experimental scenarios, the Inverse Hessian Attack (IHA) proves to be more effective against models fine-tuned on PatchCamelyon dataset in high data regime.

Method: Generate synthetic SNP sequences using time-inhomogeneous HMMs with bounded gradient contributions from each SNP sequence during training, providing strong differential privacy guarantees while handling inherent SNP correlations.

Result: Experiments on 1000 Genomes dataset show effective performance with privacy budgets ε∈[1,10] at δ=10^-4. Time-inhomogeneous HMMs significantly enhance performance, enabling synthetic datasets to closely replicate statistical properties of non-private datasets.

Conclusion: The framework enables private sharing of genomic data while providing exceptional flexibility and utility for researchers, directly addressing privacy risks from SNP correlations.

Abstract: Single nucleotide polymorphism (SNP) datasets are fundamental to genetic studies but pose significant privacy risks when shared. The correlation of SNPs with each other makes strong adversarial attacks such as masked-value reconstruction, kin, and membership inference attacks possible. Existing privacy-preserving approaches either apply differential privacy to statistical summaries of these datasets or offer complex methods that require post-processing and the usage of a publicly available dataset to suppress or selectively share SNPs. In this study, we introduce an innovative framework for generating synthetic SNP sequence datasets using samples derived from time-inhomogeneous hidden Markov models (TIHMMs). To preserve the privacy of the training data, we ensure that each SNP sequence contributes only a bounded influence during training, enabling strong differential privacy guarantees. Crucially, by operating on full SNP sequences and bounding their gradient contributions, our method directly addresses the privacy risks introduced by their inherent correlations. Through experiments conducted on the real-world 1000 Genomes dataset, we demonstrate the efficacy of our method using privacy budgets of $\varepsilon \in [1, 10]$ at $\delta=10^{-4}$. Notably, by allowing the transition models of the HMM to be dependent on the location in the sequence, we significantly enhance performance, enabling the synthetic datasets to closely replicate the statistical properties of non-private datasets. This framework facilitates the private sharing of genomic data while offering researchers exceptional flexibility and utility.

Method: Replace full SGD trajectories with smooth quadratic Bézier curves that connect initial and final model states from real training trajectories, providing noise-free, low-curvature supervision signals.

Result: Outperforms state-of-the-art condensation approaches across five clinical datasets, yielding condensed datasets that enable clinically effective model development.

Conclusion: Bézier-mode connections serve as effective surrogates for SGD paths, stabilizing gradients, accelerating convergence, and eliminating dense trajectory storage requirements while maintaining clinical utility.

Abstract: Dataset condensation (DC) enables the creation of compact, privacy-preserving synthetic datasets that can match the utility of real patient records, supporting democratised access to highly regulated clinical data for developing downstream clinical models. State-of-the-art DC methods supervise synthetic data by aligning the training dynamics of models trained on real and those trained on synthetic data, typically using full stochastic gradient descent (SGD) trajectories as alignment targets; however, these trajectories are often noisy, high-curvature, and storage-intensive, leading to unstable gradients, slow convergence, and substantial memory overhead. We address these limitations by replacing full SGD trajectories with smooth, low-loss parametric surrogates, specifically quadratic B'ezier curves that connect the initial and final model states from real training trajectories. These mode-connected paths provide noise-free, low-curvature supervision signals that stabilise gradients, accelerate convergence, and eliminate the need for dense trajectory storage. We theoretically justify B'ezier-mode connections as effective surrogates for SGD paths and empirically show that the proposed method outperforms state-of-the-art condensation approaches across five clinical datasets, yielding condensed datasets that enable clinically effective model development.

Method: ePF integrates two techniques: Entropic Annealing monitors search diversity via entropy and dynamically anneals resampling distribution when diversity drops, while Look-ahead Modulation adds a predictive guide to evaluate state potential based on successors.

Result: On challenging math benchmarks, ePF significantly outperforms strong baselines and achieves up to 50% relative improvement in task reward.

Conclusion: ePF improves PF’s resilience by balancing exploration of diverse solution spaces with exploitation of high-reward regions, leading to higher-quality solutions.

Abstract: Inference-Time Scaling (ITS) improves language models by allocating more computation at generation time. Particle Filtering (PF) has emerged as a strong ITS method for complex mathematical reasoning tasks, but it is vulnerable when guided by process reward models, which often assign overconfident scores early in the reasoning process. This causes PF to suffer from premature exploitation: it myopically commits to locally promising trajectories, prunes potentially correct hypotheses, and converges to suboptimal solutions. This failure mode, known as particle impoverishment, is especially severe under constrained computational budgets. To address this, we analyze the problem and identify two root causes: a lack of diversity in the particle set due to overconfident resampling and consequent inability to assess the potential of a reasoning path. We introduce Entropic Particle Filtering (ePF), an algorithm that integrates two new techniques to solve these issues. The first technique, Entropic Annealing (EA), directly mitigates particle impoverishment by monitoring search diversity via entropy; when diversity drops, it intervenes by dynamically annealing the resampling distribution to preserve exploration. The second, an enhancement called Look-ahead Modulation (LaM), adds a predictive guide to evaluate a state’s potential based on its successors. On several challenging math benchmarks, ePF significantly outperforms strong baselines and achieves up to a 50 % relative improvement in task reward. Together, these methods improve PF’s resilience by balancing the exploration of diverse solution spaces with the exploitation of high-reward regions, ultimately leading to higher-quality solutions.

Method: Uses modality-specific encoders and cross-modal interaction to capture spatial, temporal, and topological semantics. Implements dynamic trajectory modeling for adaptive information density regulation. Employs contrastive alignment and masked language modeling for self-supervised pretraining.

Result: Extensive experiments on three real-world datasets show MDTI consistently outperforms state-of-the-art baselines, demonstrating robustness and strong generalization abilities.

Conclusion: MDTI effectively addresses trajectory variability and multimodal integration challenges in travel time estimation, achieving superior performance through dynamic modeling and self-supervised learning techniques.

Abstract: Accurate travel time estimation (TTE) plays a crucial role in intelligent transportation systems. However, it remains challenging due to heterogeneous data sources and complex traffic dynamics. Moreover, conventional approaches typically convert trajectories into fixed-length representations, neglecting the inherent variability of real-world trajectories, which often leads to information loss or feature redundancy. To address these challenges, this paper introduces the Multimodal Dynamic Trajectory Integration (MDTI) framework–a novel multimodal trajectory representation learning approach that integrates GPS sequences, grid trajectories, and road network constraints to enhance TTE accuracy. MDTI employs modality-specific encoders and a cross-modal interaction module to capture complementary spatial, temporal, and topological semantics, while a dynamic trajectory modeling mechanism adaptively regulates information density for trajectories of varying lengths. Two self-supervised pretraining objectives, named contrastive alignment and masked language modeling, further strengthen multimodal consistency and contextual understanding. Extensive experiments on three real-world datasets demonstrate that MDTI consistently outperforms state-of-the-art baselines, confirming its robustness and strong generalization abilities. The code is publicly available at: https://github.com/freshhxy/MDTI/

Method: Uses Elliptical Slice Sampling in the source space of flow-based models, requiring only forward passes through the generative model and observation process, no gradient or Jacobian computations.

Result: Effective for designing materials with target properties and predicting protein structures from sparse inter-residue distance measurements.

Conclusion: ESS-Flow provides a practical gradient-free approach for conditional generation tasks where traditional gradient-based methods are infeasible.

Abstract: Guiding pretrained flow-based generative models for conditional generation or to produce samples with desired target properties enables solving diverse tasks without retraining on paired data. We present ESS-Flow, a gradient-free method that leverages the typically Gaussian prior of the source distribution in flow-based models to perform Bayesian inference directly in the source space using Elliptical Slice Sampling. ESS-Flow only requires forward passes through the generative model and observation process, no gradient or Jacobian computations, and is applicable even when gradients are unreliable or unavailable, such as with simulation-based observations or quantization in the generation or observation process. We demonstrate its effectiveness on designing materials with desired target properties and predicting protein structures from sparse inter-residue distance measurements.

Method: Theoretical analysis shows that any trained JEPA can compute sample probabilities efficiently using the model’s Jacobian matrix, validated empirically across datasets and methods like I-JEPA, DINOv2, and MetaCLIP.

Result: JEPAs’ anti-collapse term provably estimates data density, enabling efficient computation of sample probabilities for tasks such as outlier detection and data curation, with empirical validation across diverse datasets and methods.

Conclusion: JEPAs not only learn useful representations but also inherently estimate data density through their anti-collapse term, providing a versatile tool for density estimation and related applications.

Abstract: Joint Embedding Predictive Architectures (JEPAs) learn representations able to solve numerous downstream tasks out-of-the-box. JEPAs combine two objectives: (i) a latent-space prediction term, i.e., the representation of a slightly perturbed sample must be predictable from the original sample’s representation, and (ii) an anti-collapse term, i.e., not all samples should have the same representation. While (ii) is often considered as an obvious remedy to representation collapse, we uncover that JEPAs’ anti-collapse term does much more–it provably estimates the data density. In short, any successfully trained JEPA can be used to get sample probabilities, e.g., for data curation, outlier detection, or simply for density estimation. Our theoretical finding is agnostic of the dataset and architecture used–in any case one can compute the learned probabilities of sample $x$ efficiently and in closed-form using the model’s Jacobian matrix at $x$. Our findings are empirically validated across datasets (synthetic, controlled, and Imagenet) and across different Self Supervised Learning methods falling under the JEPA family (I-JEPA and DINOv2) and on multimodal models, such as MetaCLIP. We denote the method extracting the JEPA learned density as {\bf JEPA-SCORE}.

Method: Analyzed local order invariance, introduced KL-based metric for temporal drift, and compared explicit distribution forecasting with order-preserving methods and order-invariant multi-token approaches.

Result: Simple explicit distribution forecasting objective consistently outperforms complex implicit baselines, and mode collapse is primarily driven by distributional imbalance.

Conclusion: Provides a principled framework for selecting modeling strategies and practical guidance for building more accurate and robust forecasting systems.

Abstract: This paper studies forecasting of the future distribution of events in human action sequences, a task essential in domains like retail, finance, healthcare, and recommendation systems where the precise temporal order is often less critical than the set of outcomes. We challenge the dominant autoregressive paradigm and investigate whether explicitly modeling the future distribution or order-invariant multi-token approaches outperform order-preserving methods. We analyze local order invariance and introduce a KL-based metric to quantify temporal drift. We find that a simple explicit distribution forecasting objective consistently surpasses complex implicit baselines. We further demonstrate that mode collapse of predicted categories is primarily driven by distributional imbalance. This work provides a principled framework for selecting modeling strategies and offers practical guidance for building more accurate and robust forecasting systems.

Method: Propose MaNGO architecture that learns shared structure through meta-learning, generates latent representations using conditional neural processes (CNPs), and combines CNPs with a novel neural operator architecture to mitigate error accumulation over time.

Result: MaNGO demonstrates superior performance over existing GNS methods on dynamics prediction tasks with varying material properties, achieving accuracy on unseen material properties close to that of an oracle model.

Conclusion: The proposed meta-learning approach enables fast adaptation to new physical parameters without retraining, making physics simulations more efficient and scalable across different parameter settings.

Abstract: Accurately simulating physics is crucial across scientific domains, with applications spanning from robotics to materials science. While traditional mesh-based simulations are precise, they are often computationally expensive and require knowledge of physical parameters, such as material properties. In contrast, data-driven approaches like Graph Network Simulators (GNSs) offer faster inference but suffer from two key limitations: Firstly, they must be retrained from scratch for even minor variations in physical parameters, and secondly they require labor-intensive data collection for each new parameter setting. This is inefficient, as simulations with varying parameters often share a common underlying latent structure. In this work, we address these challenges by learning this shared structure through meta-learning, enabling fast adaptation to new physical parameters without retraining. To this end, we propose a novel architecture that generates a latent representation by encoding graph trajectories using conditional neural processes (CNPs). To mitigate error accumulation over time, we combine CNPs with a novel neural operator architecture. We validate our approach, Meta Neural Graph Operator (MaNGO), on several dynamics prediction tasks with varying material properties, demonstrating superior performance over existing GNS methods. Notably, MaNGO achieves accuracy on unseen material properties close to that of an oracle model.

Method: Developed a multi-task, modality-agnostic benchmark with 7 datasets from diverse cities across three continents, and established simple task-oriented model baselines.

Result: Created a standardized benchmark that enables evaluation of GeoAI embedders on various phenomena with underlying geographic processes.

Conclusion: The proposed benchmark provides a crucial reference point for comparing complex GeoAI solutions and addresses the limitations of existing evaluation frameworks.

Abstract: GeoAI is evolving rapidly, fueled by diverse geospatial datasets like traffic patterns, environmental data, and crowdsourced OpenStreetMap (OSM) information. While sophisticated AI models are being developed, existing benchmarks are often concentrated on single tasks and restricted to a single modality. As such, progress in GeoAI is limited by the lack of a standardized, multi-task, modality-agnostic benchmark for their systematic evaluation. This paper introduces a novel benchmark designed to assess the performance, accuracy, and efficiency of geospatial embedders. Our benchmark is modality-agnostic and comprises 7 distinct datasets from diverse cities across three continents, ensuring generalizability and mitigating demographic biases. It allows for the evaluation of GeoAI embedders on various phenomena that exhibit underlying geographic processes. Furthermore, we establish a simple and intuitive task-oriented model baselines, providing a crucial reference point for comparing more complex solutions.

Method: Three solutions: (1) inference-time hybridisation of linear-only conversions with sliding-window softmax, (2) HedgeCATs combining attention-weight transfer with targeted LoRA fine-tuning, and (3) Scheduled Sliding-window Dropout (SSD) that stochastically suppresses softmax during training.

Result: The methods maintain computational efficiency while recovering most base model performance and ensuring genuine linear attention adoption.

Conclusion: The proposed solutions restore the validity of performance attributions in hybrid conversions by preventing component collapse and ensuring balanced usage of linear attention components.

Abstract: Transformers’ quadratic computational complexity limits their scalability despite remarkable performance. While linear attention reduces this to linear complexity, pre-training such models from scratch remains, in most cases, prohibitively expensive. Recent post-training linearisation methods convert pre-trained Transformers to linear models efficiently, often using hybrid approaches that combine linear attention with sliding-window softmax. We identify a critical flaw: existing hybrid methods inadvertently bypass the linear component, relying almost entirely on SWA. Component-level diagnostics reveal this previously undetected behaviour stems from overlooked evaluation practices on common-sense benchmarks. We propose three solutions to ensure balanced component usage: (i) inference-time hybridisation of linear-only conversions with sliding-window softmax; (ii) HedgeCATs, combining attention-weight transfer with targeted LoRA fine-tuning; and (iii) Scheduled Sliding-window Dropout (SSD), which stochastically suppresses the softmax branch during training to prevent component collapse. Our methods maintain computational efficiency while recovering most base model performance and ensuring genuine linear attention adoption, restoring the validity of performance attributions in hybrid conversions.

Method: Compared three autoencoder architectures: CAE, VAE-BiLSTM, and VAE-BiLSTM-MHA trained on normal ECG samples to reconstruct non-anomalous morphology and detect disease deviations.

Result: VAE-BiLSTM-MHA achieved best performance with AUPRC of 0.81 and recall of 0.85 on CPSC dataset, outperforming other architectures.

Conclusion: Attention-augmented VAE shows superior ECG anomaly detection performance and is integrated into interactive clinical dashboard for triage support.

Abstract: Anomaly detection in 12-lead electrocardiograms (ECGs) is critical for identifying deviations associated with cardiovascular disease. This work presents a comparative analysis of three autoencoder-based architectures: convolutional autoencoder (CAE), variational autoencoder with bidirectional long short-term memory (VAE-BiLSTM), and VAE-BiLSTM with multi-head attention (VAE-BiLSTM-MHA), for unsupervised anomaly detection in ECGs. To the best of our knowledge, this study reports the first application of a VAE-BiLSTM-MHA architecture to ECG anomaly detection. All models are trained on normal ECG samples to reconstruct non-anomalous cardiac morphology and detect deviations indicative of disease. Using a unified preprocessing and evaluation pipeline on the public China Physiological Signal Challenge (CPSC) dataset, the attention-augmented VAE achieves the best performance, with an AUPRC of 0.81 and a recall of 0.85 on the held-out test set, outperforming the other architectures. To support clinical triage, this model is further integrated into an interactive dashboard that visualizes anomaly localization. In addition, a performance comparison with baseline models from the literature is provided.

Method: Introduces Carré du champ flow matching (CDC-FM) that regularizes probability paths with spatially varying, anisotropic Gaussian noise whose covariance captures local latent data manifold geometry.

Result: CDC-FM consistently offers better quality-generalization tradeoff, with significant improvements over standard FM in data-scarce regimes and non-uniformly sampled datasets across diverse domains.

Conclusion: Provides both a mathematical framework for studying data geometry, generalization and memorization interplay, and a scalable algorithm that can be integrated into existing flow matching pipelines.

Abstract: Deep generative models often face a fundamental tradeoff: high sample quality can come at the cost of memorisation, where the model reproduces training data rather than generalising across the underlying data geometry. We introduce Carr'e du champ flow matching (CDC-FM), a generalisation of flow matching (FM), that improves the quality-generalisation tradeoff by regularising the probability path with a geometry-aware noise. Our method replaces the homogeneous, isotropic noise in FM with a spatially varying, anisotropic Gaussian noise whose covariance captures the local geometry of the latent data manifold. We prove that this geometric noise can be optimally estimated from the data and is scalable to large data. Further, we provide an extensive experimental evaluation on diverse datasets (synthetic manifolds, point clouds, single-cell genomics, animal motion capture, and images) as well as various neural network architectures (MLPs, CNNs, and transformers). We demonstrate that CDC-FM consistently offers a better quality-generalisation tradeoff. We observe significant improvements over standard FM in data-scarce regimes and in highly non-uniformly sampled datasets, which are often encountered in AI for science applications. Our work provides a mathematical framework for studying the interplay between data geometry, generalisation and memorisation in generative models, as well as a robust and scalable algorithm that can be readily integrated into existing flow matching pipelines.

Method: Multi-agent architecture orchestrating deliberative debates among multiple LLM agents with specific roles for collective feature evaluation and justification generation.

Result: Achieved superior/comparable classification performance on CIC-DIAD 2024 IoT intrusion detection dataset, reduced downstream training time by 46% (statistically significant p=0.028 for XGBoost).

Conclusion: The deliberative architecture enhances both decision transparency and computational efficiency, making LLM-FS-Agent a practical solution for real-world applications.

Abstract: High-dimensional data remains a pervasive challenge in machine learning, often undermining model interpretability and computational efficiency. While Large Language Models (LLMs) have shown promise for dimensionality reduction through feature selection, existing LLM-based approaches frequently lack structured reasoning and transparent justification for their decisions. This paper introduces LLM-FS-Agent, a novel multi-agent architecture designed for interpretable and robust feature selection. The system orchestrates a deliberative “debate” among multiple LLM agents, each assigned a specific role, enabling collective evaluation of feature relevance and generation of detailed justifications. We evaluate LLM-FS-Agent in the cybersecurity domain using the CIC-DIAD 2024 IoT intrusion detection dataset and compare its performance against strong baselines, including LLM-Select and traditional methods such as PCA. Experimental results demonstrate that LLM-FS-Agent consistently achieves superior or comparable classification performance while reducing downstream training time by an average of 46% (statistically significant improvement, p = 0.028 for XGBoost). These findings highlight that the proposed deliberative architecture enhances both decision transparency and computational efficiency, establishing LLM-FS-Agent as a practical and reliable solution for real-world applications.

Method: Proposes three strategies: Greedy-Threshold (makes sampling greedy at very low confidence steps), Calibrated-TopK and Calibrated-epsilon (set truncation threshold based on estimated rank-wise correctness).

Result: Shows gains across math and general reasoning benchmarks by challenging prevailing heuristics about decoding under uncertainty.

Conclusion: Decoding should be calibrated by correctness rather than confidence alone, enabling better balance between exploration and reliability in LLM reasoning.

Abstract: Large Language Models (LLMs) are increasingly applied to complex tasks that require extended reasoning. In such settings, models often benefit from diverse chains-of-thought to arrive at multiple candidate solutions. This requires two competing objectives: to inject enough stochasticity to explore multiple reasoning chains, and to ensure sufficient accuracy and quality in each path. Existing works pursue the first objective by increasing exploration at highly uncertain steps with higher temperature or larger candidate token sets, while others improve reliability by rejecting samples with low confidence post-generation, implying that low confidence correlates with low answer quality. These two lines of thought are in conflict, as they conflate different sources of uncertainty. To resolve this, we argue that the decoding rule should be calibrated by correctness, not confidence alone. We should sample from tokens with higher estimated correctness, and reduce sampling where expected correctness is low. We propose simple strategies that achieve this goal: Greedy-Threshold makes sampling greedy at very low confidence steps. Calibrated-TopK and Calibrated-epsilon set truncation threshold based on estimated rank-wise correctness. Together, our findings challenge prevailing heuristics about decoding under uncertainty and show gains across math and general reasoning benchmarks.

Method: Explains methods for quantifying uncertainty in predictive models including linear regression, random forests, and neural networks, and introduces conformal prediction framework for generating predictions with predefined confidence intervals.

Result: Presents a systematic approach to identify and distinguish between different types of uncertainty, and demonstrates how uncertainty estimation can be practically implemented across various machine learning models.

Conclusion: Uncertainty quantification is essential for enhancing model reliability, improving business decision-making, and supporting risk-aware strategies in machine learning applications.

Abstract: This book chapter introduces the principles and practical applications of uncertainty quantification in machine learning. It explains how to identify and distinguish between different types of uncertainty and presents methods for quantifying uncertainty in predictive models, including linear regression, random forests, and neural networks. The chapter also covers conformal prediction as a framework for generating predictions with predefined confidence intervals. Finally, it explores how uncertainty estimation can be leveraged to improve business decision-making, enhance model reliability, and support risk-aware strategies.

Method: Dual-decoder architecture disentangles resonant and non-resonant signals by enforcing Kramers-Kronig causality relations via differentiable Hilbert transform loss on resonant part and smoothness prior on non-resonant part.

Result: Trained entirely on synthetic data, RamPINN demonstrates strong zero-shot generalization to real experimental data, significantly outperforming existing baselines and achieving competitive results even without ground-truth Raman spectra.

Conclusion: Scientific rules can serve as powerful inductive biases enabling robust self-supervised learning in data-limited scientific domains, bridging the gap between deep learning and scientific applications.

Abstract: Transferring the recent advancements in deep learning into scientific disciplines is hindered by the lack of the required large-scale datasets for training. We argue that in these knowledge-rich domains, the established body of scientific theory provides reliable inductive biases in the form of governing physical laws. We address the ill-posed inverse problem of recovering Raman spectra from noisy Coherent Anti-Stokes Raman Scattering (CARS) measurements, as the true Raman signal here is suppressed by a dominating non-resonant background. We propose RamPINN, a model that learns to recover Raman spectra from given CARS spectra. Our core methodological contribution is a physics-informed neural network that utilizes a dual-decoder architecture to disentangle resonant and non-resonant signals. This is done by enforcing the Kramers-Kronig causality relations via a differentiable Hilbert transform loss on the resonant and a smoothness prior on the non-resonant part of the signal. Trained entirely on synthetic data, RamPINN demonstrates strong zero-shot generalization to real-world experimental data, explicitly closing this gap and significantly outperforming existing baselines. Furthermore, we show that training with these physics-based losses alone, without access to any ground-truth Raman spectra, still yields competitive results. This work highlights a broader concept: formal scientific rules can act as a potent inductive bias, enabling robust, self-supervised learning in data-limited scientific domains.

Method: Introduced a new family of Bayesian posthoc OOD scores based on expected logit vectors, comparing 5 Bayesian and 4 deterministic posthoc OOD scores.

Result: Experiments on MNIST and CIFAR-10 with 5000 or fewer training samples showed Bayesian methods outperform corresponding deterministic methods.

Conclusion: Bayesian approaches are more effective than deterministic methods for OOD detection in small-data scenarios.

Abstract: Out-of-Distribution (OOD) detection is critical to AI reliability and safety, yet in many practical settings, only a limited amount of training data is available. Bayesian Neural Networks (BNNs) are a promising class of model on which to base OOD detection, because they explicitly represent epistemic (i.e. model) uncertainty. In the small training data regime, BNNs are especially valuable because they can incorporate prior model information. We introduce a new family of Bayesian posthoc OOD scores based on expected logit vectors, and compare 5 Bayesian and 4 deterministic posthoc OOD scores. Experiments on MNIST and CIFAR-10 In-Distributions, with 5000 training samples or less, show that the Bayesian methods outperform corresponding deterministic methods.

Method: Develop data-dependent bounds for Gibbs algorithm test error, analyze stability under Langevin Monte Carlo approximation, validate on MNIST and CIFAR-10 datasets with true and random labels.

Result: Bounds yield nontrivial predictions on true labeled data and correctly upper bound test error for random labels. Generalization in low-temperature interpolation regime is signaled by small training errors in high temperature regime.

Conclusion: The method provides effective bounds for Gibbs algorithm generalization in overparameterized settings, with generalization behavior predictable from training error patterns across temperature regimes.

Abstract: The paper provides data-dependent bounds on the test error of the Gibbs algorithm in the overparameterized interpolation regime, where low training errors are also obtained for impossible data, such as random labels in classification. The bounds are stable under approximation with Langevin Monte Carlo algorithms. Experiments on the MNIST and CIFAR-10 datasets verify that the bounds yield nontrivial predictions on true labeled data and correctly upper bound the test error for random labels. Our method indicates that generalization in the low-temperature, interpolation regime is already signaled by small training errors in the more classical high temperature regime.

Method: Developed molFTP vectorization with dummy-masking procedure to remove fragment information from held-out molecules, and introduced key-loo approximation for molecule-level leave-one-out validation.

Result: Key-loo closely approximates true molecule-level LOO with deviation below 8%, enabling near full data training while preserving unbiased cross-validation performance estimates.

Conclusion: molFTP offers fast, leakage-resistant fragment-target prevalence vectorization with practical safeguards that approximate LOO validation at significantly reduced computational cost.

Abstract: We introduce molFTP (molecular fragment-target prevalence), a compact representation that delivers strong predictive performance. To prevent feature leakage across cross-validation folds, we implement a dummy-masking procedure that removes information about fragments present in the held-out molecules. We further show that key leave-one-out (key-loo) closely approximates true molecule-level leave-one-out (LOO), with deviation below 8% on our datasets. This enables near full data training while preserving unbiased cross-validation estimates of model performance. Overall, molFTP provides a fast, leakage-resistant fragment-target prevalence vectorization with practical safeguards (dummy masking or key-LOO) that approximate LOO at a fraction of its cost.

Method: Combines Proxy Value Propagation (PVP) and Distributional Soft Actor-Critic (DSAC) to create a distributed proxy value function that encodes human intent, assigns higher returns to expert demonstrations, and penalizes actions requiring human intervention.

Result: Achieves real-world driving policy learning within practical training times. Both simulation and real-world experiments demonstrate safe, robust, and sample-efficient learning for autonomous driving.

Conclusion: The proposed H-DSAC framework effectively enables safe and efficient autonomous driving policy learning by incorporating human guidance through a reward-free, active human-in-the-loop approach.

Abstract: Autonomous driving with reinforcement learning (RL) has significant potential. However, applying RL in real-world settings remains challenging due to the need for safe, efficient, and robust learning. Incorporating human expertise into the learning process can help overcome these challenges by reducing risky exploration and improving sample efficiency. In this work, we propose a reward-free, active human-in-the-loop learning method called Human-Guided Distributional Soft Actor-Critic (H-DSAC). Our method combines Proxy Value Propagation (PVP) and Distributional Soft Actor-Critic (DSAC) to enable efficient and safe training in real-world environments. The key innovation is the construction of a distributed proxy value function within the DSAC framework. This function encodes human intent by assigning higher expected returns to expert demonstrations and penalizing actions that require human intervention. By extrapolating these labels to unlabeled states, the policy is effectively guided toward expert-like behavior. With a well-designed state space, our method achieves real-world driving policy learning within practical training times. Results from both simulation and real-world experiments demonstrate that our framework enables safe, robust, and sample-efficient learning for autonomous driving.

Method: BLISS uses a small proxy model as surrogate, formulates data selection as bilevel optimization with score model assigning importance weights, ensuring proxy model training on weighted loss leads to best validation performance.

Result: BLISS achieves 1.7× speedup in reaching same performance as state-of-the-art method under 1B model setting, demonstrating superior performance across multiple downstream tasks.

Conclusion: BLISS provides effective data selection without external models, explicitly accounting for long-term impact, and enables efficient selection of high-quality samples for LLM pretraining.

Abstract: Effective data selection is essential for pretraining large language models (LLMs), enhancing efficiency and improving generalization to downstream tasks. However, existing approaches often require leveraging external pretrained models, making it difficult to disentangle the effects of data selection from those of the external pretrained models. In addition, they often overlook the long-term impact of selected data if the model is trained to convergence, primarily due to the prohibitive cost of full-scale LLM pretraining. In this paper, we introduce BLISS (\textbf{B}ileve\textbf{L} \textbf{I}nfluence \textbf{S}coring method for data \textbf{S}election): a lightweight data selection method that operates entirely \emph{from scratch}, without relying on any external pretrained oracle models, while explicitly accounting for the long-term impact of selected data. BLISS leverages a small proxy model as a surrogate for the LLM and employs a score model to estimate the long-term influence of training samples if the proxy model is trained to convergence. We formulate data selection as a bilevel optimization problem, where the upper-level objective optimizes the score model to assign importance weights to training samples, ensuring that minimizing the lower-level objective (i.e., training the proxy model over the weighted training loss until convergence) leads to best validation performance. Once optimized, the trained score model predicts influence scores for the dataset, enabling efficient selection of high-quality samples for LLM pretraining. We validate BLISS by pretraining 410M/1B/2.8B Pythia and LLaMA-0.5B models on selected subsets of the C4 dataset. Notably, under the 1B model setting, BLISS achieves $1.7\times$ speedup in reaching the same performance as the state-of-the-art method, demonstrating superior performance across multiple downstream tasks.

Method: Created a synthetic dataset of over 18,000 scatterplots from six data generators and 17 chart designs, then evaluated proprietary models from OpenAI and Google using N-shot prompting on five distinct tasks derived from annotations of cluster bounding boxes, center coordinates, and outlier coordinates.

Result: OpenAI models and Gemini 2.5 Flash (especially with example prompting) achieve 90%+ accuracy for counting clusters and outliers. However, localization tasks show poor performance with Precision and Recall near or below 50%, except for Flash in outlier identification (65.01%). Chart design has secondary impact, but wide aspect ratios (16:9, 21:9) and random coloring should be avoided.

Conclusion: Current AI models show promise for basic scatterplot analysis tasks like counting, but significant improvements are needed for localization-related tasks. Careful chart design choices can help optimize model performance.

Abstract: AI models are increasingly used for data analysis and visualization, yet benchmarks rarely address scatterplot-specific tasks, limiting insight into performance. To address this gap for one of the most common chart types, we introduce a synthetic, annotated dataset of over 18,000 scatterplots from six data generators and 17 chart designs, and a benchmark based on it. We evaluate proprietary models from OpenAI and Google using N-shot prompting on five distinct tasks derived from annotations of cluster bounding boxes, their center coordinates, and outlier coordinates. OpenAI models and Gemini 2.5 Flash, especially when prompted with examples, are viable options for counting clusters and, in Flash’s case, outliers (90%+ Accuracy). However, the results for localization-related tasks are unsatisfactory: Precision and Recall are near or below 50%, except for Flash in outlier identification (65.01%). Furthermore, the impact of chart design on performance appears to be a secondary factor, but it is advisable to avoid scatterplots with wide aspect ratios (16:9 and 21:9) or those colored randomly. Supplementary materials are available at https://github.com/feedzai/biy-paper.

Method: Novel failure-aware IRL algorithm that focuses on misclassified or difficult examples (failures) to recover latent rewards. It learns from these failures without requiring external classifiers or supervision.

Result: Outperforms existing IRL baselines across multiple metrics in LLM detoxification. Extracts reward functions that better reflect true RLHF objectives and enables more effective re-RLHF training than standard IRL.

Conclusion: Failure-aware IRL is a robust, scalable method for auditing model alignment and reducing ambiguity in the IRL process, better capturing true incentives learned during RLHF.

Abstract: Reinforcement Learning from Human Feedback (RLHF) aligns Large Language Models (LLMs) with human preferences, yet the underlying reward signals they internalize remain hidden, posing a critical challenge for interpretability and safety. Existing approaches attempt to extract these latent incentives using Inverse Reinforcement Learning (IRL), but treat all preference pairs equally, often overlooking the most informative signals: those examples the extracted reward model misclassifies or assigns nearly equal scores, which we term \emph{failures}. We introduce a novel \emph{failure-aware} IRL algorithm that focuses on misclassified or difficult examples to recover the latent rewards defining model behaviors. By learning from these failures, our failure-aware IRL extracts reward functions that better reflect the true objectives behind RLHF. We demonstrate that failure-aware IRL outperforms existing IRL baselines across multiple metrics when applied to LLM detoxification, without requiring external classifiers or supervision. Crucially, failure-aware IRL yields rewards that better capture the true incentives learned during RLHF, enabling more effective re-RLHF training than standard IRL. This establishes failure-aware IRL as a robust, scalable method for auditing model alignment and reducing ambiguity in the IRL process.

Method: Proposes Edit Flow process that learns instantaneous edit rates (insert, delete, substitute) within a continuous-time Markov chain framework to transport noise to data more efficiently.

Result: The model demonstrates generative flexibility in various unconditional and conditional generation tasks on benchmark TPPs, effectively reducing the number of necessary edit operations.

Conclusion: Edit Flow provides a flexible and efficient non-autoregressive approach for TPP modeling that outperforms existing methods by leveraging continuous-time Markov chains with multiple edit operations.

Abstract: Temporal point processes (TPPs) are a fundamental tool for modeling event sequences in continuous time, but most existing approaches rely on autoregressive parameterizations that are limited by their sequential sampling. Recent non-autoregressive, diffusion-style models mitigate these issues by jointly interpolating between noise and data through event insertions and deletions in a discrete Markov chain. In this work, we generalize this perspective and introduce an Edit Flow process for TPPs that transports noise to data via insert, delete, and substitute edit operations. By learning the instantaneous edit rates within a continuous-time Markov chain framework, we attain a flexible and efficient model that effectively reduces the total number of necessary edit operations during generation. Empirical results demonstrate the generative flexibility of our unconditionally trained model in a wide range of unconditional and conditional generation tasks on benchmark TPPs.

Method: Uses text encoder for task metadata, learned gating module to select/blend multiple sub-policies, enabling end-to-end training across tasks. Also studied with fixed expert policies.

Result: Matches or exceeds strong multi-task baselines in success rate and sample efficiency on MetaWorld benchmarks. Learned language gate composes experts for novel tasks and unseen combinations.

Conclusion: Natural-language metadata can effectively index and recombine reusable skills within a single policy for multi-task RL.

Abstract: Multi-task reinforcement learning often relies on task metadata – such as brief natural-language descriptions – to guide behavior across diverse objectives. We present Lexical Policy Networks (LEXPOL), a language-conditioned mixture-of-policies architecture for multi-task RL. LEXPOL encodes task metadata with a text encoder and uses a learned gating module to select or blend among multiple sub-policies, enabling end-to-end training across tasks. On MetaWorld benchmarks, LEXPOL matches or exceeds strong multi-task baselines in success rate and sample efficiency, without task-specific retraining. To analyze the mechanism, we further study settings with fixed expert policies obtained independently of the gate and show that the learned language gate composes these experts to produce behaviors appropriate to novel task descriptions and unseen task combinations. These results indicate that natural-language metadata can effectively index and recombine reusable skills within a single policy.

Method: Leverages Bayesian Inverse Reinforcement Learning to recover reward distributions, quantify non-identifiability through posterior contraction analysis, provide uncertainty-aware diagnostics, and validate policy-level utility via RLHF training.

Result: Successfully audited a detoxified LLM, yielding well-calibrated and interpretable objectives that strengthen alignment guarantees. The framework demonstrates comparable training dynamics and toxicity reductions to ground-truth alignment processes.

Conclusion: Provides a practical toolkit for auditors, safety teams, and regulators to verify what LLMs are truly trying to achieve, moving toward more trustworthy and accountable AI systems.

Abstract: The objectives that Large Language Models (LLMs) implicitly optimize remain dangerously opaque, making trustworthy alignment and auditing a grand challenge. While Inverse Reinforcement Learning (IRL) can infer reward functions from behaviour, existing approaches either produce a single, overconfident reward estimate or fail to address the fundamental ambiguity of the task (non-identifiability). This paper introduces a principled auditing framework that re-frames reward inference from a simple estimation task to a comprehensive process for verification. Our framework leverages Bayesian IRL to not only recover a distribution over objectives but to enable three critical audit capabilities: (i) Quantifying and systematically reducing non-identifiability by demonstrating posterior contraction over sequential rounds of evidence; (ii) Providing actionable, uncertainty-aware diagnostics that expose spurious shortcuts and identify out-of-distribution prompts where the inferred objective cannot be trusted; and (iii) Validating policy-level utility by showing that the refined, low-uncertainty reward can be used directly in RLHF to achieve training dynamics and toxicity reductions comparable to the ground-truth alignment process. Empirically, our framework successfully audits a detoxified LLM, yielding a well-calibrated and interpretable objective that strengthens alignment guarantees. Overall, this work provides a practical toolkit for auditors, safety teams, and regulators to verify what LLMs are truly trying to achieve, moving us toward more trustworthy and accountable AI.

Method: Developed MASED metric to measure oversmoothing, derived theoretical bounds, analyzed weight matrix properties, and proposed G-Reg regularization scheme to decouple adjacency hops from weight matrices.

Result: G-Reg regularization increases MASED bounds, improves node classification accuracy at large depths, enables better performance than shallow networks in cold start scenarios, and shows trade-off between receptive field size and performance.

Conclusion: Oversmoothing increases with more trainable weight and adjacency matrices, but can be mitigated through proper regularization and decoupling of hops from weight matrices, enabling effective deep GNNs.

Abstract: In this paper, we study the factors that contribute to the effect of oversmoothing in deep Graph Neural Networks (GNNs). Specifically, our analysis is based on a new metric (Mean Average Squared Distance - $MASED$) to quantify the extent of oversmoothing. We derive layer-wise bounds on $MASED$, which aggregate to yield global upper and lower distance bounds. Based on this quantification of oversmoothing, we further analyze the importance of two different properties of the model; namely the norms of the generated node embeddings, along with the largest and smallest singular values of the weight matrices. Building on the insights drawn from the theoretical analysis, we show that oversmoothing increases as the number of trainable weight matrices and the number of adjacency matrices increases. We also use the derived layer-wise bounds on $MASED$ to form a proposal for decoupling the number of hops (i.e., adjacency depth) from the number of weight matrices. In particular, we introduce G-Reg, a regularization scheme that increases the bounds, and demonstrate through extensive experiments that by doing so node classification accuracy increases, achieving robustness at large depths. We further show that by reducing oversmoothing in deep networks, we can achieve better results in some tasks than using shallow ones. Specifically, we experiment with a ``cold start” scenario, i.e., when there is no feature information for the unlabeled nodes. Finally, we show empirically the trade-off between receptive field size (i.e., number of weight matrices) and performance, using the $MASED$ bounds. This is achieved by distributing adjacency hops across a small number of trainable layers, avoiding the extremes of under- or over-parameterization of the GNN.

Method: Using LLMs as policy-agnostic human proxies to generate synthetic data in grid-world capture games inspired by Stag Hunt. Three experiments test LLM decision-making: comparing with humans/experts, inducing risk-sensitive strategies, and testing in dynamic grid-worlds.

Result: LLMs align more closely with experts than human participants, can mirror human risk-sensitive behaviors through prompting, and generate trajectories resembling human paths in dynamic environments.

Conclusion: While LLMs cannot fully replicate human adaptability, their prompt-guided diversity provides a scalable foundation for simulating policy-agnostic teammates in heterogeneous-agent training.

Abstract: A critical challenge in modelling Heterogeneous-Agent Teams is training agents to collaborate with teammates whose policies are inaccessible or non-stationary, such as humans. Traditional approaches rely on expensive human-in-the-loop data, which limits scalability. We propose using Large Language Models (LLMs) as policy-agnostic human proxies to generate synthetic data that mimics human decision-making. To evaluate this, we conduct three experiments in a grid-world capture game inspired by Stag Hunt, a game theory paradigm that balances risk and reward. In Experiment 1, we compare decisions from 30 human participants and 2 expert judges with outputs from LLaMA 3.1 and Mixtral 8x22B models. LLMs, prompted with game-state observations and reward structures, align more closely with experts than participants, demonstrating consistency in applying underlying decision criteria. Experiment 2 modifies prompts to induce risk-sensitive strategies (e.g. “be risk averse”). LLM outputs mirror human participants’ variability, shifting between risk-averse and risk-seeking behaviours. Finally, Experiment 3 tests LLMs in a dynamic grid-world where the LLM agents generate movement actions. LLMs produce trajectories resembling human participants’ paths. While LLMs cannot yet fully replicate human adaptability, their prompt-guided diversity offers a scalable foundation for simulating policy-agnostic teammates.

Method: Use influence functions to measure the causal effect of individual CoT examples on downstream accuracy, and introduce influence-based pruning for data selection.

Result: Influence-based pruning consistently outperforms perplexity and embedding-based baselines on math reasoning tasks within a model family.

Conclusion: Influence functions provide an effective way to define and select high-quality reasoning data for fine-tuning LLMs.

Abstract: Fine-tuning large language models (LLMs) on chain-of-thought (CoT) data shows that a small amount of high-quality data can outperform massive datasets. Yet, what constitutes “quality” remains ill-defined. Existing reasoning methods rely on indirect heuristics such as problem difficulty or trace length, while instruction-tuning has explored a broader range of automated selection strategies, but rarely in the context of reasoning. We propose to define reasoning data quality using influence functions, which measure the causal effect of individual CoT examples on downstream accuracy, and introduce influence-based pruning, which consistently outperforms perplexity and embedding-based baselines on math reasoning within a model family.

Method: Construct moment tensors using input-output data to recover globally consistent estimates of mixture weights and system parameters, then refine through Kalman EM algorithm with closed-form updates for all LDS parameters.

Result: Tensor-EM achieves more reliable recovery and improved robustness on synthetic data compared to pure tensor or randomly initialized EM methods. Successfully models neural recordings from primate somatosensory cortex, clustering different conditions as separate subsystems consistent with supervised single-LDS fits.

Conclusion: MoLDS provides an effective framework for modeling complex neural data, and Tensor-EM is a reliable approach to MoLDS learning that combines the strengths of tensor methods and EM for neural applications.

Abstract: Mixtures of linear dynamical systems (MoLDS) provide a path to model time-series data that exhibit diverse temporal dynamics across trajectories. However, its application remains challenging in complex and noisy settings, limiting its effectiveness for neural data analysis. Tensor-based moment methods can provide global identifiability guarantees for MoLDS, but their performance degrades under noise and complexity. Commonly used expectation-maximization (EM) methods offer flexibility in fitting latent models but are highly sensitive to initialization and prone to poor local minima. Here, we propose a tensor-based method that provides identifiability guarantees for learning MoLDS, which is followed by EM updates to combine the strengths of both approaches. The novelty in our approach lies in the construction of moment tensors using the input-output data to recover globally consistent estimates of mixture weights and system parameters. These estimates can then be refined through a Kalman EM algorithm, with closed-form updates for all LDS parameters. We validate our framework on synthetic benchmarks and real-world datasets. On synthetic data, the proposed Tensor-EM method achieves more reliable recovery and improved robustness compared to either pure tensor or randomly initialized EM methods. We then analyze neural recordings from the primate somatosensory cortex while a non-human primate performs reaches in different directions. Our method successfully models and clusters different conditions as separate subsystems, consistent with supervised single-LDS fits for each condition. Finally, we apply this approach to another neural dataset where monkeys perform a sequential reaching task. These results demonstrate that MoLDS provides an effective framework for modeling complex neural data, and that Tensor-EM is a reliable approach to MoLDS learning for these applications.

Method: The study examines the roles of locality and compositionality in data, tasks, and deep learning representations to understand how neural networks encode and exploit latent structure.

Result: The research addresses how the underlying structure of learnable data enables neural networks to overcome dimensionality challenges and how this structure quantitatively impacts performance metrics like generalization.

Conclusion: Understanding the nature of latent structure in learnable data, particularly through locality and compositionality, is crucial for explaining how neural networks successfully learn high-dimensional tasks despite statistical challenges.

Abstract: Deep neural networks have achieved remarkable success, yet our understanding of how they learn remains limited. These models can learn high-dimensional tasks, which is generally statistically intractable due to the curse of dimensionality. This apparent paradox suggests that learnable data must have an underlying latent structure. What is the nature of this structure? How do neural networks encode and exploit it, and how does it quantitatively impact performance - for instance, how does generalization improve with the number of training examples? This thesis addresses these questions by studying the roles of locality and compositionality in data, tasks, and deep learning representations.

Method: Proposed Stratified GRPO with Stratified Advantage Normalization (SAN) that partitions trajectories into homogeneous strata based on structural properties and computes advantages locally within each stratum, then linearly blends with global estimator for practical stability.

Result: Extensive experiments show Stratified GRPO outperforms GRPO by up to 11.3 points, achieving higher training rewards, greater stability, and more effective search policies on diverse single-hop and multi-hop QA benchmarks.

Conclusion: Stratification provides a principled remedy for structural heterogeneity in RL for LLM search agents, eliminating cross-stratum bias and enabling more effective training of complex search strategies.

Abstract: Large language model (LLM) agents increasingly rely on external tools such as search engines to solve complex, multi-step problems, and reinforcement learning (RL) has become a key paradigm for training them. However, the trajectories of search agents are structurally heterogeneous, where variations in the number, placement, and outcomes of search calls lead to fundamentally different answer directions and reward distributions. Standard policy gradient methods, which use a single global baseline, suffer from what we identify and formalize as cross-stratum bias-an “apples-to-oranges” comparison of heterogeneous trajectories. This cross-stratum bias distorts credit assignment and hinders exploration of complex, multi-step search strategies. To address this, we propose Stratified GRPO, whose central component, Stratified Advantage Normalization (SAN), partitions trajectories into homogeneous strata based on their structural properties and computes advantages locally within each stratum. This ensures that trajectories are evaluated only against their true peers. Our analysis proves that SAN eliminates cross-stratum bias, yields conditionally unbiased unit-variance estimates inside each stratum, and retains the global unbiasedness and unit-variance properties enjoyed by standard normalization, resulting in a more pure and scale-stable learning signal. To improve practical stability under finite-sample regimes, we further linearly blend SAN with the global estimator. Extensive experiments on diverse single-hop and multi-hop question-answering benchmarks demonstrate that Stratified GRPO consistently and substantially outperforms GRPO by up to 11.3 points, achieving higher training rewards, greater training stability, and more effective search policies. These results establish stratification as a principled remedy for structural heterogeneity in RL for LLM search agents.

Method: PGD approximates Jensen-Shannon distance by fitting binary classifiers to distinguish between real and generated graphs, using graph descriptors as features. The data log-likelihood of these classifiers provides a variational lower bound on the JS distance.

Result: PGD metrics are constrained to [0,1] interval and are comparable across different graph descriptors. The framework provides a theoretically grounded summary metric that combines individual metrics for a tight lower bound on distribution distance.

Conclusion: PGD offers more robust and insightful evaluation compared to MMD metrics, addressing limitations of existing methods and providing comparable absolute performance measures for graph generative models.

Abstract: Existing methods for evaluating graph generative models primarily rely on Maximum Mean Discrepancy (MMD) metrics based on graph descriptors. While these metrics can rank generative models, they do not provide an absolute measure of performance. Their values are also highly sensitive to extrinsic parameters, namely kernel and descriptor parametrization, making them incomparable across different graph descriptors. We introduce PolyGraph Discrepancy (PGD), a new evaluation framework that addresses these limitations. It approximates the Jensen-Shannon distance of graph distributions by fitting binary classifiers to distinguish between real and generated graphs, featurized by these descriptors. The data log-likelihood of these classifiers approximates a variational lower bound on the JS distance between the two distributions. Resulting metrics are constrained to the unit interval [0,1] and are comparable across different graph descriptors. We further derive a theoretically grounded summary metric that combines these individual metrics to provide a maximally tight lower bound on the distance for the given descriptors. Thorough experiments demonstrate that PGD provides a more robust and insightful evaluation compared to MMD metrics. The PolyGraph framework for benchmarking graph generative models is made publicly available at https://github.com/BorgwardtLab/polygraph-benchmark.

Method: RGSD performs contrastive pretraining to embed motions on a unit hypersphere, clustering reference trajectories into distinct directions. This grounding enables simultaneous imitation of reference behaviors and discovery of semantically related diverse behaviors.

Result: On a simulated SMPL humanoid with 359-D observations and 69-D actions, RGSD learns structured skills (walking, running, punching, side stepping) and discovers related novel behaviors. It outperforms imitation-based skill acquisition baselines in downstream control tasks.

Conclusion: Lightweight reference-guided grounding offers a practical path to discovering semantically rich and structured skills in high-DoF systems.

Abstract: Scaling unsupervised skill discovery algorithms to high-DoF agents remains challenging. As dimensionality increases, the exploration space grows exponentially, while the manifold of meaningful skills remains limited. Therefore, semantic meaningfulness becomes essential to effectively guide exploration in high-dimensional spaces. In this work, we present Reference-Grounded Skill Discovery (RGSD), a novel algorithm that grounds skill discovery in a semantically meaningful latent space using reference data. RGSD first performs contrastive pretraining to embed motions on a unit hypersphere, clustering each reference trajectory into a distinct direction. This grounding enables skill discovery to simultaneously involve both imitation of reference behaviors and the discovery of semantically related diverse behaviors. On a simulated SMPL humanoid with 359-D observations and 69-D actions, RGSD learns structured skills including walking, running, punching, and side stepping, and also discovers related novel behaviors. In downstream control tasks, RGSD outperforms imitation-based skill acquisition baselines. Our results suggest that lightweight reference-guided grounding offers a practical path to discovering semantically rich and structured skills in high-DoF systems.

Method: Proposed faithfulness metrics including model agreement for predictive consistency and chi-squared tests to detect statistically significant changes in predictive patterns across overall dataset and demographic subgroups.

Result: Applied quantization and pruning to ANNs on three socially meaningful datasets, finding that high accuracy doesn’t ensure faithfulness, and statistical tests revealed subtle but significant behavioral shifts missed by standard metrics like Accuracy and Equalized Odds.

Conclusion: The proposed metrics provide a practical method to ensure efficiency gains from compression don’t compromise fairness or faithfulness, essential for trustworthy AI in critical applications.

Abstract: In real-world applications, computational constraints often require transforming large models into smaller, more efficient versions through model compression. While these techniques aim to reduce size and computational cost without sacrificing performance, their evaluations have traditionally focused on the trade-off between size and accuracy, overlooking the aspect of model faithfulness. This limited view is insufficient for high-stakes domains like healthcare, finance, and criminal justice, where compressed models must remain faithful to the behavior of their original counterparts. This paper presents a novel approach to evaluating faithfulness in compressed models, moving beyond standard metrics. We introduce and demonstrate a set of faithfulness metrics that capture how model behavior changes post-compression. Our contributions include introducing techniques to assess predictive consistency between the original and compressed models using model agreement, and applying chi-squared tests to detect statistically significant changes in predictive patterns across both the overall dataset and demographic subgroups, thereby exposing shifts that aggregate fairness metrics may obscure. We demonstrate our approaches by applying quantization and pruning to artificial neural networks (ANNs) trained on three diverse and socially meaningful datasets. Our findings show that high accuracy does not guarantee faithfulness, and our statistical tests detect subtle yet significant shifts that are missed by standard metrics, such as Accuracy and Equalized Odds. The proposed metrics provide a practical and more direct method for ensuring that efficiency gains through compression do not compromise the fairness or faithfulness essential for trustworthy AI.

Method: Developed lm-Meter as a lightweight profiler that captures real-time latency at phase and kernel levels without auxiliary devices, implemented on commercial mobile platforms with minimal system overhead.

Result: lm-Meter achieves high profiling accuracy with only 2.58% throughput reduction in prefill and 0.99% in decode under constrained conditions. It reveals phase/kernel-level bottlenecks and identifies optimization opportunities.

Conclusion: lm-Meter provides unprecedented visibility into LLM runtime behavior on constrained platforms, enabling informed optimization and accelerating the democratization of on-device LLM systems.

Abstract: Large Language Models (LLMs) are increasingly integrated into everyday applications, but their prevalent cloud-based deployment raises growing concerns around data privacy and long-term sustainability. Running LLMs locally on mobile and edge devices (on-device LLMs) offers the promise of enhanced privacy, reliability, and reduced communication costs. However, realizing this vision remains challenging due to substantial memory and compute demands, as well as limited visibility into performance-efficiency trade-offs on resource-constrained hardware. We propose lm-Meter, the first lightweight, online latency profiler tailored for on-device LLM inference. lm-Meter captures fine-grained, real-time latency at both phase (e.g., embedding, prefill, decode, softmax, sampling) and kernel levels without auxiliary devices. We implement lm-Meter on commercial mobile platforms and demonstrate its high profiling accuracy with minimal system overhead, e.g., only 2.58% throughput reduction in prefill and 0.99% in decode under the most constrained Powersave governor. Leveraging lm-Meter, we conduct comprehensive empirical studies revealing phase- and kernel-level bottlenecks in on-device LLM inference, quantifying accuracy-efficiency trade-offs, and identifying systematic optimization opportunities. lm-Meter provides unprecedented visibility into the runtime behavior of LLMs on constrained platforms, laying the foundation for informed optimization and accelerating the democratization of on-device LLM systems. Code and tutorials are available at https://github.com/amai-gsu/LM-Meter.

Method: Extends existing prior-data fitted networks through continued pre-training on synthetic data sampled from a customized prior, creating TabPFN-Wide that scales to high-dimensional data.

Result: TabPFN-Wide matches or exceeds base model performance, shows improved robustness to noise, scales beyond 50,000 features while maintaining interpretability, and identifies biologically relevant features that overlap with known findings.

Conclusion: Prior-informed adaptation effectively enhances foundation models for high-dimensional data, providing interpretable feature importance analysis critical for biomedical applications and suggesting potential starting points for future biological studies.

Abstract: Revealing novel insights from the relationship between molecular measurements and pathology remains a very impactful application of machine learning in biomedicine. Data in this domain typically contain only a few observations but thousands of potentially noisy features, posing challenges for conventional machine learning approaches. While prior-data fitted networks emerge as foundation models for tabular data, they are currently not suited to handle large feature counts (>500). Although feature reduction enables their application, it hinders feature importance analysis. We propose a strategy that extends existing models through continued pre-training on synthetic data sampled from a customized prior. The resulting model, TabPFN-Wide, matches or exceeds its base model’s performance while exhibiting improved robustness to noise. It seamlessly scales beyond 50,000 features, regardless of noise levels, while maintaining inherent interpretability, which is critical for biomedical applications. Our results show that prior-informed adaptation is suitable to enhance the capability of foundation models for high-dimensional data. On real-world biomedical datasets many of the most relevant features identified by the model overlap with previous biological findings, while others propose potential starting points for future studies.

Method: Develops a general theory of higher-order feature attribution built on the foundation of Integrated Gradients (IG), extending existing explainable AI frameworks to capture feature interactions.

Result: Discovers natural connections between IG-based feature attribution and statistics/topological signal processing, provides theoretical results to establish the theory, and validates the approach on examples.

Conclusion: The proposed higher-order feature attribution theory successfully addresses the limitations of existing methods in handling feature interactions and establishes meaningful connections to other mathematical domains.

Abstract: Feature attributions are post-training analysis methods that assess how various input features of a machine learning model contribute to an output prediction. Their interpretation is straightforward when features act independently, but becomes less direct when the predictive model involves interactions such as multiplicative relationships or joint feature contributions. In this work, we propose a general theory of higher-order feature attribution, which we develop on the foundation of Integrated Gradients (IG). This work extends existing frameworks in the literature on explainable AI. When using IG as the method of feature attribution, we discover natural connections to statistics and topological signal processing. We provide several theoretical results that establish the theory, and we validate our theory on a few examples.

Method: Theoretical derivation of a lower bound on negative log-likelihood that relates model performance to entropy rates of diffusion processes, with numerical validation on synthetic datasets and investigation of bound tightness.

Result: Established a fundamental connection between diffusion models and thermodynamics, derived performance limits based on entropy rates, and provided insights drawing parallels to Maxwell’s demon with implications for thermodynamic computing hardware.

Conclusion: The framework successfully connects generative modeling performance to fundamental physical principles through stochastic thermodynamics, providing new theoretical understanding of diffusion models from a thermodynamic perspective.

Abstract: We establish a fundamental connection between score-based diffusion models and non-equilibrium thermodynamics by deriving performance limits based on entropy rates. Our main theoretical contribution is a lower bound on the negative log-likelihood of the data that relates model performance to entropy rates of diffusion processes. We numerically validate this bound on a synthetic dataset and investigate its tightness. By building a bridge to entropy rates - system, intrinsic, and exchange entropy - we provide new insights into the thermodynamic operation of these models, drawing parallels to Maxwell’s demon and implications for thermodynamic computing hardware. Our framework connects generative modeling performance to fundamental physical principles through stochastic thermodynamics.

Method: Proposes graph-aware parametric GP using random feature kernel approximation for scalability, ensemble of GPs with adaptive weights for incremental data, and online conformal prediction for robust coverage.

Result: Experimental results show improved coverage and efficient prediction sets compared to existing baselines through adaptive ensembling and conformal prediction thresholding.

Conclusion: The proposed method effectively addresses scalability and coverage issues in graph uncertainty quantification by combining efficient GP approximation with online conformal prediction.

Abstract: Uncertainty quantification (UQ) over graphs arises in a number of safety-critical applications in network science. The Gaussian process (GP), as a classical Bayesian framework for UQ, has been developed to handle graph-structured data by devising topology-aware kernel functions. However, such GP-based approaches are limited not only by the prohibitive computational complexity, but also the strict modeling assumptions that might yield poor coverage, especially with labels arriving on the fly. To effect scalability, we devise a novel graph-aware parametric GP model by leveraging the random feature (RF)-based kernel approximation, which is amenable to efficient recursive Bayesian model updates. To further allow for adaptivity, an ensemble of graph-aware RF-based scalable GPs have been leveraged, with per-GP weight adapted to data arriving incrementally. To ensure valid coverage with robustness to model mis-specification, we wed the GP-based set predictors with the online conformal prediction framework, which post-processes the prediction sets using adaptive thresholds. Experimental results the proposed method yields improved coverage and efficient prediction sets over existing baselines by adaptively ensembling the GP models and setting the key threshold parameters in CP.

Method: Formal analysis of generation processes at abstraction level, using measurable criteria like computational hardness and learnability to compare auto-regressive next-token prediction and masked diffusion methods.

Result: Demonstrated that extending generation capabilities beyond current auto-regressive and masked diffusion approaches to include rewriting and length-variable editing provides significant theoretical and empirical advantages.

Conclusion: Advanced generation processes with rewriting and length-variable editing capabilities offer important improvements for frontier LLMs working on hard problems across diverse domains including coding and science.

Abstract: This paper formally studies generation processes, including auto-regressive next-token prediction and masked diffusion, that abstract beyond architectural specifics. At this level of abstraction, we quantify their benefits and limitations through measurable criteria such as computational hardness and learnability. In particular, we demonstrate that allowing generation to proceed beyond autoregression and current masked diffusion, with capabilities to rewrite and length-variable edit, can bring significant theoretical and empirical advantages, with important implications for frontier LLMs that aspire to tackle increasingly hard problems and work universally across domains beyond natural language, such as coding and science.

Method: Comprehensive analysis of quantization degradation across open-source LLM training trajectories up to 32B parameters and 15T tokens, plus controlled experiments training models up to 100B tokens with specific hyperparameter interventions.

Result: Quantization errors are driven by interplay between learning rate and other hyperparameters. Once learning rates decay, validation loss and quantization error diverge independently of training data scale. Strategic training interventions can improve quantization quality.

Conclusion: Increasing dataset scale doesn’t inherently compromise quantization effectiveness; strategic training hyperparameter interventions can improve quantization quality at scale.

Abstract: While post-training quantization is widely adopted for efficient deployment of large language models, the mechanisms underlying quantization robustness remain unclear. We conduct a comprehensive analysis of quantization degradation across open-source language model training trajectories up to 32B parameters and 15T training tokens to accurately assess the relationship between training dynamics and quantization performance. Our key finding is that quantization errors in large-scale training runs are driven by a complex interplay between learning rate and other training hyperparameters. Specifically, once learning rates decay, validation loss and quantization error diverge, largely independent of training data scale. To investigate interventions on the training dynamics and identify specific configurations that can modulate quantization robustness favorably, we train our own models in controlled experiments up to 100B tokens. Our results challenge the assumption that increasing dataset scale inherently compromises quantization effectiveness, demonstrating instead that strategic training hyperparameter interventions can improve quantization quality at scale.

Method: Analyzes various deep neural network architectures including convolutional networks and diffusion models, explains their structures and working principles, and provides detailed comparisons of different models.

Result: Showcases milestone examples from DeepDream to recent models like Stable Diffusion and DALL-E 3, highlighting strengths and limitations of each approach and demonstrating remarkable progress in AI art generation.

Conclusion: Deep neural networks have made significant advances in AI-generated art in a short time, exemplifying the successful interaction between art and computer science through increasingly sophisticated image generation capabilities.

Abstract: This paper delves into the fascinating field of AI-generated art and explores the various deep neural network architectures and models that have been utilized to create it. From the classic convolutional networks to the cutting-edge diffusion models, we examine the key players in the field. We explain the general structures and working principles of these neural networks. Then, we showcase examples of milestones, starting with the dreamy landscapes of DeepDream and moving on to the most recent developments, including Stable Diffusion and DALL-E 3, which produce mesmerizing images. We provide a detailed comparison of these models, highlighting their strengths and limitations, and examining the remarkable progress that deep neural networks have made so far in a short period of time. With a unique blend of technical explanations and insights into the current state of AI-generated art, this paper exemplifies how art and computer science interact.

Method: CentroBind uses adaptively adjustable centroid-based anchors generated from all available modalities, theoretically capturing intra-modal learning, inter-modal learning, and multi-modal alignment while constructing unified representations.

Result: Experiments on synthetic and real-world datasets show that adaptive anchor methods like CentroBind consistently outperform fixed anchor binding methods.

Conclusion: Adaptive anchor binding methods provide superior performance over fixed anchor approaches by creating more balanced and rich representation spaces that better capture multi-modal relationships.

Abstract: A unified representation space in multi-modal learning is essential for effectively integrating diverse data sources, such as text, images, and audio, to enhance efficiency and performance across various downstream tasks. Recent binding methods, such as ImageBind, typically rely on a single, fixed anchor modality for aligning multi-modal data. We mathematically analyze these fixed anchor binding methods and uncover significant limitations: (1) over-reliance on the choice of the anchor modality, (2) inadequate capture of intra-modal information, and (3) failure to account for cross-modal correlation among non-anchored modalities. To address these issues, we propose the need for adaptive anchor binding methods, exemplified by our framework CentroBind. The proposed method uses adaptively adjustable centroid-based anchors generated from all available modalities, leading to a balanced and rich representation space. We theoretically demonstrate that our approach captures three critical properties of multi-modal learning – intra-modal learning, inter-modal learning, and multi-modal alignment – while constructing a unified representation that spans all modalities. Experiments on both synthetic and real-world datasets show that adaptive anchor methods such as CentroBind consistently outperform fixed anchor binding methods, verifying our analysis.

Method: Proposes a benchmark framework with nested cross-validation and various unsupervised model selection scores, evaluating shallow algorithms (reweighting, mapping, subspace alignment) on simulated datasets with controlled shifts and real-world datasets across images, text, biomedical, and tabular data.

Result: The benchmark highlights the importance of realistic validation and provides practical guidance for real-life applications, offering key insights into model selection approaches.

Conclusion: SKADA-bench enables comprehensive, fair evaluation of DA methods across diverse modalities and is designed as an open-source, reproducible framework that can be easily extended with new methods, datasets, and criteria.

Abstract: Unsupervised Domain Adaptation (DA) consists of adapting a model trained on a labeled source domain to perform well on an unlabeled target domain with some data distribution shift. While many methods have been proposed in the literature, fair and realistic evaluation remains an open question, particularly due to methodological difficulties in selecting hyperparameters in the unsupervised setting. With SKADA-bench, we propose a framework to evaluate DA methods on diverse modalities, beyond computer vision task that have been largely explored in the literature. We present a complete and fair evaluation of existing shallow algorithms, including reweighting, mapping, and subspace alignment. Realistic hyperparameter selection is performed with nested cross-validation and various unsupervised model selection scores, on both simulated datasets with controlled shifts and real-world datasets across diverse modalities, such as images, text, biomedical, and tabular data. Our benchmark highlights the importance of realistic validation and provides practical guidance for real-life applications, with key insights into the choice and impact of model selection approaches. SKADA-bench is open-source, reproducible, and can be easily extended with novel DA methods, datasets, and model selection criteria without requiring re-evaluating competitors. SKADA-bench is available on Github at https://github.com/scikit-adaptation/skada-bench.

Method: Introduced an empirical analysis framework to measure completeness and selectivity, enabling direct comparisons between different families of causal probing methods (linear vs. nonlinear, concept removal vs. counterfactual interventions).

Result: All methods show a clear tradeoff between completeness and selectivity; more complete and reliable methods have greater impact on LLM behavior; nonlinear interventions are almost always more reliable than linear interventions.

Conclusion: The proposed framework provides systematic evaluation of causal probing methods, revealing inherent tradeoffs and demonstrating the superiority of nonlinear interventions in reliability.

Abstract: Causal probing aims to analyze foundation models by examining how intervening on their representation of various latent properties impacts their outputs. Recent works have cast doubt on the theoretical basis of several leading causal probing methods, but it has been unclear how to systematically evaluate the effectiveness of these methods in practice. To address this, we define two key causal probing desiderata: completeness (how thoroughly the representation of the target property has been transformed) and selectivity (how little non-targeted properties have been impacted). We find that there is an inherent tradeoff between the two, which we define as reliability, their harmonic mean. We introduce an empirical analysis framework to measure and evaluate these quantities, allowing us to make the first direct comparisons between different families of leading causal probing methods (e.g., linear vs. nonlinear, or concept removal vs. counterfactual interventions). We find that: (1) all methods show a clear tradeoff between completeness and selectivity; (2) more complete and reliable methods have a greater impact on LLM behavior; and (3) nonlinear interventions are almost always more reliable than linear interventions.

Method: DIST2Loss transforms continuous exponential family distributions from inherent distance metrics into discrete categorical optimization targets compatible with existing model architectures, enabling models to learn and preserve distance relationships during token generation.

Result: Empirical evaluations show consistent performance gains in diverse multimodal applications including visual grounding, robotic manipulation, generative reward modeling, and image generation using vector-quantized features.

Conclusion: DIST2Loss effectively improves model performance by incorporating distance awareness, with particularly strong benefits in low-data regimes and resource-constrained scenarios.

Abstract: As large language models expand beyond natural language to domains such as mathematics, multimodal understanding, and embodied agents, tokens increasingly reflect metric relationships rather than purely linguistic meaning. We introduce DIST2Loss, a distance-aware framework designed to train autoregressive discrete models by leveraging predefined distance relationships among output tokens. At its core, DIST2Loss transforms continuous exponential family distributions derived from inherent distance metrics into discrete, categorical optimization targets compatible with the models’ architectures. This approach enables the models to learn and preserve meaningful distance relationships during token generation while maintaining compatibility with existing architectures. Empirical evaluations show consistent performance gains in diverse multimodal applications, including visual grounding, robotic manipulation, generative reward modeling, and image generation using vector-quantized features. These improvements are most notable in low-data regimes, demonstrating DIST2Loss’s strength under resource constraints.

Method: The paper conducts a structured review of explainable clustering methods, identifies key criteria to distinguish between different approaches, and creates a taxonomy with an open repository.

Result: The survey provides insights to help researchers select appropriate explainable clustering methods for specific contexts and promotes development of efficient yet transparent clustering algorithms.

Conclusion: An open repository organizes representative interpretable clustering methods under the proposed taxonomy, available at https://github.com/hulianyu/Awesome-Interpretable-Clustering for easy reference.

Abstract: In recent years, much of the research on clustering algorithms has primarily focused on enhancing their accuracy and efficiency, frequently at the expense of interpretability. However, as these methods are increasingly being applied in high-stakes domains such as healthcare, finance, and autonomous systems, the need for transparent and interpretable clustering outcomes has become a critical concern. This is not only necessary for gaining user trust but also for satisfying the growing ethical and regulatory demands in these fields. Ensuring that decisions derived from clustering algorithms can be clearly understood and justified is now a fundamental requirement. To address this need, this paper provides a comprehensive and structured review of the current state of explainable clustering algorithms, identifying key criteria to distinguish between various methods. These insights can effectively assist researchers in making informed decisions about the most suitable explainable clustering methods for specific application contexts, while also promoting the development and adoption of clustering algorithms that are both efficient and transparent. For convenient access and reference, an open repository organizes representative and emerging interpretable clustering methods under the taxonomy proposed in this survey, available at https://github.com/hulianyu/Awesome-Interpretable-Clustering

Method: Used Linear Regression, Extreme Gradient Boosting (XGB), and Genetic Algorithm Optimization applied to XGB to forecast solar irradiation using data from three SURFRAD network stations in the US.

Result: Models predict Global Horizontal Index (GHI) and are compared, with Genetic Algorithm Optimization further improving XGB’s prediction accuracy.

Conclusion: Machine learning techniques, particularly Genetic Algorithm-optimized XGBoost, can effectively forecast solar irradiation with high accuracy for power grid management.

Abstract: Renewable energy forecasting is attaining greater importance due to its constant increase in contribution to the electrical power grids. Solar energy is one of the most significant contributors to renewable energy and is dependent on solar irradiation. For the effective management of electrical power grids, forecasting models that predict solar irradiation, with high accuracy, are needed. In the current study, Machine Learning techniques such as Linear Regression, Extreme Gradient Boosting and Genetic Algorithm Optimization are used to forecast solar irradiation. The data used for training and validation is recorded from across three different geographical stations in the United States that are part of the SURFRAD network. A Global Horizontal Index (GHI) is predicted for the models built and compared. Genetic Algorithm Optimization is applied to XGB to further improve the accuracy of solar irradiation prediction.

Method: Proposes diffusion expansion theory of f-divergence family and introduces key theories to overcome intractability, resulting in an equivalent yet tractable loss for training one-step diffusion models.

Result: Achieves record-breaking FID scores: 1.46 (CIFAR10 unconditional), 1.38 (CIFAR10 conditional), and 1.02 (ImageNet-64x64), outperforming 79-step teacher diffusion by significant margin. Also shows decent performance in text-to-3D generation.

Conclusion: Uni-Instruct provides both theoretical unification and empirical advances in one-step diffusion distillation, potentially benefiting future studies on diffusion model knowledge transfer.

Abstract: In this paper, we unify more than 10 existing one-step diffusion distillation approaches, such as Diff-Instruct, DMD, SIM, SiD, $f$-distill, etc, inside a theory-driven framework which we name the \textbf{\emph{Uni-Instruct}}. Uni-Instruct is motivated by our proposed diffusion expansion theory of the $f$-divergence family. Then we introduce key theories that overcome the intractability issue of the original expanded $f$-divergence, resulting in an equivalent yet tractable loss that effectively trains one-step diffusion models by minimizing the expanded $f$-divergence family. The novel unification introduced by Uni-Instruct not only offers new theoretical contributions that help understand existing approaches from a high-level perspective but also leads to state-of-the-art one-step diffusion generation performances. On the CIFAR10 generation benchmark, Uni-Instruct achieves record-breaking Frechet Inception Distance (FID) values of \textbf{\emph{1.46}} for unconditional generation and \textbf{\emph{1.38}} for conditional generation. On the ImageNet-$64\times 64$ generation benchmark, Uni-Instruct achieves a new SoTA one-step generation FID of \textbf{\emph{1.02}}, which outperforms its 79-step teacher diffusion with a significant improvement margin of 1.33 (1.02 vs 2.35). We also apply Uni-Instruct on broader tasks like text-to-3D generation. For text-to-3D generation, Uni-Instruct gives decent results, which slightly outperforms previous methods, such as SDS and VSD, in terms of both generation quality and diversity. Both the solid theoretical and empirical contributions of Uni-Instruct will potentially help future studies on one-step diffusion distillation and knowledge transferring of diffusion models.

Method: The method estimates the Brenier optimal transport map from the current prior state distribution to the posterior distribution at the next time step, using neural networks to model complex distributions and stochastic optimization for scalability.

Result: Extensive numerical experiments show the OT method outperforms SIR particle filters and ensemble Kalman filters in sample efficiency, high-dimensional scalability, and ability to capture complex multi-modal distributions.

Conclusion: The optimal transport formulation provides a viable alternative to conventional particle filters, offering advantages in handling complex distributions and high-dimensional problems without requiring analytical likelihood forms.

Abstract: This paper is concerned with the problem of nonlinear filtering, i.e., computing the conditional distribution of the state of a stochastic dynamical system given a history of noisy partial observations. Conventional sequential importance resampling (SIR) particle filters suffer from fundamental limitations, in scenarios involving degenerate likelihoods or high-dimensional states, due to the weight degeneracy issue. In this paper, we explore an alternative method, which is based on estimating the Brenier optimal transport (OT) map from the current prior distribution of the state to the posterior distribution at the next time step. Unlike SIR particle filters, the OT formulation does not require the analytical form of the likelihood. Moreover, it allows us to harness the approximation power of neural networks to model complex and multi-modal distributions and employ stochastic optimization algorithms to enhance scalability. Extensive numerical experiments are presented that compare the OT method to the SIR particle filter and the ensemble Kalman filter, evaluating the performance in terms of sample efficiency, high-dimensional scalability, and the ability to capture complex and multi-modal distributions.

Method: Propose SOLID framework that secures bottom layers in hardware-secured devices, introduces an efficient metric to determine optimal number of secured layers, and analyzes trade-off between protection and customization.

Result: Extensive experiments on models from 1.3B to 70B parameters show SOLID outperforms baselines, providing stronger protection against distillation attacks while maintaining comparable customization performance.

Conclusion: Securing bottom layers before transition layers provides better protection than top layers, and SOLID effectively balances model confidentiality with downstream customization needs for on-premises LLM deployment.

Abstract: Privacy-sensitive users require deploying large language models (LLMs) within their own infrastructure (on-premises) to safeguard private data and enable customization. However, vulnerabilities in local environments can lead to unauthorized access and potential model theft. To address this, prior research on small models has explored securing only the output layer within hardware-secured devices to balance model confidentiality and customization. Yet this approach fails to protect LLMs effectively. In this paper, we discover that (1) query-based distillation attacks targeting the secured top layer can produce a functionally equivalent replica of the victim model; (2) securing the same number of layers, bottom layers before a transition layer provide stronger protection against distillation attacks than top layers, with comparable effects on customization performance; and (3) the number of secured layers creates a trade-off between protection and customization flexibility. Based on these insights, we propose SOLID, a novel deployment framework that secures a few bottom layers in a secure environment and introduces an efficient metric to optimize the trade-off by determining the ideal number of hidden layers. Extensive experiments on five models (1.3B to 70B parameters) demonstrate that SOLID outperforms baselines, achieving a better balance between protection and downstream customization.

Method: The authors propose and investigate three types of multi-trigger attacks: parallel, sequential, and hybrid attacks, where multiple triggers can coexist, overwrite, or cross-activate each other.

Result: MTBAs successfully break the prevalent shortcut assumption underlying most existing backdoor detection and removal methods, rendering them ineffective against these more sophisticated attacks.

Conclusion: The paper highlights the security risk of MTBAs, provides a multi-trigger backdoor poisoning dataset for future research, and discusses potential defense strategies against these advanced attacks.

Abstract: Backdoor attacks have become a significant threat to the pre-training and deployment of deep neural networks (DNNs). Although numerous methods for detecting and mitigating backdoor attacks have been proposed, most rely on identifying and eliminating the ``shortcut" created by the backdoor, which links a specific source class to a target class. However, these approaches can be easily circumvented by designing multiple backdoor triggers that create shortcuts everywhere and therefore nowhere specific. In this study, we explore the concept of Multi-Trigger Backdoor Attacks (MTBAs), where multiple adversaries leverage different types of triggers to poison the same dataset. By proposing and investigating three types of multi-trigger attacks including \textit{parallel}, \textit{sequential}, and \textit{hybrid} attacks, we demonstrate that 1) multiple triggers can coexist, overwrite, or cross-activate one another, and 2) MTBAs easily break the prevalent shortcut assumption underlying most existing backdoor detection/removal methods, rendering them ineffective. Given the security risk posed by MTBAs, we have created a multi-trigger backdoor poisoning dataset to facilitate future research on detecting and mitigating these attacks, and we also discuss potential defense strategies against MTBAs. Our code is available at https://github.com/bboylyg/Multi-Trigger-Backdoor-Attacks.

Method: Develop a theoretical connection between the proposed metric and FL training dynamics, then evaluate on synthetic, benchmark, and medical imaging datasets.

Result: The metric shows robust and interpretable relationship with model performance and can be calculated in privacy-preserving manner without model training.

Conclusion: This first federated dataset similarity metric can better facilitate successful collaborations between sites in FL scenarios.

Abstract: Federated Learning is increasingly used in domains such as healthcare to facilitate collaborative model training without data-sharing. However, datasets located in different sites are often non-identically distributed, leading to degradation of model performance in FL. Most existing methods for assessing these distribution shifts are limited by being dataset or task-specific. Moreover, these metrics can only be calculated by exchanging data, a practice restricted in many FL scenarios. To address these challenges, we propose a novel metric for assessing dataset similarity. Our metric exhibits several desirable properties for FL: it is dataset-agnostic, is calculated in a privacy-preserving manner, and is computationally efficient, requiring no model training. In this paper, we first establish a theoretical connection between our metric and training dynamics in FL. Next, we extensively evaluate our metric on a range of datasets including synthetic, benchmark, and medical imaging datasets. We demonstrate that our metric shows a robust and interpretable relationship with model performance and can be calculated in privacy-preserving manner. As the first federated dataset similarity metric, we believe this metric can better facilitate successful collaborations between sites.

Method: Uses a multi-agent framework with LLMs that decomposes benchmark creation into planning, generation, verification, and evaluation phases, with agents interacting and incorporating developer feedback.

Result: Successfully created benchmarks for planning, constraint satisfaction, and causal reasoning across language and vision modalities, enabling study of state-of-the-art models.

Conclusion: BenchAgents provides an automated, scalable approach to benchmark creation that reveals new insights into model failure modes and differences.

Abstract: Evaluation insights are limited by the availability of high-quality benchmarks. As models evolve, there is a need to create benchmarks that can measure progress on new and complex generative capabilities. However, manually creating new benchmarks is slow and expensive, restricting comprehensive evaluations for any capability. We introduce BenchAgents, a multi-agent framework that methodically leverages large language models (LLMs) to automate evaluation benchmark creation while inherently ensuring data and (evaluation) metric quality. BenchAgents decomposes the benchmark creation process into planning, generation, verification, and evaluation, each of which is ] orchestrated via LLM agents. These agents interact with each other and utilize feedback from benchmark developers to improve and flexibly control data diversity and quality. We use BenchAgents to create benchmarks to evaluate capabilities related to planning, constraint satisfaction, and causal reasoning spanning both language and vision modalities. We then use these benchmarks to study state-of-the-art models and extract new insights into common failure modes and model differences.

Method: The authors compare standard PCA with sparse PCA (SPCA) as front-end feature extractors, provide theoretical robustness certificates for linear heads, and analyze sparsity effects through Lipschitz composition arguments.

Result: SPCA consistently outperforms PCA under strong white-box and black-box attacks while maintaining competitive clean accuracy, with sparser projections reducing adversarial leverage.

Conclusion: Sparse PCA provides effective adversarial robustness through sparser projections that reduce input sensitivity, with benefits persisting beyond linear settings as demonstrated by both theory and experiments.

Abstract: Deep neural networks perform remarkably well on image classification tasks but remain vulnerable to carefully crafted adversarial perturbations. This work revisits linear dimensionality reduction as a simple, data-adapted defense. We empirically compare standard Principal Component Analysis (PCA) with its sparse variant (SPCA) as front-end feature extractors for downstream classifiers, and we complement these experiments with a theoretical analysis. On the theory side, we derive exact robustness certificates for linear heads applied to SPCA features: for both $\ell_\infty$ and $\ell_2$ threat models (binary and multiclass), the certified radius grows as the dual norms of $W^\top u$ shrink, where $W$ is the projection and $u$ the head weights. We further show that for general (non-linear) heads, sparsity reduces operator-norm bounds through a Lipschitz composition argument, predicting lower input sensitivity. Empirically, with a small non-linear network after the projection, SPCA consistently degrades more gracefully than PCA under strong white-box and black-box attacks while maintaining competitive clean accuracy. Taken together, the theory identifies the mechanism (sparser projections reduce adversarial leverage) and the experiments verify that this benefit persists beyond the linear setting. Our code is available at https://github.com/killian31/SPCARobustness.

Method: The authors propose a classification system with two comparator types: CP (standard comparator requiring identical non-protected attributes) and MM (allowing dissimilar non-protected attributes while removing protected attribute effects). They illustrate these concepts using a real-world example.

Result: The paper demonstrates that MM comparators provide an alternative approach to discrimination testing that can handle more complex scenarios where complainant-comparator pairs may differ in non-protected attributes.

Conclusion: The MM comparator represents a more sophisticated approach to discrimination testing that leverages causal modeling and offers significant potential for machine learning applications in this domain.

Abstract: Testing for individual discrimination involves deriving a profile, the comparator, similar to the one making the discrimination claim, the complainant, based on a protected attribute, such as race or gender, and comparing their decision outcomes. The complainant-comparator pair is central to discrimination testing. Most discrimination testing tools rely on this pair to establish evidence for discrimination. In this work we revisit the role of the comparator in discrimination testing. We first argue for the inherent causal modeling nature of deriving the comparator. We then introduce a two-kinds classification for the comparator: the ceteris paribus, orwith all else equal,'' (CP) comparator and the mutatis mutandis, or with the appropriate adjustments being made,’’ (MM) comparator. The CP comparator is the standard comparator, representing an idealized comparison for establishing discrimination as it aims for a complainant-comparator pair that only differs on membership to the protected attribute. As an alternative to it, we define the MM comparator, which requires that the comparator represents the``what would have been of’’ the complainant without the effects of the protected attribute on the non-protected attributes. Under the MM comparator, the complainant-comparator pair can be dissimilar in terms of the non-protected attributes, departing from an idealized comparison. Notably, the MM comparator is a more complex kind of comparator and its implementation offers an impactful venue for machine learning methods. We illustrate these two comparators and their impact on discrimination testing using a real-world example.

Method: Proposes HOG-Diff framework that follows a coarse-to-fine generation curriculum guided by higher-order topology and implemented via diffusion bridges, with theoretical guarantees stronger than classical diffusion frameworks.

Result: Extensive experiments on molecular and generic graph generation tasks show the method consistently outperforms or remains competitive with state-of-the-art baselines.

Conclusion: HOG-Diff provides a principled framework for graph generation that effectively captures topological properties through higher-order guidance, demonstrating superior performance across various graph generation tasks.

Abstract: Graph generation is a critical yet challenging task as empirical analyses require a deep understanding of complex, non-Euclidean structures. Diffusion models have recently made significant achievements in graph generation, but these models are typically adapted from image generation frameworks and overlook inherent higher-order topology, leaving them ill-suited for capturing the topological properties of graphs. In this work, we propose Higher-order Guided Diffusion (HOG-Diff), a principled framework that progressively generates plausible graphs with inherent topological structures. HOG-Diff follows a coarse-to-fine generation curriculum guided by higher-order topology and implemented via diffusion bridges. We further prove that our model exhibits a stronger theoretical guarantee than classical diffusion frameworks. Extensive experiments on both molecular and generic graph generation tasks demonstrate that our method consistently outperforms or remains competitive with state-of-the-art baselines. Our code is available at https://github.com/Yiminghh/HOG-Diff.

Method: Propose UniAug - a universal graph structure augmentor built on a discrete diffusion model. Pre-train the diffusion model on thousands of graphs across domains to learn structural patterns, then use guided generation for adaptive graph structure augmentation in downstream tasks.

Result: By leveraging the pre-trained diffusion model for structure augmentation, the method consistently achieves performance improvements across various downstream tasks in a plug-and-play manner.

Conclusion: This represents the first demonstration of a data-scaling graph structure augmentor that works effectively across different graph domains, enabling the ‘better with more’ phenomenon for graph data.

Abstract: Models for natural language and images benefit from data scaling behavior: the more data fed into the model, the better they perform. This ‘better with more’ phenomenon enables the effectiveness of large-scale pre-training on vast amounts of data. However, current graph pre-training methods struggle to scale up data due to heterogeneity across graphs. To achieve effective data scaling, we aim to develop a general model that is able to capture diverse data patterns of graphs and can be utilized to adaptively help the downstream tasks. To this end, we propose UniAug, a universal graph structure augmentor built on a diffusion model. We first pre-train a discrete diffusion model on thousands of graphs across domains to learn the graph structural patterns. In the downstream phase, we provide adaptive enhancement by conducting graph structure augmentation with the help of the pre-trained diffusion model via guided generation. By leveraging the pre-trained diffusion model for structure augmentation, we consistently achieve performance improvements across various downstream tasks in a plug-and-play manner. To the best of our knowledge, this study represents the first demonstration of a data-scaling graph structure augmentor on graphs across domains.

Method: Created the Gemstones dataset with over 4000 transformer checkpoints (up to 2B parameters) featuring diverse architectural shapes, including ablations over learning rate and cooldown parameters.

Result: Found that scaling law prescriptions are highly sensitive to experimental design process and specific model checkpoints used during fitting, challenging the robustness of traditional scaling law studies.

Conclusion: Scaling law prescriptions depend significantly on architectural and hyperparameter choices, and the released Gemstones dataset enables more complex scaling studies to better understand these relationships.

Abstract: Scaling laws are typically fit using a family of models with a narrow range of frozen hyperparameter choices. In this work we study scaling laws using multiple architectural shapes and hyperparameter choices, highlighting their impact on resulting prescriptions. As a primary artifact of our research, we release the Gemstones: an open-source scaling law dataset, consisting of over 4000 checkpoints from transformers with up to 2 billion parameters and diverse architectural shapes; including ablations over learning rate and cooldown. Our checkpoints enable more complex studies of scaling, such as analyzing the relationship between width and depth. By examining our model suite, we find that the prescriptions of scaling laws can be highly sensitive to the experimental design process and the specific model checkpoints used during fitting.

Method: Combines existing ToMATo clustering algorithm with a bootstrapping procedure to identify significant peaks in estimated density functions, with default parameter choices provided.

Result: AuToMATo performs favorably against parameter-free clustering algorithms and often significantly outperforms even the best parameter selections for other algorithms, especially when used with Mapper.

Conclusion: AuToMATo is an effective out-of-the-box clustering solution with an open-source Python implementation compatible with scikit-learn, particularly valuable for topological data analysis applications.

Abstract: We present AuToMATo, a novel clustering algorithm based on persistent homology. While AuToMATo is not parameter-free per se, we provide default choices for its parameters that make it into an out-of-the-box clustering algorithm that performs well across the board. AuToMATo combines the existing ToMATo clustering algorithm with a bootstrapping procedure in order to separate significant peaks of an estimated density function from non-significant ones. We perform a thorough comparison of AuToMATo (with its parameters fixed to their defaults) against many other state-of-the-art clustering algorithms. We find not only that AuToMATo compares favorably against parameter-free clustering algorithms, but in many instances also significantly outperforms even the best selection of parameters for other algorithms. AuToMATo is motivated by applications in topological data analysis, in particular the Mapper algorithm, where it is desirable to work with a clustering algorithm that does not need tuning of its parameters. Indeed, we provide evidence that AuToMATo performs well when used with Mapper. Finally, we provide an open-source implementation of AuToMATo in Python that is fully compatible with the standard scikit-learn architecture.

Method: Proposed CeFeGNN with two key components: 1) Embedding learnable cell attributes to upgrade message passing from first-order to higher-order (volume+edge to node), 2) A feature-enhanced block to improve performance and alleviate over-smoothing.

Result: Extensive experiments on various PDE systems and real-world datasets show CeFeGNN achieves superior performance compared to other baseline methods.

Conclusion: The proposed CeFeGNN framework successfully enhances GNN representation learning for spatiotemporal dynamics by incorporating volumetric information and feature enhancement, demonstrating state-of-the-art performance.

Abstract: Data-driven learning of physical systems has kindled significant attention, where many neural models have been developed. In particular, mesh-based graph neural networks (GNNs) have demonstrated significant potential in modeling spatiotemporal dynamics across arbitrary geometric domains. However, the existing node-edge message-passing and aggregation mechanism in GNNs limits the representation learning ability. In this paper, we proposed a dual-module framework, Cell-embedded and Feature-enhanced Graph Neural Network (aka, CeFeGNN), for learning spatiotemporal dynamics. Specifically, we embed learnable cell attributions to the common node-edge message passing process, which better captures the spatial dependency of regional features. Such a strategy essentially upgrades the local aggregation scheme from first order (e.g., from edge to node) to a higher order (e.g., from volume and edge to node), which takes advantage of volumetric information in message passing. Meanwhile, a novel feature-enhanced block is designed to further improve the model’s performance and alleviate the over-smoothness problem. Extensive experiments on various PDE systems and one real-world dataset demonstrate that CeFeGNN achieves superior performance compared with other baselines.

Method: Uses dense neural networks with dynamic sparse input layers and employs entry-based pruning that compares neurons based on their relative change when they enter the network.

Result: Outperforms current state-of-the-art methods on 13 datasets, particularly improves average accuracy on low-dimensional datasets, and achieves lower runtime than competing approaches.

Conclusion: EntryPrune demonstrates superior performance over traditional feature selection methods and provides an effective framework for feature selection using neural networks with entry-based pruning.

Abstract: There is an ongoing effort to develop feature selection algorithms to improve interpretability, reduce computational resources, and minimize overfitting in predictive models. Neural networks stand out as architectures on which to build feature selection methods, and recently, neuron pruning and regrowth have emerged from the sparse neural network literature as promising new tools. We introduce EntryPrune, a novel supervised feature selection algorithm using a dense neural network with a dynamic sparse input layer. It employs entry-based pruning, a novel approach that compares neurons based on their relative change induced when they have entered the network. Extensive experiments on 13 different datasets show that our approach generally outperforms the current state-of-the-art methods, and in particular improves the average accuracy on low-dimensional datasets. Furthermore, we show that EntryPruning surpasses traditional techniques such as magnitude pruning within the EntryPrune framework and that EntryPrune achieves lower runtime than competing approaches. Our code is available at https://github.com/flxzimmer/entryprune.

Method: Dynamic uncertainty-aware aggregation that weights subset models based on accuracy and prediction uncertainty measured on a shared validation set.

Result: Significantly improves generalization on benchmark datasets across multiple domains, even when combined with other regularization methods.

Conclusion: DUA-D2C provides a theoretically grounded and effective approach to combat overfitting in deep learning through intelligent model aggregation.

Abstract: Overfitting remains a significant challenge in deep learning, often arising from data outliers, noise, and limited training data. To address this, the Divide2Conquer (D2C) method was previously proposed, which partitions training data into multiple subsets and trains identical models independently on each. This strategy enables learning more consistent patterns while minimizing the influence of individual outliers and noise. However, D2C’s standard aggregation typically treats all subset models equally or based on fixed heuristics (like data size), potentially underutilizing information about their varying generalization capabilities. Building upon this foundation, we introduce Dynamic Uncertainty-Aware Divide2Conquer (DUA-D2C), an advanced technique that refines the aggregation process. DUA-D2C dynamically weights the contributions of subset models based on their performance on a shared validation set, considering both accuracy and prediction uncertainty. This intelligent aggregation allows the central model to preferentially learn from subsets yielding more generalizable and confident edge models, thereby more effectively combating overfitting. Empirical evaluations on benchmark datasets spanning multiple domains demonstrate that DUA-D2C significantly improves generalization. Our analysis includes evaluations of decision boundaries, loss curves, and other performance metrics, highlighting the effectiveness of DUA-D2C. This study demonstrates that DUA-D2C improves generalization performance even when applied on top of other regularization methods, establishing it as a theoretically grounded and effective approach to combating overfitting in modern deep learning. Our codes are publicly available at: https://github.com/Saiful185/DUA-D2C.

Method: Frames prompt optimization as structured AutoML using successive halving to efficiently navigate combinatorial spaces of agentic/non-agentic prompting patterns and demonstrations, implemented via PDL programming language.

Result: Achieved consistent accuracy gains (9.21±15.46 percentage points, up to 67.5pp) across three tasks and seven LLMs (3B-70B parameters), with selected strategies varying across models and tasks.

Conclusion: AutoPDL successfully automates LLM prompt optimization, producing human-readable PDL programs that enable source-to-source optimization and human-in-the-loop refinement.

Abstract: The performance of large language models (LLMs) depends on how they are prompted, with choices spanning both the high-level prompting pattern (e.g., Zero-Shot, CoT, ReAct, ReWOO) and the specific prompt content (instructions and few-shot demonstrations). Manually tuning this combination is tedious, error-prone, and specific to a given LLM and task. Therefore, this paper proposes AutoPDL, an automated approach to discovering good LLM agent configurations. Our approach frames this as a structured AutoML problem over a combinatorial space of agentic and non-agentic prompting patterns and demonstrations, using successive halving to efficiently navigate this space. We introduce a library implementing common prompting patterns using the PDL prompt programming language. AutoPDL solutions are human-readable, editable, and executable PDL programs that use this library. This approach also enables source-to-source optimization, allowing human-in-the-loop refinement and reuse. Evaluations across three tasks and seven LLMs (ranging from 3B to 70B parameters) show consistent accuracy gains ($9.21\pm15.46$ percentage points), up to 67.5pp, and reveal that selected prompting strategies vary across models and tasks.

Method: Tested models in two environments: ‘Feature World’ for simple information gathering and a text-based ‘Alchemy’ environment for complex meta-learning, using prompting strategies like observation summarization.

Result: Models performed well in simple tasks but initially failed in complex Alchemy environment. With observation summarization prompts, they achieved emergent meta-learning, improved across trials, and adapted to rule changes. Gemini 2.5 performed best, followed by Claude 3.7, while ChatGPT-4o and o4-mini struggled.

Conclusion: The main challenge for foundation models is integrating knowledge through adaptive strategies over time rather than selecting informative actions. There’s no intrinsic barrier to future models mastering these abilities with proper prompting techniques.

Abstract: Foundation models excel at single-turn reasoning but struggle with multi-turn exploration in dynamic environments, a requirement for many real-world challenges. We evaluated these models on their ability to learn from experience, adapt, and gather information. First, in “Feature World,” a simple setting for testing information gathering, models performed near-optimally. However, to test more complex, multi-trial learning, we implemented a text-based version of the “Alchemy” environment, a benchmark for meta-learning. Here, agents must deduce a latent causal structure by integrating information across many trials. In this setting, recent foundation models initially failed to improve their performance over time. Crucially, we found that prompting the models to summarize their observations at regular intervals enabled an emergent meta-learning process. This allowed them to improve across trials and even adaptively re-learn when the environment’s rules changed unexpectedly. While most models handled the simple task, Alchemy revealed stark differences in robustness: Gemini 2.5 performed best, followed by Claude 3.7, while ChatGPT-4o and o4-mini struggled. This underscores Alchemy’s value as a benchmark. Our findings demonstrate that the biggest challenge for foundation models is not selecting informative actions in the moment, but integrating knowledge through adaptive strategies over time. Encouragingly, there appears to be no intrinsic barrier to future models mastering these abilities.

Method: Leverages linear representation hypothesis and concept vectors to add logical implication by steering model generation behavior through vector additions to activations.

Result: Enables hand-engineered reasoning capabilities by integrating neuro-symbolic logic into pre-trained transformer models.

Conclusion: LIMS unlocks new reasoning capabilities in transformers through interpretable logical implication steering based on concept vectors.

Abstract: The field of mechanistic interpretability in pre-trained transformer models has demonstrated substantial evidence supporting the ‘’linear representation hypothesis’’, which is the idea that high level concepts are encoded as vectors in the space of activations of a model. Studies also show that model generation behavior can be steered toward a given concept by adding the concept’s vector to the corresponding activations. We show how to leverage these properties to build a form of logical implication into models, enabling transparent and interpretable adjustments that induce a chosen generation behavior in response to the presence of any given concept. Our method, Logical Implication Model Steering (LIMS), unlocks new hand engineered reasoning capabilities by integrating neuro-symbolic logic into pre-trained transformer models.

Method: The authors provide theoretical proof of failure under label blindness conditions, define a new Adjacent OOD detection task, and conduct experiments to validate their theory using existing unlabeled OOD methods.

Result: Experiments demonstrate that existing unlabeled OOD detection methods fail under the label blindness conditions identified in the theoretical framework.

Conclusion: The paper reveals fundamental limitations in current unlabeled OOD detection approaches and proposes Adjacent OOD detection as a more realistic benchmark to address safety gaps in future research.

Abstract: Out-of-distribution (OOD) detection methods have recently become more prominent, serving as a core element in safety-critical autonomous systems. One major purpose of OOD detection is to reject invalid inputs that could lead to unpredictable errors and compromise safety. Due to the cost of labeled data, recent works have investigated the feasibility of self-supervised learning (SSL) OOD detection, unlabeled OOD detection, and zero shot OOD detection. In this work, we identify a set of conditions for a theoretical guarantee of failure in unlabeled OOD detection algorithms from an information-theoretic perspective. These conditions are present in all OOD tasks dealing with real-world data: I) we provide theoretical proof of unlabeled OOD detection failure when there exists zero mutual information between the learning objective and the in-distribution labels, a.k.a. ’label blindness’, II) we define a new OOD task - Adjacent OOD detection - that tests for label blindness and accounts for a previously ignored safety gap in all OOD detection benchmarks, and III) we perform experiments demonstrating that existing unlabeled OOD methods fail under conditions suggested by our label blindness theory and analyze the implications for future research in unlabeled OOD methods.

Method: Two-pronged strategy: (1) server-side adaptive aggregation that dynamically adjusts client contributions, and (2) client-side regularization to enhance individual privacy robustness.

Result: Achieved 57.14% improvement in group fairness for privacy risk on CIFAR-10 compared to state-of-the-art, with minimal impact on utility. Effectively mitigated source inference attack risks.

Conclusion: FinP successfully establishes fairness in privacy within FL systems without compromising utility, demonstrating significant improvements in privacy risk fairness.

Abstract: Ensuring fairness in machine learning extends to the critical dimension of privacy, particularly in human-centric federated learning (FL) settings where decentralized data necessitates an equitable distribution of privacy risk across clients. This paper introduces FinP, a novel framework specifically designed to address disparities in privacy risk by mitigating disproportionate vulnerability to source inference attacks (SIA). FinP employs a two-pronged strategy: (1) server-side adaptive aggregation, which dynamically adjusts client contributions to the global model to foster fairness, and (2) client-side regularization, which enhances the privacy robustness of individual clients. This comprehensive approach directly tackles both the symptoms and underlying causes of privacy unfairness in FL. Extensive evaluations on the Human Activity Recognition (HAR) and CIFAR-10 datasets demonstrate FinP’s effectiveness, achieving improvement in fairness-in-privacy on HAR and CIFAR-10 with minimal impact on utility. FinP improved group fairness with respect to disparity in privacy risk using equal opportunity in CIFAR-10 by 57.14% compared to the state-of-the-art. Furthermore, FinP significantly mitigates SIA risks on CIFAR-10, underscoring its potential to establish fairness in privacy within FL systems without compromising utility.

Method: Introduced hyperschedules for token-specific noise schedules, hybrid token-wise noising processes (absorbing-uniform interpolation), and novel inference algorithms with KV-caching compatible attention masks.

Result: Achieved state-of-the-art perplexity and generated diverse, high-quality sequences across standard benchmarks.

Conclusion: The approach presents a promising path for autoregressive diffusion-based sequence generation, effectively combining strengths of both paradigms.

Abstract: We present significant extensions to diffusion-based sequence generation models, blurring the line with autoregressive language models. We introduce hyperschedules, which assign distinct noise schedules to individual token positions, generalizing both autoregressive models (e.g., GPT) and conventional diffusion models (e.g., SEDD, MDLM) as special cases. Second, we propose two hybrid token-wise noising processes that interpolate between absorbing and uniform processes, enabling the model to fix past mistakes, and we introduce a novel inference algorithm that leverages this new feature in a simplified context inspired from MDLM. To support efficient training and inference, we design attention masks compatible with KV-caching. Our methods achieve state-of-the-art perplexity and generate diverse, high-quality sequences across standard benchmarks, suggesting a promising path for autoregressive diffusion-based sequence generation. See code and resources at https://hdlm-colm.github.io/

Method: Proposes a framework based on optimal policy in KL-controlled reinforcement learning to incorporate reward and risk/similarity signals into MBRD, enabling sample-efficient variant that adjusts generation count based on problem difficulty.

Result: Empirical demonstration on math (MATH-500) and coding (HumanEval) tasks shows advantages including higher robustness, improved accuracy, and well-understood asymptotic behavior compared to traditional methods.

Conclusion: The proposed framework provides a simple, well-defined mechanism for leveraging reward and similarity signals in MBRD, offering superior performance and sample efficiency while enabling comprehensive accuracy-compute trade-off analysis.

Abstract: Inference scaling helps LLMs solve complex reasoning problems through extended runtime computation. On top of long chain-of-thought (long-CoT) models, purely inference-time techniques such as best-of-N (BoN) sampling, majority voting, or more generally, minimum Bayes risk decoding (MBRD), can further improve LLM accuracy by generating multiple candidate solutions and aggregating over them. These methods typically leverage additional signals in the form of reward models and risk/similarity functions that compare generated samples, e.g., exact match in some normalized space or standard similarity metrics such as Rouge. Here we present a novel method for incorporating reward and risk/similarity signals into MBRD. Based on the concept of optimal policy in KL-controlled reinforcement learning, our framework provides a simple and well-defined mechanism for leveraging such signals, offering several advantages over traditional inference-time methods: higher robustness, improved accuracy, and well-understood asymptotic behavior. In addition, it allows for the development of a sample-efficient variant of MBRD that can adjust the number of samples to generate according to the difficulty of the problem, without relying on majority vote counts. We empirically demonstrate the advantages of our approach on math (MATH-$500$) and coding (HumanEval) tasks using recent open-source models. We also present a comprehensive analysis of its accuracy-compute trade-offs.

Method: Conducted comprehensive empirical study of mathematical reasoning benchmarks, analyzed sensitivity to implementation choices, and proposed standardized evaluation framework with best practices and reporting standards.

Result: Found that performance gains in recent studies often depend on unclear comparisons. RL approaches yield only modest improvements (below prior claims) and are prone to overfitting, while SFT methods show stronger generalization.

Conclusion: Established more rigorous foundations for reasoning evaluation by releasing code, prompts, and model outputs, and demonstrated the need for standardized practices to ensure reliable progress in language model reasoning.

Abstract: Reasoning has emerged as the next major frontier for language models (LMs), with rapid advances from both academic and industrial labs. However, this progress often outpaces methodological rigor, with many evaluations relying on benchmarking practices that lack transparency, robustness, or statistical grounding. In this work, we conduct a comprehensive empirical study and find that current mathematical reasoning benchmarks are highly sensitive to subtle implementation choices–including decoding parameters, random seeds, prompt formatting, and even hardware and software configurations. Performance gains reported in recent studies frequently hinge on unclear comparisons or unreported sources of variance. To address these issues, we propose a standardized evaluation framework with clearly defined best practices and reporting standards. Using this framework, we reassess recent methods and find that most reinforcement learning (RL) approaches yield only modest improvements–far below prior claims–and are prone to overfitting, especially on small-scale benchmarks like AIME'24. In contrast, supervised finetuning (SFT) methods show consistently stronger generalization in the settings we study. To foster reproducibility, we release all code, prompts, and model outputs, for reasoning benchmarks, establishing more rigorous foundations for future work.

Method: Comprehensive analysis of both rule-based and model-based verifiers in static evaluation and RL training scenarios using mathematical reasoning as a case study across multiple datasets.

Result: Rule-based verifiers fail to recognize equivalent answers in different formats (high false negatives), while model-based verifiers achieve higher accuracy but are vulnerable to hacking and misclassify patterns as correct after fine-tuning, leading to artificially inflated rewards.

Conclusion: Both rule-based and model-based verifiers face unique challenges - rule-based ones struggle with answer equivalence, while model-based ones are susceptible to exploitation during optimization, highlighting the need for more accurate and robust reward systems in RL.

Abstract: Trustworthy verifiers are essential for the success of reinforcement learning with verifiable reward (RLVR), which is the core methodology behind various large reasoning models such as DeepSeek-R1. In complex domains like mathematical reasoning, rule-based verifiers have been widely adopted in previous works to train strong reasoning models. However, the reliability of these verifiers and their impact on the RL training process remain poorly understood. In this work, we take mathematical reasoning as a case study and conduct a comprehensive analysis of various verifiers in both static evaluation and RL training scenarios. First, we find that current open-source rule-based verifiers often fail to recognize equivalent answers presented in different formats across multiple commonly used mathematical datasets, resulting in non-negligible false negative rates. This limitation adversely affects RL training performance and becomes more pronounced as the policy model gets stronger. Subsequently, we investigate model-based verifiers as a potential solution to address these limitations. While the static evaluation shows that model-based verifiers achieve significantly higher verification accuracy, further analysis and RL results imply that they are highly susceptible to hacking, where they misclassify certain patterns in responses as correct, particularly after fine-tuning. This vulnerability is exploited during policy model optimization, leading to artificially inflated rewards. Our findings underscore the unique challenges inherent to both rule-based and model-based verifiers and provide insights toward developing more accurate and robust reward systems for reinforcement learning.

Method: Combines 19-layer deep convolutional autoencoder (DCAE) with three gradient boosting classifiers (AdaBoost, XGBoost, LightGBM) for end-to-end AF detection without manual feature extraction.

Result: DCAE-LightGBM model achieved 95.20% F1-score, 99.99% sensitivity, and 4-second inference latency, outperforming existing methods.

Conclusion: The hybrid DCAE-boosting system provides reliable automated AF detection suitable for clinical deployment, with DCAE integration significantly enhancing boosting model performance.

Abstract: Atrial fibrillation (AF) is a prevalent cardiac arrhythmia associated with elevated health risks, where timely detection is pivotal for mitigating stroke-related morbidity. This study introduces an innovative hybrid methodology integrating unsupervised deep learning and gradient boosting models to improve AF detection. A 19-layer deep convolutional autoencoder (DCAE) is coupled with three boosting classifiers-AdaBoost, XGBoost, and LightGBM (LGBM)-to harness their complementary advantages while addressing individual limitations. The proposed framework uniquely combines DCAE with gradient boosting, enabling end-to-end AF identification devoid of manual feature extraction. The DCAE-LGBM model attains an F1-score of 95.20%, sensitivity of 99.99%, and inference latency of four seconds, outperforming existing methods and aligning with clinical deployment requirements. The DCAE integration significantly enhances boosting models, positioning this hybrid system as a reliable tool for automated AF detection in clinical settings.

Method: OCB framework integrates concept-based models with pre-trained object-centric foundation models, using strategies for aggregating object-concept encodings.

Result: OCB outperforms traditional CBMs on complex image datasets and enables interpretable decisions for complex visual tasks.

Conclusion: The Object-Centric Concept Bottlenecks framework successfully addresses limitations of traditional CBMs by leveraging object-centric representations, achieving both improved performance and interpretability.

Abstract: Developing high-performing, yet interpretable models remains a critical challenge in modern AI. Concept-based models (CBMs) attempt to address this by extracting human-understandable concepts from a global encoding (e.g., image encoding) and then applying a linear classifier on the resulting concept activations, enabling transparent decision-making. However, their reliance on holistic image encodings limits their expressiveness in object-centric real-world settings and thus hinders their ability to solve complex vision tasks beyond single-label classification. To tackle these challenges, we introduce Object-Centric Concept Bottlenecks (OCB), a framework that combines the strengths of CBMs and pre-trained object-centric foundation models, boosting performance and interpretability. We evaluate OCB on complex image datasets and conduct a comprehensive ablation study to analyze key components of the framework, such as strategies for aggregating object-concept encodings. The results show that OCB outperforms traditional CBMs and allows one to make interpretable decisions for complex visual tasks.

Method: Learn MAD as the minimum number of actions to transition between states, constructing an embedding space where distances correspond to MAD values using self-supervised learning that handles both symmetric and asymmetric approximations.

Result: The approach efficiently learns accurate MAD representations across diverse environments (deterministic/stochastic, discrete/continuous, noisy observations) and significantly outperforms existing state representation methods in quality.

Conclusion: MAD provides a geometrically meaningful metric that enables critical RL tasks like goal-conditioned learning and reward shaping, with the framework successfully learning state representations from pure trajectories.

Abstract: This paper presents a state representation framework for Markov decision processes (MDPs) that can be learned solely from state trajectories, requiring neither reward signals nor the actions executed by the agent. We propose learning the minimum action distance (MAD), defined as the minimum number of actions required to transition between states, as a fundamental metric that captures the underlying structure of an environment. MAD naturally enables critical downstream tasks such as goal-conditioned reinforcement learning and reward shaping by providing a dense, geometrically meaningful measure of progress. Our self-supervised learning approach constructs an embedding space where the distances between embedded state pairs correspond to their MAD, accommodating both symmetric and asymmetric approximations. We evaluate the framework on a comprehensive suite of environments with known MAD values, encompassing both deterministic and stochastic dynamics, as well as discrete and continuous state spaces, and environments with noisy observations. Empirical results demonstrate that the proposed approach not only efficiently learns accurate MAD representations across these diverse settings but also significantly outperforms existing state representation methods in terms of representation quality.

Method: Introduces ProjNCE, a generalization of InfoNCE loss that incorporates projection functions and an adjustment term for negative pairs, enabling flexible projection strategies for class embeddings.

Result: ProjNCE consistently outperforms both supervised contrastive learning and standard cross-entropy training on image and audio datasets.

Conclusion: The work refines supervised contrastive learning from both information-theoretic and projection perspectives, offering broadly applicable improvements for contrastive learning objectives.

Abstract: Self-supervised contrastive learning (SSCL) has emerged as a powerful paradigm for representation learning and has been studied from multiple perspectives, including mutual information and geometric viewpoints. However, supervised contrastive (SupCon) approaches have received comparatively little attention in this context: for instance, while InfoNCE used in SSCL is known to form a lower bound on mutual information (MI), the relationship between SupCon and MI remains unexplored. To address this gap, we introduce ProjNCE, a generalization of the InfoNCE loss that unifies supervised and self-supervised contrastive objectives by incorporating projection functions and an adjustment term for negative pairs. We prove that ProjNCE constitutes a valid MI bound and affords greater flexibility in selecting projection strategies for class embeddings. Building on this flexibility, we further explore the centroid-based class embeddings in SupCon by exploring a variety of projection methods. Extensive experiments on image and audio datasets demonstrate that ProjNCE consistently outperforms both SupCon and standard cross-entropy training. Our work thus refines SupCon along two complementary perspectives–information-theoretic and projection viewpoints–and offers broadly applicable improvements whenever SupCon serves as the foundational contrastive objective.

Method: VCL maximizes the evidence lower bound by interpreting InfoNCE loss as a reconstruction term and adding KL divergence regularization to a uniform prior on the unit hypersphere. It models the approximate posterior as a projected normal distribution.

Result: VCL mitigates dimensional collapse, enhances mutual information with class labels, and matches or outperforms deterministic baselines in classification accuracy while providing meaningful uncertainty estimates.

Conclusion: VCL provides a probabilistic foundation for contrastive learning, serving as a new basis for contrastive approaches with built-in uncertainty quantification.

Abstract: Deterministic embeddings learned by contrastive learning (CL) methods such as SimCLR and SupCon achieve state-of-the-art performance but lack a principled mechanism for uncertainty quantification. We propose Variational Contrastive Learning (VCL), a decoder-free framework that maximizes the evidence lower bound (ELBO) by interpreting the InfoNCE loss as a surrogate reconstruction term and adding a KL divergence regularizer to a uniform prior on the unit hypersphere. We model the approximate posterior $q_\theta(z|x)$ as a projected normal distribution, enabling the sampling of probabilistic embeddings. Our two instantiation–VSimCLR and VSupCon–replace deterministic embeddings with samples from $q_\theta(z|x)$ and incorporate a normalized KL term into the loss. Experiments on multiple benchmarks demonstrate that VCL mitigates dimensional collapse, enhances mutual information with class labels, and matches or outperforms deterministic baselines in classification accuracy, all the while providing meaningful uncertainty estimates through the posterior model. VCL thus equips contrastive learning with a probabilistic foundation, serving as a new basis for contrastive approaches.

Method: Extend emergent misalignment experiments across diverse conditions, apply model diffing using sparse autoencoders to compare internal representations, and test mitigation strategies with minimal benign fine-tuning.

Result: Found emergent misalignment occurs in various fine-tuning scenarios, identified toxic persona features that control misalignment, and demonstrated efficient restoration of alignment with few benign samples.

Conclusion: Emergent misalignment is a general phenomenon with identifiable internal mechanisms, but can be efficiently mitigated, providing insights for AI safety.

Abstract: Understanding how language models generalize behaviors from their training to a broader deployment distribution is an important problem in AI safety. Betley et al. discovered that fine-tuning GPT-4o on intentionally insecure code causes “emergent misalignment,” where models give stereotypically malicious responses to unrelated prompts. We extend this work, demonstrating emergent misalignment across diverse conditions, including reinforcement learning on reasoning models, fine-tuning on various synthetic datasets, and in models without safety training. To investigate the mechanisms behind this generalized misalignment, we apply a “model diffing” approach using sparse autoencoders to compare internal model representations before and after fine-tuning. This approach reveals several “misaligned persona” features in activation space, including a toxic persona feature which most strongly controls emergent misalignment and can be used to predict whether a model will exhibit such behavior. Additionally, we investigate mitigation strategies, discovering that fine-tuning an emergently misaligned model on just a few hundred benign samples efficiently restores alignment.

Method: Three sequential blocks: (1) learning shared patterns, (2) partitioning tasks into outliers/non-outliers with regularized parameters, (3) fine-tuning task-specific predictors within gradient boosting framework.

Result: Successfully isolates outliers, transfers knowledge among related tasks, reduces prediction errors for each task individually, and achieves overall performance gains across all tasks.

Conclusion: R-MTGB demonstrates robustness, adaptability, and reliable convergence in challenging multi-task learning environments with outlier tasks.

Abstract: Multi-task learning (MTL) has shown effectiveness in exploiting shared information across tasks to improve generalization. MTL assumes tasks share similarities that can improve performance. In addition, boosting algorithms have demonstrated exceptional performance across diverse learning problems, primarily due to their ability to focus on hard-to-learn instances and iteratively reduce residual errors. This makes them a promising approach for learning multi-task problems. However, real-world MTL scenarios often involve tasks that are not well-aligned (known as outlier or adversarial tasks), which do not share beneficial similarities with others and can, in fact, deteriorate the performance of the overall model. To overcome this challenge, we propose Robust-Multi-Task Gradient Boosting (R-MTGB), a novel boosting framework that explicitly models and adapts to task heterogeneity during training. R-MTGB structures the learning process into three sequential blocks: (1) learning shared patterns, (2) partitioning tasks into outliers and non-outliers with regularized parameters, and (3) fine-tuning task-specific predictors. This architecture enables R-MTGB to automatically detect and penalize outlier tasks while promoting effective knowledge transfer among related tasks. Our method integrates these mechanisms seamlessly within gradient boosting, allowing robust handling of noisy or adversarial tasks without sacrificing accuracy. Extensive experiments on both synthetic benchmarks and real-world datasets demonstrate that our approach successfully isolates outliers, transfers knowledge, and consistently reduces prediction errors for each task individually, and achieves overall performance gains across all tasks. These results highlight robustness, adaptability, and reliable convergence of R-MTGB in challenging MTL environments.

Method: CAPO uses an off-the-shelf LLM as a Generative Process Reward Model (LLM-as-GenPRM) to generate step-wise critiques in one pass, providing deterministic token-level credits. It employs voting mechanisms to enhance accuracy and robustness.

Result: CAPO consistently outperforms supervised learning and RL-based fine-tuning methods across four mathematical benchmarks and three out-of-domain benchmarks. It helps models learn correct reasoning pathways leading to correct answers.

Conclusion: CAPO provides an efficient and effective solution for fine-grained credit assignment in RL, enabling better reasoning capabilities in LLMs without requiring expensive process supervision or suffering from unreliable probabilistic rewards.

Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has improved the reasoning abilities of Large Language Models (LLMs) by using rule-based binary feedback. However, current RLVR methods typically assign the same reward to every token. This coarse-grained feedback hampers precise credit assignment, making it hard for models to identify which reasoning steps lead to success or failure, and often results in suboptimal policies. Methods like PPO provide credit assignment by value estimation, but yield inaccurate and unverifiable signals due to limited sampling. On the other hand, methods using Process Reward Models can provide step-wise rewards but suffer from several key limitations: they require high-quality process supervision labels, the feedback is unreliable due to probabilistic reward modeling, and their application in online reinforcement learning (RL) is time-consuming. To overcome these limitations, we introduce a simple but efficient method-Credit Assignment Policy Optimization (CAPO). Instead of training auxiliary models, CAPO directly leverages an off-the-shelf, general-purpose LLM as a Generative Process Reward Model (LLM-as-GenPRM) to generate all step-wise critique by one pass only based on the correctness of the step itself, providing deterministic token-level credits to refine the tokens that were originally assigned identical rule-based rewards. To further enhance the accuracy and robustness, we employ voting mechanisms that scale with the number of generated critiques. Extensive experiments on various backbones like Llama and Qwen models show that CAPO consistently outperforms supervised learning-based and RL-based fine-tuning methods across four challenging mathematical benchmarks and three out-of-domain benchmarks. Further analysis shows that CAPO can help the model to foster the learning of correct reasoning pathways leading to correct answers.

Method: Pref-GUIDE Individual compares agent behaviors within short windows and filters ambiguous feedback. Pref-GUIDE Voting aggregates reward models across users to form consensus preferences.

Result: Across three challenging environments, Pref-GUIDE significantly outperforms scalar-feedback baselines, with the voting variant exceeding expert-designed dense rewards.

Conclusion: By reframing scalar feedback as structured preferences with population feedback, Pref-GUIDE offers a scalable and principled approach for harnessing human input in online reinforcement learning.

Abstract: Training reinforcement learning agents with human feedback is crucial when task objectives are difficult to specify through dense reward functions. While prior methods rely on offline trajectory comparisons to elicit human preferences, such data is unavailable in online learning scenarios where agents must adapt on the fly. Recent approaches address this by collecting real-time scalar feedback to guide agent behavior and train reward models for continued learning after human feedback becomes unavailable. However, scalar feedback is often noisy and inconsistent, limiting the accuracy and generalization of learned rewards. We propose Pref-GUIDE, a framework that transforms real-time scalar feedback into preference-based data to improve reward model learning for continual policy training. Pref-GUIDE Individual mitigates temporal inconsistency by comparing agent behaviors within short windows and filtering ambiguous feedback. Pref-GUIDE Voting further enhances robustness by aggregating reward models across a population of users to form consensus preferences. Across three challenging environments, Pref-GUIDE significantly outperforms scalar-feedback baselines, with the voting variant exceeding even expert-designed dense rewards. By reframing scalar feedback as structured preferences with population feedback, Pref-GUIDE offers a scalable and principled approach for harnessing human input in online reinforcement learning.

Method: Novel data-driven scheduling heuristic that combines machine learning with problem-specific characteristics to ensure feasible solutions, with systematic ML model exploration and detailed model selection process.

Result: Significantly outperforms state-of-the-art in terms of optimality gap, number of optimal solutions, and adaptability across varied data scenarios, demonstrating flexibility for practical applications.

Conclusion: The approach provides strong performance and adaptability, with systematic model selection addressing a common gap in similar studies and offering insights into optimal model choices.

Abstract: Existing research on single-machine scheduling is largely focused on exact algorithms, which perform well on typical instances but can significantly deteriorate on certain regions of the problem space. In contrast, data-driven approaches provide strong and scalable performance when tailored to the structure of specific datasets. Leveraging this idea, we focus on a single-machine scheduling problem where each job is defined by its weight, duration, due date, and deadline, aiming to minimize the total weight of tardy jobs. We introduce a novel data-driven scheduling heuristic that combines machine learning with problem-specific characteristics, ensuring feasible solutions, which is a common challenge for ML-based algorithms. Experimental results demonstrate that our approach significantly outperforms the state-of-the-art in terms of optimality gap, number of optimal solutions, and adaptability across varied data scenarios, highlighting its flexibility for practical applications. In addition, we conduct a systematic exploration of ML models, addressing a common gap in similar studies by offering a detailed model selection process and providing insights into why the chosen model is the best fit.

Method: Proposed a novel taxonomy of harms and created BAD-ACTS benchmark with 4 agentic system implementations across different environments and 188 harmful action examples. Analyzed robustness against attacks where one adversarial agent manipulates others.

Result: Attack has high success rate, demonstrating even a single adversarial agent can significantly impact system security. Simple prompting-based defenses remain ineffective, but proposed message monitoring defense shows better effectiveness.

Conclusion: BAD-ACTS provides a diverse testbed for security research of agentic systems, highlighting vulnerabilities and the need for robust defense mechanisms.

Abstract: Ensuring the safe use of agentic systems requires a thorough understanding of the range of malicious behaviors these systems may exhibit when under attack. In this paper, we evaluate the robustness of LLM-based agentic systems against attacks that aim to elicit harmful actions from agents. To this end, we propose a novel taxonomy of harms for agentic systems and a novel benchmark, BAD-ACTS, for studying the security of agentic systems with respect to a wide range of harmful actions. BAD-ACTS consists of 4 implementations of agentic systems in distinct application environments, as well as a dataset of 188 high-quality examples of harmful actions. This enables a comprehensive study of the robustness of agentic systems across a wide range of categories of harmful behaviors, available tools, and inter-agent communication structures. Using this benchmark, we analyze the robustness of agentic systems against an attacker that controls one of the agents in the system and aims to manipulate other agents to execute a harmful target action. Our results show that the attack has a high success rate, demonstrating that even a single adversarial agent within the system can have a significant impact on the security. This attack remains effective even when agents use a simple prompting-based defense strategy. However, we additionally propose a more effective defense based on message monitoring. We believe that this benchmark provides a diverse testbed for the security research of agentic systems. The benchmark can be found at github.com/JNoether/BAD-ACTS

Method: Analyzed 6 state-of-the-art text encoders across 26 downstream tasks, randomly removing embedding dimensions, and studied the phenomenon by examining representation space usage and dimension distribution.

Result: Removing up to 50% of embedding dimensions resulted in less than 10% performance drop in retrieval and classification tasks. Found that many uniformly distributed dimensions actually improve performance when removed, explaining the minor performance impact.

Conclusion: The phenomenon of minimal performance degradation when truncating embeddings is not due to ineffective representation space usage, but rather the presence of many uniformly distributed dimensions that hinder performance. This effect extends to generative tasks using large language models.

Abstract: In this paper, we study the surprising impact that truncating text embeddings has on downstream performance. We consistently observe across 6 state-of-the-art text encoders and 26 downstream tasks, that randomly removing up to 50% of embedding dimensions results in only a minor drop in performance, less than 10%, in retrieval and classification tasks. Given the benefits of using smaller-sized embeddings, as well as the potential insights about text encoding, we study this phenomenon and find that, contrary to what is suggested in prior work, this is not the result of an ineffective use of representation space. Instead, we find that a large number of uniformly distributed dimensions actually cause an increase in performance when removed. This would explain why, on average, removing a large number of embedding dimensions results in a marginal drop in performance. We make similar observations when truncating the embeddings used by large language models to make next-token predictions on generative tasks, suggesting that this phenomenon is not isolated to classification or retrieval tasks.

Method: 1) Selective modeling using single Gaussian and GMMs guided by cross-validation; 2) Novel BPA transformation from possibility distributions; 3) Enhanced EKNN classifier with belief structure-based BPA generation.

Result: Proposed classifier outperforms best existing evidential classifier with 4.86% average accuracy improvement while maintaining low variance.

Conclusion: The enhanced attribute fusion-based classifier demonstrates superior effectiveness and robustness in handling uncertainty for multi-attribute classification tasks.

Abstract: Dempster-Shafer Theory (DST) provides a powerful framework for modeling uncertainty and has been widely applied to multi-attribute classification tasks. However, traditional DST-based attribute fusion-based classifiers suffer from oversimplified membership function modeling and limited exploitation of the belief structure brought by basic probability assignment (BPA), reducing their effectiveness in complex real-world scenarios. This paper presents an enhanced attribute fusion-based classifier that addresses these limitations through two key innovations. First, we adopt a selective modeling strategy that utilizes both single Gaussian and Gaussian Mixture Models (GMMs) for membership function construction, with model selection guided by cross-validation and a tailored evaluation metric. Second, we introduce a novel method to transform the possibility distribution into a BPA by combining simple BPAs derived from normalized possibility distributions, enabling a much richer and more flexible representation of uncertain information. Furthermore, we apply the belief structure-based BPA generation method to the evidential K-Nearest Neighbors (EKNN) classifier, enhancing its ability to incorporate uncertainty information into decision-making. Comprehensive experiments on benchmark datasets are conducted to evaluate the performance of the proposed attribute fusion-based classifier and the enhanced evidential K-Nearest Neighbors classifier in comparison with both evidential classifiers and conventional machine learning classifiers. The results demonstrate that the proposed classifier outperforms the best existing evidential classifier, achieving an average accuracy improvement of 4.86%, while maintaining low variance, thus confirming its superior effectiveness and robustness.

Method: Developed DynaGuard suite with dynamic guardian models and DynaBench dataset, offering both rapid detection and chain-of-thought reasoning for policy violations.

Result: DynaGuard surpasses static models in detection accuracy on traditional safety categories and is competitive with frontier reasoning models on free-form policy violations while being much faster.

Conclusion: DynaGuard provides a critical tool for language model guardrails with its dynamic policy evaluation capabilities and superior performance.

Abstract: Guardian models play a crucial role in ensuring the safety and ethical behavior of user-facing AI applications by enforcing guardrails and detecting harmful content. While standard guardian models are limited to predefined, static harm categories, we introduce DynaGuard, a suite of dynamic guardian models offering novel flexibility by evaluating text based on user-defined policies, and DynaBench, a dataset for training and evaluating dynamic guardian models. Our models provide both rapid detection of policy violations and a chain-of-thought reasoning option that articulate and justify model outputs. Critically, DynaGuard not only surpasses static models in detection accuracy on traditional safety categories, but is competitive with frontier reasoning models on free-form policy violations, all in a fraction of the time. This makes DynaGuard an critical tool for language model guardrails.

Method: Leverages historical experiment outcomes with SHAP explanations and employs LLM-as-judge evaluation to control output format, accuracy, and completeness for zero-shot hyperparameter optimization.

Result: Achieves competitive or superior performance to traditional HPO methods while drastically reducing computational cost, with 99.6-99.9% response time reduction and 2.4-15.7x faster training times on medical imaging datasets.

Conclusion: MetaLLMiX provides an efficient, locally-deployable alternative to API-based approaches that maintains accuracy while significantly reducing computational requirements.

Abstract: Effective model and hyperparameter selection remains a major challenge in deep learning, often requiring extensive expertise and computation. While AutoML and large language models (LLMs) promise automation, current LLM-based approaches rely on trial and error and expensive APIs, which provide limited interpretability and generalizability. We propose MetaLLMiX, a zero-shot hyperparameter optimization framework combining meta-learning, explainable AI, and efficient LLM reasoning. By leveraging historical experiment outcomes with SHAP explanations, MetaLLMiX recommends optimal hyperparameters and pretrained models without additional trials. We further employ an LLM-as-judge evaluation to control output format, accuracy, and completeness. Experiments on eight medical imaging datasets using nine open-source lightweight LLMs show that MetaLLMiX achieves competitive or superior performance to traditional HPO methods while drastically reducing computational cost. Our local deployment outperforms prior API-based approaches, achieving optimal results on 5 of 8 tasks, response time reductions of 99.6-99.9%, and the fastest training times on 6 datasets (2.4-15.7x faster), maintaining accuracy within 1-5% of best-performing baselines.

Method: Developed a graph representation of mixed bundling model, trained GCN to learn optimal bundle patterns, proposed two inference strategies, and used local-search to improve solution quality.

Result: Achieved near-optimal solutions (better than 97%) with fraction of computational time for small-medium problems, superior solutions for larger problems compared to BSP, and handled up to 30+ products even with non-additive utilities.

Conclusion: The GCN-based framework is effective and efficient for bundle pricing problems, providing high-quality solutions across various problem sizes and challenging scenarios.

Abstract: Bundle pricing refers to designing several product combinations (i.e., bundles) and determining their prices in order to maximize the expected profit. It is a classic problem in revenue management and arises in many industries, such as e-commerce, tourism, and video games. However, the problem is typically intractable due to the exponential number of candidate bundles. In this paper, we explore the usage of graph convolutional networks (GCNs) in solving the bundle pricing problem. Specifically, we first develop a graph representation of the mixed bundling model (where every possible bundle is assigned with a specific price) and then train a GCN to learn the latent patterns of optimal bundles. Based on the trained GCN, we propose two inference strategies to derive high-quality feasible solutions. A local-search technique is further proposed to improve the solution quality. Numerical experiments validate the effectiveness and efficiency of our proposed GCN-based framework. Using a GCN trained on instances with 5 products, our methods consistently achieve near-optimal solutions (better than 97%) with only a fraction of computational time for problems of small to medium size. It also achieves superior solutions for larger size of problems compared with other heuristic methods such as bundle size pricing (BSP). The method can also provide high quality solutions for instances with more than 30 products even for the challenging cases where product utilities are non-additive.

Method: Extends Direct Preference Optimization (DPO) by incorporating adaptive margins derived from preference-over-preference supervision. Uses ordinal signals about which of two preferences reflects a stronger distinction to infer per-datapoint margins. Proposes two sampling strategies for gathering preference-over-preference labels.

Result: DPO-PoP outperforms vanilla DPO, DPO with fixed margins, and DPO with ground-truth margins on the UltraFeedback dataset. Shows a tradeoff between discriminative and generative performance - improving test classification accuracy (especially for weaker preferences) can reduce generative quality.

Conclusion: Modeling preference strength through adaptive margins improves alignment and generalization. The proposed DPO-PoP method effectively leverages preference-over-preference annotations to achieve better performance while navigating the tradeoff between discriminative and generative capabilities.

Abstract: Margin-based optimization is fundamental to improving generalization and robustness in classification tasks. In the context of reward model learning from preferences within Reinforcement Learning from Human Feedback (RLHF), existing methods typically rely on no margins, fixed margins, or margins that are simplistic functions of preference ratings. However, such formulations often fail to account for the varying strengths of different preferences, for example some preferences are associated with larger margins between responses, or they rely on noisy margin information derived from ratings. We argue that modeling the strength of preferences can lead to better generalization and more faithful alignment. Furthermore, many existing methods that use adaptive margins assume access to accurate preference scores, which can be difficult for humans to provide reliably. We propose an approach that leverages preferences over preferences, that is annotations indicating which of two preferences reflects a stronger distinction. We use this ordinal signal to infer adaptive margins on a per-datapoint basis. We introduce an extension to Direct Preference Optimization (DPO), DPO-PoP, that incorporates adaptive margins from preference-over-preference supervision, enabling improved discriminative and generative performance. Empirically, our method outperforms vanilla DPO, DPO with fixed margins, and DPO with ground-truth margins on the UltraFeedback dataset. Additionally, we show that there is a tradeoff between discriminative and generative performance: improving test classification accuracy, particularly by correctly labeling weaker preferences at the expense of stronger ones, can lead to a decline in generative quality. To navigate this tradeoff, we propose two sampling strategies to gather preference-over-preference labels: one favoring discriminative performance and one favoring generative performance.

Method: A two-stage training strategy: first using maximum likelihood estimation to learn latent representations from both partial-task aware (PTA) and all-task aware (ATA) setups, then optimizing KL divergence to identify and learn shared latent variables.

Result: Theoretical guarantee and empirical validations show that identifying and learning shared representations effectively mitigates catastrophic interference, with practical performance improvements across synthetic and benchmark datasets.

Conclusion: Formulating catastrophic interference as an identification problem and learning shared latent variables through the proposed framework provides both theoretical guarantees and practical solutions to mitigate forgetting in machine learning systems.

Abstract: Catastrophic interference, also known as catastrophic forgetting, is a fundamental challenge in machine learning, where a trained learning model progressively loses performance on previously learned tasks when adapting to new ones. In this paper, we aim to better understand and model the catastrophic interference problem from a latent representation learning point of view, and propose a novel theoretical framework that formulates catastrophic interference as an identification problem. Our analysis demonstrates that the forgetting phenomenon can be quantified by the distance between partial-task aware (PTA) and all-task aware (ATA) setups. Building upon recent advances in identifiability theory, we prove that this distance can be minimized through identification of shared latent variables between these setups. When learning, we propose our method \ourmeos with two-stage training strategy: First, we employ maximum likelihood estimation to learn the latent representations from both PTA and ATA configurations. Subsequently, we optimize the KL divergence to identify and learn the shared latent variables. Through theoretical guarantee and empirical validations, we establish that identifying and learning these shared representations can effectively mitigate catastrophic interference in machine learning systems. Our approach provides both theoretical guarantees and practical performance improvements across both synthetic and benchmark datasets.

Method: Analyzed zero-shot capabilities and fine-tuning performance of time series foundation models on forecasting tasks, comparing them with smaller dedicated models.

Result: Foundation models’ zero-shot performance is domain-dependent, and fine-tuned versions don’t consistently provide substantial improvements over smaller specialized models relative to their computational costs.

Conclusion: Time series data’s diversity makes foundation models less effective than in language/vision domains, with dedicated smaller models often being more practical for forecasting tasks.

Abstract: Foundation Models are designed to serve as versatile embedding machines, with strong zero shot capabilities and superior generalization performance when fine-tuned on diverse downstream tasks. While this is largely true for language and vision foundation models, we argue that the inherent diversity of time series data makes them less suited for building effective foundation models. We demonstrate this using forecasting as our downstream task. We show that the zero-shot capabilities of a time series foundation model are significantly influenced and tied to the specific domains it has been pretrained on. Furthermore, when applied to unseen real-world time series data, fine-tuned foundation models do not consistently yield substantially better results, relative to their increased parameter count and memory footprint, than smaller, dedicated models tailored to the specific forecasting task at hand.

Method: SoftAdaClip introduces a differentially private training method that replaces hard clipping with a smooth, tanh-based transformation to preserve relative gradient magnitudes while bounding sensitivity.

Result: SoftAdaClip reduces subgroup disparities by up to 87% compared to DP-SGD and up to 48% compared to Adaptive-DPSGD, with statistically significant reductions across datasets including MIMIC-III, GOSSIS-eICU, and Adult Income.

Conclusion: Integrating smooth transformations with adaptive mechanisms is crucial for achieving fair and private model training, as demonstrated by SoftAdaClip’s ability to significantly reduce subgroup disparities while maintaining differential privacy.

Abstract: Differential privacy (DP) provides strong protection for sensitive data, but often reduces model performance and fairness, especially for underrepresented groups. One major reason is gradient clipping in DP-SGD, which can disproportionately suppress learning signals for minority subpopulations. Although adaptive clipping can enhance utility, it still relies on uniform hard clipping, which may restrict fairness. To address this, we introduce SoftAdaClip, a differentially private training method that replaces hard clipping with a smooth, tanh-based transformation to preserve relative gradient magnitudes while bounding sensitivity. We evaluate SoftAdaClip on various datasets, including MIMIC-III (clinical text), GOSSIS-eICU (structured healthcare), and Adult Income (tabular data). Our results show that SoftAdaClip reduces subgroup disparities by up to 87% compared to DP-SGD and up to 48% compared to Adaptive-DPSGD, and these reductions in subgroup disparities are statistically significant. These findings underscore the importance of integrating smooth transformations with adaptive mechanisms to achieve fair and private model training.

Method: RainSeer uses a physics-informed two-stage architecture: 1) Structure-to-Point Mapper for spatial alignment by projecting radar structures to ground-level rainfall, and 2) Geo-Aware Rain Decoder that captures semantic transformation of hydro-meteors through descent, melting, and evaporation using causal spatiotemporal attention.

Result: RainSeer achieved consistent improvements on RAIN-F (Korea) and MeteoNet (France) datasets, reducing MAE by over 13.31% and significantly enhancing structural fidelity in reconstructed rainfall fields compared to state-of-the-art baselines.

Conclusion: RainSeer successfully addresses the challenges of translating volumetric radar fields to point-wise rainfall observations and bridging the physical disconnect between aloft hydro-meteors and ground-level precipitation, providing more accurate rainfall reconstruction.

Abstract: Reconstructing high-resolution rainfall fields is essential for flood forecasting, hydrological modeling, and climate analysis. However, existing spatial interpolation methods-whether based on automatic weather station (AWS) measurements or enhanced with satellite/radar observations often over-smooth critical structures, failing to capture sharp transitions and localized extremes. We introduce RainSeer, a structure-aware reconstruction framework that reinterprets radar reflectivity as a physically grounded structural prior-capturing when, where, and how rain develops. This shift, however, introduces two fundamental challenges: (i) translating high-resolution volumetric radar fields into sparse point-wise rainfall observations, and (ii) bridging the physical disconnect between aloft hydro-meteors and ground-level precipitation. RainSeer addresses these through a physics-informed two-stage architecture: a Structure-to-Point Mapper performs spatial alignment by projecting mesoscale radar structures into localized ground-level rainfall, through a bidirectional mapping, and a Geo-Aware Rain Decoder captures the semantic transformation of hydro-meteors through descent, melting, and evaporation via a causal spatiotemporal attention mechanism. We evaluate RainSeer on two public datasets-RAIN-F (Korea, 2017-2019) and MeteoNet (France, 2016-2018)-and observe consistent improvements over state-of-the-art baselines, reducing MAE by over 13.31% and significantly enhancing structural fidelity in reconstructed rainfall fields.

Method: Exploits resource asymmetry between prompt guards and main LLMs by encoding jailbreak prompts that lightweight guards cannot decode but main models can interpret.

Result: Consistently jailbreaks production models (Google Gemini, DeepSeek Chat, Grok, Mistral Le Chat) while maintaining response quality, even under highly protected chat interfaces.

Conclusion: Reveals inherent attack surface in lightweight prompt guards and underscores need to shift defenses from blocking malicious inputs to preventing malicious outputs.

Abstract: As large language models (LLMs) advance, ensuring AI safety and alignment is paramount. One popular approach is prompt guards, lightweight mechanisms designed to filter malicious queries while being easy to implement and update. In this work, we introduce a new attack that circumvents such prompt guards, highlighting their limitations. Our method consistently jailbreaks production models while maintaining response quality, even under the highly protected chat interfaces of Google Gemini (2.5 Flash/Pro), DeepSeek Chat (DeepThink), Grok (3), and Mistral Le Chat (Magistral). The attack exploits a resource asymmetry between the prompt guard and the main LLM, encoding a jailbreak prompt that lightweight guards cannot decode but the main model can. This reveals an attack surface inherent to lightweight prompt guards in modern LLM architectures and underscores the need to shift defenses from blocking malicious inputs to preventing malicious outputs. We additionally identify other critical alignment issues, such as copyrighted data extraction, training data extraction, and malicious response leakage during thinking.

Method: Adapted Score identity Distillation (SiD) with multistep generation and inference time noise modulation to train few-step protein backbone generators that reduce sampling time while maintaining performance.

Result: The distilled few-step generators achieve more than 20-fold improvement in sampling speed while maintaining similar levels of designability, diversity, and novelty as the Proteina teacher model.

Conclusion: This reduction in inference cost enables large-scale in silico protein design, bringing diffusion-based models closer to real-world protein engineering applications.

Abstract: Diffusion- and flow-based generative models have recently demonstrated strong performance in protein backbone generation tasks, offering unprecedented capabilities for de novo protein design. However, while achieving notable performance in generation quality, these models are limited by their generating speed, often requiring hundreds of iterative steps in the reverse-diffusion process. This computational bottleneck limits their practical utility in large-scale protein discovery, where thousands to millions of candidate structures are needed. To address this challenge, we explore the techniques of score distillation, which has shown great success in reducing the number of sampling steps in the vision domain while maintaining high generation quality. However, a straightforward adaptation of these methods results in unacceptably low designability. Through extensive study, we have identified how to appropriately adapt Score identity Distillation (SiD), a state-of-the-art score distillation strategy, to train few-step protein backbone generators which significantly reduce sampling time, while maintaining comparable performance to their pretrained teacher model. In particular, multistep generation combined with inference time noise modulation is key to the success. We demonstrate that our distilled few-step generators achieve more than a 20-fold improvement in sampling speed, while achieving similar levels of designability, diversity, and novelty as the Proteina teacher model. This reduction in inference cost enables large-scale in silico protein design, thereby bringing diffusion-based models closer to real-world protein engineering applications.

Method: Four-agent framework: Curator performs LLM-guided diagnostics and preprocessing; Planner narrows model choices using multi-modal diagnostics; Forecaster handles model fitting, validation, and adaptive ensemble selection; Reporter synthesizes the process into transparent reports.

Result: TSci outperforms both statistical and LLM-based baselines on eight benchmarks, reducing forecast error by 10.4% and 38.2% respectively, while producing transparent and interpretable reports.

Conclusion: TSci successfully automates time series forecasting workflows, providing a general, domain-agnostic solution that reduces human intervention while improving accuracy and interpretability through its multi-agent framework.

Abstract: Time series forecasting is central to decision-making in domains as diverse as energy, finance, climate, and public health. In practice, forecasters face thousands of short, noisy series that vary in frequency, quality, and horizon, where the dominant cost lies not in model fitting, but in the labor-intensive preprocessing, validation, and ensembling required to obtain reliable predictions. Prevailing statistical and deep learning models are tailored to specific datasets or domains and generalize poorly. A general, domain-agnostic framework that minimizes human intervention is urgently in demand. In this paper, we introduce TimeSeriesScientist (TSci), the first LLM-driven agentic framework for general time series forecasting. The framework comprises four specialized agents: Curator performs LLM-guided diagnostics augmented by external tools that reason over data statistics to choose targeted preprocessing; Planner narrows the hypothesis space of model choice by leveraging multi-modal diagnostics and self-planning over the input; Forecaster performs model fitting and validation and, based on the results, adaptively selects the best model configuration as well as ensemble strategy to make final predictions; and Reporter synthesizes the whole process into a comprehensive, transparent report. With transparent natural-language rationales and comprehensive reports, TSci transforms the forecasting workflow into a white-box system that is both interpretable and extensible across tasks. Empirical results on eight established benchmarks demonstrate that TSci consistently outperforms both statistical and LLM-based baselines, reducing forecast error by an average of 10.4% and 38.2%, respectively. Moreover, TSci produces a clear and rigorous report that makes the forecasting workflow more transparent and interpretable.

Method: Developed OligoICP method that uses TabPFN’s predicted inter-quantile range (IQR) as an uncertainty measure to select and average ensemble models with the lowest mean IQR for siRNA efficacy prediction.

Result: TabPFN with sequence-based features surpassed specialized state-of-the-art predictors. Model’s IQR showed negative correlation with true prediction error. OligoICP achieved superior performance compared to naive ensembling or single models trained on all data.

Conclusion: Model uncertainty serves as a powerful, label-free heuristic for optimizing biomolecule efficacy predictions, enabling effective model selection without ground truth labels.

Abstract: In-context learners like TabPFN are promising for biomolecule efficacy prediction, where established molecular feature sets and relevant experimental results can serve as powerful contextual examples. However, their performance is highly sensitive to the provided context, making strategies like post-hoc ensembling of models trained on different data subsets a viable approach. An open question is how to select the best models for the ensemble without access to ground truth labels. In this study, we investigate an uncertainty-guided strategy for model selection. We demonstrate on an siRNA knockdown efficacy task that a TabPFN model using straightforward sequence-based features can surpass specialized state-of-the-art predictors. We also show that the model’s predicted inter-quantile range (IQR), a measure of its uncertainty, has a negative correlation with true prediction error. We developed the OligoICP method, which selects and averages an ensemble of models with the lowest mean IQR for siRNA efficacy prediction, achieving superior performance compared to naive ensembling or using a single model trained on all available data. This finding highlights model uncertainty as a powerful, label-free heuristic for optimizing biomolecule efficacy predictions.

Method: Introduce a novel algorithm for detecting stable and unstable manifolds in PLRNNs using ReLU activation functions. The algorithm traces boundaries between basins of attraction and finds homoclinic points (intersections between stable/unstable manifolds).

Result: Demonstrated ability to characterize multistability by tracing basin boundaries, established existence of chaos in PLRNNs by finding homoclinic points, and applied to electrophysiological recordings from cortical neurons to gain insights into underlying dynamics.

Conclusion: The proposed algorithm provides a powerful tool for understanding RNN dynamics, enabling characterization of multistability and chaos detection, with practical applications in analyzing neural data and advancing explainable AI.

Abstract: Recurrent Neural Networks (RNNs) have found widespread applications in machine learning for time series prediction and dynamical systems reconstruction, and experienced a recent renaissance with improved training algorithms and architectural designs. Understanding why and how trained RNNs produce their behavior is important for scientific and medical applications, and explainable AI more generally. An RNN’s dynamical repertoire depends on the topological and geometrical properties of its state space. Stable and unstable manifolds of periodic points play a particularly important role: They dissect a dynamical system’s state space into different basins of attraction, and their intersections lead to chaotic dynamics with fractal geometry. Here we introduce a novel algorithm for detecting these manifolds, with a focus on piecewise-linear RNNs (PLRNNs) employing rectified linear units (ReLUs) as their activation function. We demonstrate how the algorithm can be used to trace the boundaries between different basins of attraction, and hence to characterize multistability, a computationally important property. We further show its utility in finding so-called homoclinic points, the intersections between stable and unstable manifolds, and thus establish the existence of chaos in PLRNNs. Finally we show for an empirical example, electrophysiological recordings from a cortical neuron, how insights into the underlying dynamics could be gained through our method.

Method: Uses a VAE to encode text reasoning steps into structured latent thought tokens, then applies a latent diffusion model with blockwise bidirectional attention to denoise and iteratively refine reasoning trajectories.

Result: LaDiR consistently improves accuracy, diversity, and interpretability over existing autoregressive, diffusion-based, and latent reasoning methods across mathematical reasoning and planning benchmarks.

Conclusion: LaDiR reveals a new paradigm for text reasoning with latent diffusion, enabling efficient parallel generation of diverse reasoning trajectories with adaptive test-time compute.

Abstract: Large Language Models (LLMs) demonstrate their reasoning ability through chain-of-thought (CoT) generation. However, LLM’s autoregressive decoding may limit the ability to revisit and refine earlier tokens in a holistic manner, which can also lead to inefficient exploration for diverse solutions. In this paper, we propose LaDiR (Latent Diffusion Reasoner), a novel reasoning framework that unifies the expressiveness of continuous latent representation with the iterative refinement capabilities of latent diffusion models for an existing LLM. We first construct a structured latent reasoning space using a Variational Autoencoder (VAE) that encodes text reasoning steps into blocks of thought tokens, preserving semantic information and interpretability while offering compact but expressive representations. Subsequently, we utilize a latent diffusion model that learns to denoise a block of latent thought tokens with a blockwise bidirectional attention mask, enabling longer horizon and iterative refinement with adaptive test-time compute. This design allows efficient parallel generation of diverse reasoning trajectories, allowing the model to plan and revise the reasoning process holistically. We conduct evaluations on a suite of mathematical reasoning and planning benchmarks. Empirical results show that LaDiR consistently improves accuracy, diversity, and interpretability over existing autoregressive, diffusion-based, and latent reasoning methods, revealing a new paradigm for text reasoning with latent diffusion.

Method: DP-HYPE performs distributed voting based on local hyperparameter evaluations from clients to select hyperparameters supported by the majority, while ensuring client-level differential privacy.

Result: The algorithm provides strong privacy guarantees independent of hyperparameter count, achieves high utility even with small privacy budgets, and is implemented in the Flower framework for distributed ML.

Conclusion: DP-HYPE enables efficient, privacy-preserving hyperparameter tuning that scales well and maintains high utility across various data distributions including non-iid settings.

Abstract: The tuning of hyperparameters in distributed machine learning can substantially impact model performance. When the hyperparameters are tuned on sensitive data, privacy becomes an important challenge and to this end, differential privacy has emerged as the de facto standard for provable privacy. A standard setting when performing distributed learning tasks is that clients agree on a shared setup, i.e., find a compromise from a set of hyperparameters, like the learning rate of the model to be trained. Yet, prior work on differentially private hyperparameter tuning either uses computationally expensive cryptographic protocols, determines hyperparameters separately for each client, or applies differential privacy locally, which can lead to undesirable utility-privacy trade-offs. In this work, we present our algorithm DP-HYPE, which performs a distributed and privacy-preserving hyperparameter search by conducting a distributed voting based on local hyperparameter evaluations of clients. In this way, DP-HYPE selects hyperparameters that lead to a compromise supported by the majority of clients, while maintaining scalability and independence from specific learning tasks. We prove that DP-HYPE preserves the strong notion of differential privacy called client-level differential privacy and, importantly, show that its privacy guarantees do not depend on the number of hyperparameters. We also provide bounds on its utility guarantees, that is, the probability of reaching a compromise, and implement DP-HYPE as a submodule in the popular Flower framework for distributed machine learning. In addition, we evaluate performance on multiple benchmark data sets in iid as well as multiple non-iid settings and demonstrate high utility of DP-HYPE even under small privacy budgets.

Method: Information decomposition framework using partial information decomposition of time-delayed mutual information (TDMI), applied to experiments with a guessing game using three randomized interventions: control, persona assignment, and persona plus strategic thinking instruction.

Result: Control groups showed temporal synergy but little coordinated alignment. Persona assignment created stable identity-linked differentiation. Persona plus strategic thinking instruction produced both identity-linked differentiation and goal-directed complementarity across agents.

Conclusion: Multi-agent LLM systems can be steered from mere aggregates to higher-order collectives through prompt design, mirroring principles of collective intelligence in human groups that require both shared objectives and complementary contributions.

Abstract: When are multi-agent LLM systems merely a collection of individual agents versus an integrated collective with higher-order structure? We introduce an information-theoretic framework to test – in a purely data-driven way – whether multi-agent systems show signs of higher-order structure. This information decomposition lets us measure whether dynamical emergence is present in multi-agent LLM systems, localize it, and distinguish spurious temporal coupling from performance-relevant cross-agent synergy. We implement both a practical criterion and an emergence capacity criterion operationalized as partial information decomposition of time-delayed mutual information (TDMI). We apply our framework to experiments using a simple guessing game without direct agent communication and only minimal group-level feedback with three randomized interventions. Groups in the control condition exhibit strong temporal synergy but only little coordinated alignment across agents. Assigning a persona to each agent introduces stable identity-linked differentiation. Combining personas with an instruction to ``think about what other agents might do’’ shows identity-linked differentiation and goal-directed complementarity across agents. Taken together, our framework establishes that multi-agent LLM systems can be steered with prompt design from mere aggregates to higher-order collectives. Our results are robust across emergence measures and entropy estimators, and not explained by coordination-free baselines or temporal dynamics alone. Without attributing human-like cognition to the agents, the patterns of interaction we observe mirror well-established principles of collective intelligence in human groups: effective performance requires both alignment on shared objectives and complementary contributions across members.

Method: Extended Epstein’s Agent_Zero framework with eight behavioral enhancements: age-based impulse control, memory-based risk estimation, affect-cognition coupling, endogenous destructive radius, fight-or-flight dynamics, affective homophily, retaliatory damage, and multi-agent coordination. Implemented in Python using Mesa ABM framework.

Result: The model produces emergent dynamics including protest asymmetries, escalation cycles, and localized retaliation. Results show how small variations in memory, reactivity, and affective alignment can amplify or dampen unrest through feedback loops.

Conclusion: AgentZero++ provides a flexible and extensible platform for analyzing affective contagion and psychologically grounded collective action, explicitly modeling emotional thresholds, identity-driven behavior, and adaptive networks.

Abstract: We present AgentZero++, an agent-based model that integrates cognitive, emotional, and social mechanisms to simulate decentralized collective violence in spatially distributed systems. Building on Epstein’s Agent_Zero framework, we extend the original model with eight behavioral enhancements: age-based impulse control; memory-based risk estimation; affect-cognition coupling; endogenous destructive radius; fight-or-flight dynamics; affective homophily; retaliatory damage; and multi-agent coordination. These additions allow agents to adapt based on internal states, previous experiences, and social feedback, producing emergent dynamics such as protest asymmetries, escalation cycles, and localized retaliation. Implemented in Python using the Mesa ABM framework, AgentZero++ enables modular experimentation and visualization of how micro-level cognitive heterogeneity shapes macro-level conflict patterns. Our results highlight how small variations in memory, reactivity, and affective alignment can amplify or dampen unrest through feedback loops. By explicitly modeling emotional thresholds, identity-driven behavior, and adaptive networks, this work contributes a flexible and extensible platform for analyzing affective contagion and psychologically grounded collective action.

Method: Model app’s UI transition structure as UI Transition Graph (UTG), reformulate UI automation as pathfinding problem, use symbolic planner to generate optimal high-level plan.

Result: Improves success rates of state-of-the-art UI agents by up to 14% and reduces action steps by 37.7% on AndroidWorld benchmark.

Conclusion: AGENT+P effectively enhances existing UI agents through symbolic planning and UTG modeling, preventing redundant exploration and improving automation performance.

Abstract: Large Language Model (LLM)-based UI agents show great promise for UI automation but often hallucinate in long-horizon tasks due to their lack of understanding of the global UI transition structure. To address this, we introduce AGENT+P, a novel framework that leverages symbolic planning to guide LLM-based UI agents. Specifically, we model an app’s UI transition structure as a UI Transition Graph (UTG), which allows us to reformulate the UI automation task as a pathfinding problem on the UTG. This further enables an off-the-shelf symbolic planner to generate a provably correct and optimal high-level plan, preventing the agent from redundant exploration and guiding the agent to achieve the automation goals. AGENT+P is designed as a plug-and-play framework to enhance existing UI agents. Evaluation on the AndroidWorld benchmark demonstrates that AGENT+P improves the success rates of state-of-the-art UI agents by up to 14% and reduces the action steps by 37.7%.

Method: Integrates world models with communication channels using bidirectional message exchange with contrastive learning for message alignment. Uses temporal extension of collective predictive coding.

Result: Outperforms non-communicative models when agents have divergent perceptual capabilities, achieving second-best coordination after centralized models. Facilitates emergence of meaningful symbol systems that accurately reflect environmental states.

Conclusion: Decentralized communication effectively supports coordination while developing shared environmental representations, demonstrating the value of simultaneous communication and coordination in multi-agent systems.

Abstract: We propose a fully decentralized multi-agent world model that enables both symbol emergence for communication and coordinated behavior through temporal extension of collective predictive coding. Unlike previous research that focuses on either communication or coordination separately, our approach achieves both simultaneously. Our method integrates world models with communication channels, enabling agents to predict environmental dynamics, estimate states from partial observations, and share critical information through bidirectional message exchange with contrastive learning for message alignment. Using a two-agent trajectory drawing task, we demonstrate that our communication-based approach outperforms non-communicative models when agents have divergent perceptual capabilities, achieving the second-best coordination after centralized models. Importantly, our decentralized approach with constraints preventing direct access to other agents’ internal states facilitates the emergence of more meaningful symbol systems that accurately reflect environmental states. These findings demonstrate the effectiveness of decentralized communication for supporting coordination while developing shared representations of the environment.

Method: Proposes QLLM algorithm with TFCAF concept, representing credit allocation as direct nonlinear functional formulation using a coder-evaluator framework to guide LLM code generation and refinement.

Result: Extensive experiments show QLLM consistently outperforms state-of-the-art baselines on standard MARL benchmarks with strong generalization capability.

Conclusion: QLLM is a promising and versatile solution for complex multi-agent scenarios, maintaining compatibility with various MARL algorithms using mixing networks.

Abstract: Credit assignment has remained a fundamental challenge in multi-agent reinforcement learning (MARL). Previous studies have primarily addressed this issue through value decomposition methods under the centralized training with decentralized execution paradigm, where neural networks are utilized to approximate the nonlinear relationship between individual Q-values and the global Q-value. Although these approaches have achieved considerable success in various benchmark tasks, they still suffer from several limitations, including imprecise attribution of contributions, limited interpretability, and poor scalability in high-dimensional state spaces. To address these challenges, we propose a novel algorithm, \textbf{QLLM}, which facilitates the automatic construction of credit assignment functions using large language models (LLMs). Specifically, the concept of \textbf{TFCAF} is introduced, wherein the credit allocation process is represented as a direct and expressive nonlinear functional formulation. A custom-designed \textit{coder-evaluator} framework is further employed to guide the generation, verification, and refinement of executable code by LLMs, significantly mitigating issues such as hallucination and shallow reasoning during inference. Extensive experiments conducted on several standard MARL benchmarks demonstrate that the proposed method consistently outperforms existing state-of-the-art baselines. Moreover, QLLM exhibits strong generalization capability and maintains compatibility with a wide range of MARL algorithms that utilize mixing networks, positioning it as a promising and versatile solution for complex multi-agent scenarios.

Method: Three-step approach: (1) Multi-Expert Collaborative Reasoning to compensate missing modalities, (2) Incomplete Modality Adapters to handle new feature distributions, (3) Modality Missing Learning with adaptive weighting and contrastive learning.

Result: Superior performance on three real-world benchmarks across two languages compared to state-of-the-art methods while maintaining simplicity.

Conclusion: MMLNet effectively addresses modality incompleteness in fake news detection, improving robustness and helping curb malicious misinformation spread.

Abstract: Multimodal fake news detection (MFND) has become an urgent task with the emergence of huge multimodal fake content on social media platforms. Previous studies mainly focus on complex feature extraction and fusion to learn discriminative information from multimodal content. However, in real-world applications, multimedia news may naturally lose some information during dissemination, resulting in modality incompleteness, which is detrimental to the generalization and robustness of existing models. To this end, we propose a novel generic and robust multimodal fusion strategy, termed Multi-expert Modality-incomplete Learning Network (MMLNet), which is simple yet effective. It consists of three key steps: (1) Multi-Expert Collaborative Reasoning to compensate for missing modalities by dynamically leveraging complementary information through multiple experts. (2) Incomplete Modality Adapters compensates for the missing information by leveraging the new feature distribution. (3) Modality Missing Learning leveraging an label-aware adaptive weighting strategy to learn a robust representation with contrastive learning. We evaluate MMLNet on three real-world benchmarks across two languages, demonstrating superior performance compared to state-of-the-art methods while maintaining relative simplicity. By ensuring the accuracy of fake news detection in incomplete modality scenarios caused by information propagation, MMLNet effectively curbs the spread of malicious misinformation. Code is publicly available at https://github.com/zhyhome/MMLNet.

Method: Propose a diffusion model with three conditioning signals: panoramic saliency map for regions of interest, bounding-box-aware signed distance map for target viewpoint, and descriptive caption of entire scene.

Result: Generates spatially-aware viewpoint videos and audios that are coherently influenced by broader environmental context, introducing strong controllability for realistic immersive audio-visual generation.

Conclusion: This is the first framework for controllable audio-visual generation from 360-degree environments, demonstrating effectiveness through audiovisual examples.

Abstract: The generation of sounding videos has seen significant advancements with the advent of diffusion models. However, existing methods often lack the fine-grained control needed to generate viewpoint-specific content from larger, immersive 360-degree environments. This limitation restricts the creation of audio-visual experiences that are aware of off-camera events. To the best of our knowledge, this is the first work to introduce a framework for controllable audio-visual generation, addressing this unexplored gap. Specifically, we propose a diffusion model by introducing a set of powerful conditioning signals derived from the full 360-degree space: a panoramic saliency map to identify regions of interest, a bounding-box-aware signed distance map to define the target viewpoint, and a descriptive caption of the entire scene. By integrating these controls, our model generates spatially-aware viewpoint videos and audios that are coherently influenced by the broader, unseen environmental context, introducing a strong controllability that is essential for realistic and immersive audio-visual generation. We show audiovisual examples proving the effectiveness of our framework.

Method: Introduces parameter-efficient front-ends using prompt-tuning fused with signal processing transforms (FourierPT-XLSR, WSPT-XLSR, Partial-WSPT-XLSR), and WaveSP-Net combining Partial-WSPT-XLSR front-end with bidirectional Mamba back-end to inject multi-resolution features into prompt embeddings.

Result: WaveSP-Net outperforms state-of-the-art models on Deepfake-Eval-2024 and SpoofCeleb benchmarks with low trainable parameters and significant performance gains.

Conclusion: The proposed parameter-efficient approach effectively enhances localization of synthetic artifacts without altering frozen model parameters, providing superior performance for speech deepfake detection.

Abstract: Modern front-end design for speech deepfake detection relies on full fine-tuning of large pre-trained models like XLSR. However, this approach is not parameter-efficient and may lead to suboptimal generalization to realistic, in-the-wild data types. To address these limitations, we introduce a new family of parameter-efficient front-ends that fuse prompt-tuning with classical signal processing transforms. These include FourierPT-XLSR, which uses the Fourier Transform, and two variants based on the Wavelet Transform: WSPT-XLSR and Partial-WSPT-XLSR. We further propose WaveSP-Net, a novel architecture combining a Partial-WSPT-XLSR front-end and a bidirectional Mamba-based back-end. This design injects multi-resolution features into the prompt embeddings, which enhances the localization of subtle synthetic artifacts without altering the frozen XLSR parameters. Experimental results demonstrate that WaveSP-Net outperforms several state-of-the-art models on two new and challenging benchmarks, Deepfake-Eval-2024 and SpoofCeleb, with low trainable parameters and notable performance gains. The code and models are available at https://github.com/xxuan-acoustics/WaveSP-Net.

Method: Generates pseudo-labels via majority voting from predictions, then optimizes model using reinforcement learning with confidence-based weighting to handle noisy labels and multiple-attempt sampling to stabilize training.

Result: Achieved significant improvements of 4.42% for 7B model and 11.04% for 3B model on MMAU, MMAR, and MMSU benchmarks, with adapted 3B model outperforming unadapted 7B model.

Conclusion: Test-time adaptation is highly effective for audio understanding, enabling smaller models to outperform larger unadapted models through on-the-fly learning from unlabeled test data.

Abstract: Large Audio Language Models (LALMs) demonstrate impressive general audio understanding, but once deployed, they are static and fail to improve with new real-world audio data. As traditional supervised fine-tuning is costly, we introduce a novel framework for test-time audio understanding, AQA-TTRL, where an LALM evolves on-the-fly using only unlabeled test data. It first generates pseudo-labels from the prediction via majority voting, then optimizes the model via reinforcement learning. To handle the inherent noise in these self-generated labels, we introduce a confidence-based weighting method to adjust training signals. Furthermore, a multiple-attempt sampling operation mitigates advantage collapse and stabilizes training. On the MMAU (test-mini/test), MMAR, and MMSU benchmarks, AQA-TTRL achieves significant average improvements of 4.42% for the Qwen2.5-Omni 7B model and 11.04% for the 3B model. Notably, the adapted 3B model consistently outperforms the direct inference of the unadapted 7B model, highlighting the effectiveness of previously unexplored test-time adaptations in audio understanding.

Method: Uses reinforcement learning (Proximal Policy Optimization) to train policies that control vocal tract articulators (tongue, lips, jaw) based on acoustic feedback from Sylber audio perceiver. Articulatory trajectories are decoded to audio using SPARC decoder.

Result: Successfully trained on six target syllables with similarity scores exceeding 0.85. Generated audio was accurately transcribed by humans for syllables like “please”, “loot”, and “cat”, demonstrating intelligibility.

Conclusion: The framework successfully demonstrates that speech can be generated through explicit articulatory control, providing more interpretable and physically-grounded speech production compared to black-box transformer models.

Abstract: Current speech production systems predominantly rely on large transformer models that operate as black boxes, providing little interpretability or grounding in the physical mechanisms of human speech. We address this limitation by proposing a new framework: speech generation through explicit articulatory control. This reframes speech as a motor control task similar to robotic manipulation. Our approach uses reinforcement learning to train a policy that directly controls the movements of vocal tract articulators, such as the tongue, lips, and jaw, to produce syllable-level speech. Specifically, we employ the Proximal Policy Optimization algorithm to learn optimal articulatory movements based on acoustic feedback provided by our audio perceiver, Sylber. The resulting articulatory trajectories are decoded into audio using SPARC, a pre-trained articulatory-to-speech decoder. We train this framework on six target syllables, and it demonstrates successful convergence, with similarity scores between the policy-generated audio and the target syllables exceeding 0.85. Accurate human transcription of the audio for syllables such as “please”, “loot”, and “cat” demonstrates the intelligibility of this framework.

Method: The study investigates the inconsistency via modifications to speaker embedding in speech generation process, discovering a subspace whose removal alters machine perception while preserving human perception. Experiments conducted on FACodec and Diff-HierVC speech generation models.

Result: Developed an asynchronous voice anonymization achieving 100% human perception preservation rate while obscuring machine perception.

Conclusion: The research successfully identified and leveraged the perception inconsistency to create effective voice anonymization that maintains human-perceived speaker identity while confusing machine recognition systems.

Abstract: Given the speech generation framework that represents the speaker attribute with an embedding vector, asynchronous voice anonymization can be achieved by modifying the speaker embedding derived from the original speech. However, the inconsistency between machine and human perceptions of the speaker attribute within the speaker embedding remains unexplored, limiting its performance in asynchronous voice anonymization. To this end, this study investigates this inconsistency via modifications to speaker embedding in the speech generation process. Experiments conducted on the FACodec and Diff-HierVC speech generation models discover a subspace whose removal alters machine perception while preserving its human perception of the speaker attribute in the generated speech. With these findings, an asynchronous voice anonymization is developed, achieving 100% human perception preservation rate while obscuring the machine perception. Audio samples can be found in https://voiceprivacy.github.io/speaker-embedding-eigen-decomposition/.

Method: Proposed CxNet uses two DNNs with a neural forward filtering module in between. The first DNN predicts direct-path signals and reverberant speech, then the filtering module estimates linear filters and convolves them with direct-path estimates to generate discriminative features that help the second DNN better estimate reverberant speech.

Result: Evaluation on the SMS-WSJ dataset demonstrates the effectiveness of the proposed CxNet approach for separating mixed speakers while preserving individual reverberation characteristics.

Conclusion: By explicitly modeling the linear filter relationship between direct-path signals and reverberant speech, CxNet successfully leverages physical constraints to improve multi-speaker separation in reverberant environments while maintaining reverberation characteristics.

Abstract: We address monaural multi-speaker-image separation in reverberant conditions, aiming at separating mixed speakers but preserving the reverberation of each speaker. A straightforward approach for this task is to directly train end-to-end DNN systems to predict the reverberant speech of each speaker based on the input mixture. Although effective, this approach does not explicitly exploit the physical constraint that reverberant speech can be reproduced by convolving the direct-path signal with a linear filter. To address this, we propose CxNet, a two-DNN system with a neural forward filtering module in between. The first DNN is trained to jointly predict the direct-path signal and reverberant speech. Based on the direct-path estimate, the neural forward filtering module estimates the linear filter, and the estimated filter is then convolved with the direct-path estimate to obtain another estimate of reverberant speech, which is utilized as a discriminative feature to help the second DNN better estimate the reverberant speech. By explicitly modeling the linear filter, CxNet could leverage the physical constraint between the direct-path signal and reverberant speech to capture crucial information about reverberation tails. Evaluation results on the SMS-WSJ dataset show the effectiveness of the proposed algorithms.

Method: Used null hypothesis significance testing framework to systematically assess statistical independence between MFCCs and three prosodic features: energy, fundamental frequency (F0), and voicing.

Result: Demonstrated that it is statistically implausible that MFCCs are independent of any of the three prosodic features.

Conclusion: MFCCs inherently carry valuable prosodic information, which can inform the design of future models in speech analysis and recognition.

Abstract: Mel-frequency cepstral coefficients (MFCCs) are an important feature in speech processing. A deeper understanding of their properties can contribute to the work that is being done with both classical and deep learning models. This study challenges the long-held assumption that MFCCs lack relevant temporal information by investigating their relationship with speech prosody. Using a null hypothesis significance testing framework, a systematic assessment is made about the statistical independence between MFCCs and the three prosodic features: energy, fundamental frequency (F0), and voicing. The results demonstrate that it is statistically implausible that the MFCCs are independent of any of these three prosodic features. This finding suggests that MFCCs inherently carry valuable prosodic information, which can inform the design of future models in speech analysis and recognition.

Method: Proposes three key approaches: (1) Retain all emotional ratings using soft-label distributions and train models on individual annotator ratings, (2) Implement an “all-inclusive rule” that aggregates all ratings to maximize diversity, allowing co-occurring emotions, (3) Use a penalization matrix to discourage unlikely emotion combinations during training.

Result: Experiments on four English emotion databases show superior performance over conventional majority and plurality labeling approaches. Models trained with individual annotator ratings improve performance on consensus-labeled tests, and the all-inclusive rule outperforms traditional aggregation methods.

Conclusion: Embracing minority ratings, multiple annotator perspectives, and multi-emotion predictions leads to more robust and human-aligned speech emotion recognition systems that better reflect the subjective nature of emotion perception.

Abstract: Over the past two decades, speech emotion recognition (SER) has received growing attention. To train SER systems, researchers collect emotional speech databases annotated by crowdsourced or in-house raters who select emotions from predefined categories. However, disagreements among raters are common. Conventional methods treat these disagreements as noise, aggregating labels into a single consensus target. While this simplifies SER as a single-label task, it ignores the inherent subjectivity of human emotion perception. This dissertation challenges such assumptions and asks: (1) Should minority emotional ratings be discarded? (2) Should SER systems learn from only a few individuals’ perceptions? (3) Should SER systems predict only one emotion per sample? Psychological studies show that emotion perception is subjective and ambiguous, with overlapping emotional boundaries. We propose new modeling and evaluation perspectives: (1) Retain all emotional ratings and represent them with soft-label distributions. Models trained on individual annotator ratings and jointly optimized with standard SER systems improve performance on consensus-labeled tests. (2) Redefine SER evaluation by including all emotional data and allowing co-occurring emotions (e.g., sad and angry). We propose an ``all-inclusive rule’’ that aggregates all ratings to maximize diversity in label representation. Experiments on four English emotion databases show superior performance over majority and plurality labeling. (3) Construct a penalization matrix to discourage unlikely emotion combinations during training. Integrating it into loss functions further improves performance. Overall, embracing minority ratings, multiple annotators, and multi-emotion predictions yields more robust and human-aligned SER systems.

Method: A fully discrete speech chain with semantic-token ASR coupled with two-stage TTS: autoregressive text-to-semantic model co-trained with ASR, and masked-generative semantic-to-acoustic model for synthesis only. Uses straight-through argmax/Gumbel-Softmax for end-to-end feedback and dynamic weight averaging to balance supervised ASR.

Result: TokenChain surpasses baseline accuracy 2-6 epochs earlier, yields 5-13% lower equal-epoch error with stable T2S on LibriSpeech, and reduces relative ASR WER by 56% and T2S WER by 31% on TED-LIUM with minimal forgetting.

Conclusion: Chain learning remains effective with token interfaces and models, enabling improved joint ASR and TTS performance through discrete representations and end-to-end feedback mechanisms.

Abstract: Machine Speech Chain, simulating the human perception-production loop, proves effective in jointly improving ASR and TTS. We propose TokenChain, a fully discrete speech chain coupling semantic-token ASR with a two-stage TTS: an autoregressive text-to-semantic model co-trained with ASR and a masked-generative semantic-to-acoustic model for synthesis only. End-to-end feedback across the text interface is enabled with straight-through argmax/Gumbel-Softmax and balanced with supervised ASR via dynamic weight averaging. Ablations examine optimal temperature schedules for in- and cross-domain transfer. Evaluation reveals TokenChain surpasses baseline accuracy 2-6 epochs earlier and yields 5-13% lower equal-epoch error with stable T2S on LibriSpeech, and reduces relative ASR WER by 56% and T2S WER by 31% on TED-LIUM with minimal forgetting, showing that chain learning remains effective with token interfaces and models.

Method: Used AASIST-based approach with WavLM large frontend and RawBoost augmentation, trained on multilingual dataset of 256,600 samples from 9 languages and 70+ TTS systems across multiple datasets.

Result: Achieved second place in Task 1 (unmodified audio detection) and Task 3 (laundered audio detection), demonstrating strong generalization and robustness.

Conclusion: The systematic exploration of SSL front-ends, training data compositions, and audio length configurations enables robust deepfake detection across challenging scenarios.

Abstract: The SAFE Challenge evaluates synthetic speech detection across three tasks: unmodified audio, processed audio with compression artifacts, and laundered audio designed to evade detection. We systematically explore self-supervised learning (SSL) front-ends, training data compositions, and audio length configurations for robust deepfake detection. Our AASIST-based approach incorporates WavLM large frontend with RawBoost augmentation, trained on a multilingual dataset of 256,600 samples spanning 9 languages and over 70 TTS systems from CodecFake, MLAAD v5, SpoofCeleb, Famous Figures, and MAILABS. Through extensive experimentation with different SSL front-ends, three training data versions, and two audio lengths, we achieved second place in both Task 1 (unmodified audio detection) and Task 3 (laundered audio detection), demonstrating strong generalization and robustness.

Method: Used X-vector system from Kaldi for audio-only trials in closed-set condition. For audio-visual results, used visual modality models only. Employed pretrained models on VoxBlink2 and VoxCeleb2 datasets, and trained X-vector models from scratch using CTS superset dataset.

Result: Submitted results for both open-set and closed-set conditions to the competition website, with performance analysis provided in the report.

Conclusion: The team successfully developed and submitted speaker recognition systems for NIST SRE 2024 evaluation, demonstrating the effectiveness of their approach across different conditions and modalities.

Abstract: The CL-UZH team submitted one system each for the fixed and open conditions of the NIST SRE 2024 challenge. For the closed-set condition, results for the audio-only trials were achieved using the X-vector system developed with Kaldi. For the audio-visual results we used only models developed for the visual modality. Two sets of results were submitted for the open-set and closed-set conditions, one based on a pretrained model using the VoxBlink2 and VoxCeleb2 datasets. An Xvector-based model was trained from scratch using the CTS superset dataset for the closed set. In addition to the submission of the results of the SRE24 evaluation to the competition website, we talked about the performance of the proposed systems on the SRE24 evaluation in this report.

Method: Proposed MuFFIN with interactive hierarchical neural architecture, phoneme-contrastive ordinal regularization for phoneme-discriminative features, and a training objective to handle MDD data imbalance by perturbing phoneme classifier outputs with phoneme-specific variations.

Result: Experiments on Speechocean762 benchmark show state-of-the-art performance on both APA and MDD tasks, outperforming several cutting-edge baselines.

Conclusion: The proposed MuFFIN model effectively integrates MDD and APA tasks, demonstrating that joint modeling with appropriate regularization and training objectives can significantly improve pronunciation assessment performance.

Abstract: Computer-assisted pronunciation training (CAPT) manages to facilitate second-language (L2) learners to practice pronunciation skills by offering timely and instructive feedback. To examine pronunciation proficiency from multiple facets, existing methods for CAPT broadly fall into two categories: mispronunciation detection and diagnosis (MDD) as well as automatic pronunciation assessment (APA). The former aims to pinpoint phonetic pronunciation errors and provide diagnostic feedback, while the latter seeks instead to quantify pronunciation proficiency pertaining to various aspects. Despite the natural complementarity between MDD and APA, researchers and practitioners, however, often treat them as independent tasks with disparate modeling paradigms. In light of this, we in this paper first introduce MuFFIN, a Multi-Faceted pronunciation Feedback model with an Interactive hierarchical Neural architecture, to jointly address the tasks of MDD and APA. To better capture the nuanced distinctions between phonemes in the feature space, a novel phoneme-contrastive ordinal regularization mechanism is then put forward to optimize the proposed model to generate more phoneme-discriminative features while factoring in the ordinality of the aspect scores. In addition, to address the intricate data imbalance problem in MDD, we design a simple yet effective training objective, which is specifically tailored to perturb the outputs of a phoneme classifier with the phoneme-specific variations, so as to better render the distribution of predicted phonemes meanwhile considering their mispronunciation characteristics. A series of experiments conducted on the Speechocean762 benchmark dataset demonstrates the efficacy of our method in relation to several cutting-edge baselines, showing state-of-the-art performance on both the APA and MDD tasks.

Method: Combines real-time EEG signals from wearable skullcap with structural MRI data using Enhanced Vision Transformer (ViT++) with Patch-Level Attention Regularization and Adaptive Threshold Mechanism, plus Bidirectional LSTM for EEG pattern analysis over time.

Result: Achieved 94.6% precision, 93.2% recall, and Dice score of 0.91 for tumor localization and brain state classification, with interactive 3D visualization and tumor growth prediction capabilities.

Conclusion: Sets a new standard for real-time, interpretable neurodiagnostics and paves the way for future advancements in intelligent brain health monitoring.

Abstract: Neuro-oncological prognostics are now vital in modern clinical neuroscience because brain tumors pose significant challenges in detection and management. To tackle this issue, we propose a cognitive digital twin framework that combines real-time EEG signals from a wearable skullcap with structural MRI data for dynamic and personalized tumor monitoring. At the heart of this framework is an Enhanced Vision Transformer (ViT++) that includes innovative components like Patch-Level Attention Regularization (PLAR) and an Adaptive Threshold Mechanism to improve tumor localization and understanding. A Bidirectional LSTM-based neural classifier analyzes EEG patterns over time to classify brain states such as seizure, interictal, and healthy. Grad-CAM-based heatmaps and a three.js-powered 3D visualization module provide interactive anatomical insights. Furthermore, a tumor kinetics engine predicts volumetric growth by looking at changes in MRI trends and anomalies from EEG data. With impressive accuracy metrics of 94.6% precision, 93.2% recall, and a Dice score of 0.91, this framework sets a new standard for real-time, interpretable neurodiagnostics. It paves the way for future advancements in intelligent brain health monitoring.

Method: Adapted Hierarchical Functional Maximal Correlation Algorithm (HFMCA) to graph-structured fMRI data, using density ratio decomposition in reproducing kernel Hilbert space (RKHS) for pretraining to learn robust representations.

Result: Evaluations across five neuroimaging datasets show the method produces competitive embeddings for classification tasks and enables effective knowledge transfer to unseen datasets.

Conclusion: The adapted HFMCA approach provides a theoretically grounded method for learning robust and generalizable representations from fMRI data, overcoming domain variability challenges in neuroimaging analysis.

Abstract: Functional magnetic resonance imaging (fMRI) analysis faces significant challenges due to limited dataset sizes and domain variability between studies. Traditional self-supervised learning methods inspired by computer vision often rely on positive and negative sample pairs, which can be problematic for neuroimaging data where defining appropriate contrasts is non-trivial. We propose adapting a recently developed Hierarchical Functional Maximal Correlation Algorithm (HFMCA) to graph-structured fMRI data, providing a theoretically grounded approach that measures statistical dependence via density ratio decomposition in a reproducing kernel Hilbert space (RKHS),and applies HFMCA-based pretraining to learn robust and generalizable representations. Evaluations across five neuroimaging datasets demonstrate that our adapted method produces competitive embeddings for various classification tasks and enables effective knowledge transfer to unseen datasets. Codebase and supplementary material can be found here: https://github.com/fr30/mri-eigenencoder

Method: Used single-slice SAM2 prompts to generate pseudo-labels, pooled across datasets, and refined through three sequential nnU-Net models. Evaluated with Dice similarity coefficient and statistical tests (TOST, ICC) for automated measurements.

Result: Achieved DSCs of 0.94-0.96 on MRI and 0.92-0.93 on CT, outperforming existing methods. Automated measurements were statistically equivalent to expert references with high ICCs (0.86-1.00) for muscle volume, CT attenuation, and fat ratio.

Conclusion: nnsam2 is a state-of-the-art few-shot framework that provides efficient, generalizable, and reproducible segmentation across multimodal, multicenter cohorts, with open code and data release.

Abstract: Purpose: To develop and validate No-New SAM2 (nnsam2) for few-shot segmentation of lumbar paraspinal muscles using only a single annotated slice per dataset, and to assess its statistical comparability with expert measurements across multi-sequence MRI and multi-protocol CT. Methods: We retrospectively analyzed 1,219 scans (19,439 slices) from 762 participants across six datasets. Six slices (one per dataset) served as labeled examples, while the remaining 19,433 slices were used for testing. In this minimal-supervision setting, nnsam2 used single-slice SAM2 prompts to generate pseudo-labels, which were pooled across datasets and refined through three sequential, independent nnU-Net models. Segmentation performance was evaluated using the Dice similarity coefficient (DSC), and automated measurements-including muscle volume, fat ratio, and CT attenuation-were assessed with two one-sided tests (TOST) and intraclass correlation coefficients (ICC). Results: nnsam2 outperformed vanilla SAM2, its medical variants, TotalSegmentator, and the leading few-shot method, achieving DSCs of 0.94-0.96 on MR images and 0.92-0.93 on CT. Automated and expert measurements were statistically equivalent for muscle volume (MRI/CT), CT attenuation, and Dixon fat ratio (TOST, P < 0.05), with consistently high ICCs (0.86-1.00). Conclusion: We developed nnsam2, a state-of-the-art few-shot framework for multi-modality LPM segmentation, producing muscle volume (MRI/CT), attenuation (CT), and fat ratio (Dixon MRI) measurements that were statistically comparable to expert references. Validated across multimodal, multicenter, and multinational cohorts, and released with open code and data, nnsam2 demonstrated high annotation efficiency, robust generalizability, and reproducibility.

Method: Uses a multi-layer perceptron that maps pixel coordinates and transmit steering angles to complex-valued apodization weights for each receive channel in an implicit neural representation framework.

Result: Experiments on in vivo porcine ICE data show the learned apodization suppresses clutter and enhances contrast, producing reconstructions that closely match 26-angle compounded DW ground truths.

Conclusion: Implicit neural representations offer a powerful framework for ultrasound image enhancement, enabling high-quality ICE reconstructions with significantly fewer transmits.

Abstract: Ultrafast intracardiac echocardiography (ICE) uses unfocused transmissions to capture cardiac motion at frame rates exceeding 1 kHz. While this enables real-time visualization of rapid dynamics, image quality is often degraded by diffraction artifacts, requiring many transmits to achieve satisfying resolution and contrast. To address this limitation, we propose an implicit neural representation (INR) framework to encode complex-valued receive apodization weights in a continuous manner, enabling high-quality ICE reconstructions from only three diverging wave (DW) transmits. Our method employs a multi-layer perceptron that maps pixel coordinates and transmit steering angles to complex-valued apodization weights for each receive channel. Experiments on a large in vivo porcine ICE imaging dataset show that the learned apodization suppresses clutter and enhances contrast, yielding reconstructions closely matching 26-angle compounded DW ground truths. Our study suggests that INRs could offer a powerful framework for ultrasound image enhancement.

Method: Proposed a modulated INR framework with coordinate-based neural network conditioned on latent embeddings from time-delayed I/Q data, using complex Gabor wavelet activation and conditioner network to capture oscillatory and phase-sensitive signal characteristics.

Result: Outperformed state-of-the-art methods on in vivo intracardiac echocardiography dataset.

Conclusion: INR-based modeling shows advantages for ultrasound reconstruction and has broader potential applications across other medical imaging modalities.

Abstract: Ultrafast ultrasound imaging enables visualization of rapid physiological dynamics by acquiring data at exceptionally high frame rates. However, this speed often comes at the cost of spatial resolution and image quality due to unfocused wave transmissions and associated artifacts. In this work, we propose a novel modulated Implicit Neural Representation (INR) framework that leverages a coordinate-based neural network conditioned on latent embeddings extracted from time-delayed I/Q channel data for high-quality ultrasound image reconstruction. Our method integrates complex Gabor wavelet activation and a conditioner network to capture the oscillatory and phase-sensitive nature of I/Q ultrasound signals. We evaluate the framework on an in vivo intracardiac echocardiography (ICE) dataset and demonstrate that it outperforms the compared state-of-the-art methods. We believe these findings not only highlight the advantages of INR-based modeling for ultrasound image reconstruction, but also point to broader opportunities for applying INR frameworks across other medical imaging modalities.

Method: Developed an end-to-end pipeline with ISO/IEC 29794-6 quality compliance at acquisition, using a custom Android app for real-time framing and sharpness evaluation. Created LightIrisNet (MobileNetV3-based multi-task segmentation) for on-device processing and IrisFormer (transformer matcher) adapted to VIS domain.

Result: OSIRIS achieved TAR of 97.9% at FAR=0.01 (EER=0.76%), while IrisFormer trained only on UBIRIS.v2 achieved EER of 0.057% on CUVIRIS dataset (752 compliant images from 47 subjects).

Conclusion: Standardized capture protocols and VIS-adapted lightweight models enable accurate and practical iris recognition on smartphones, with released acquisition app, trained models, and dataset subset to support reproducibility.

Abstract: Smartphone-based iris recognition in the visible spectrum (VIS) remains difficult due to illumination variability, pigmentation differences, and the absence of standardized capture controls. This work presents a compact end-to-end pipeline that enforces ISO/IEC 29794-6 quality compliance at acquisition and demonstrates that accurate VIS iris recognition is feasible on commodity devices. Using a custom Android application performing real-time framing, sharpness evaluation, and feedback, we introduce the CUVIRIS dataset of 752 compliant images from 47 subjects. A lightweight MobileNetV3-based multi-task segmentation network (LightIrisNet) is developed for efficient on-device processing, and a transformer matcher (IrisFormer) is adapted to the VIS domain. Under a standardized protocol and comparative benchmarking against prior CNN baselines, OSIRIS attains a TAR of 97.9% at FAR=0.01 (EER=0.76%), while IrisFormer, trained only on UBIRIS.v2, achieves an EER of 0.057% on CUVIRIS. The acquisition app, trained models, and a public subset of the dataset are released to support reproducibility. These results confirm that standardized capture and VIS-adapted lightweight models enable accurate and practical iris recognition on smartphones.

Method: Proposed HP-FINE network fine-tuning based on high-pass filtering forward model with low-frequency preservation regularization. Compared different architectures (Unet, Progressive Unet, Big Unet), output types (field vs susceptibility), and pre-training strategies with filtering augmentation.

Result: Low-frequency regularization in HP-FINE substantially improved prediction accuracy. Recovered field output preserved ROI values in prospective datasets, unlike susceptibility output. Progressive Unet with multiple losses outperformed other architectures in ROI preservation.

Conclusion: HP-FINE with low-frequency regularization and recovered field output effectively improves QSM prediction generalization and preserves ROI values, with Progressive Unet showing best performance when pre-trained with multiple losses.

Abstract: Purpose: To improve the generalization ability of deep learning based predictions of quantitative susceptibility mapping (QSM) from high-pass filtered phase (HPFP) data. Methods: A network fine-tuning step called HP-FINE is proposed, which is based on the high-pass filtering forward model with low-frequency preservation regularization. Several comparisons were conducted:

1. HP-FINE with and without low-frequency regularization, 2. three 3D network architectures (Unet, Progressive Unet, and Big Unet), 3. two types of network output (recovered field and susceptibility), and 4. pre-training with and without the filtering augmentation. HPFP datasets with diverse high-pass filters, another acquisition voxel size, and prospective acquisition were used to assess the accuracy of QSM predictions. In the retrospective datasets, quantitative metrics (PSNR, SSIM, RMSE and HFEN) were used for evaluation. In the prospective dataset, statistics of ROI linear regression and Bland-Altman analysis were used for evaluation. Results: In the retrospective datasets, adding low-frequency regularization in HP-FINE substantially improved prediction accuracy compared to the pre-trained results, especially when combined with the filtering augmentation and recovered field output. In the prospective datasets, HP-FINE with low-frequency regularization and recovered field output demonstrated the preservation of ROI values, a result that was not achieved when using susceptibility as the output. Furthermore, Progressive Unet pre-trained with a combination of multiple losses outperformed both Unet and Progressive Unet pre-trained with a single loss in terms of preserving ROI values.

Method: RimSet combines RimSeg (unsupervised segmentation using level-set methodology) with radiomic measurements using Local Binary Pattern texture descriptors, validated on simulated and in vivo datasets.

Result: RimSeg achieved 78.7% Dice score, RimSet detected rim+ lesions with partial ROC AUC of 0.808 and PR AUC of 0.737, outperforming existing methods. QSMRim-Net showed low error (0.85 MSE) and high correlation (0.91) with expert annotations.

Conclusion: RimSet provides an effective automated solution for quantitative characterization of rim+ lesions in MS, demonstrating superior performance compared to existing methods.

Abstract: Background: Rim+ lesions in multiple sclerosis (MS), detectable via Quantitative Susceptibility Mapping (QSM), correlate with increased disability. Existing literature lacks quantitative analysis of these lesions. We introduce RimSet for quantitative identification and characterization of rim+ lesions on QSM. Methods: RimSet combines RimSeg, an unsupervised segmentation method using level-set methodology, and radiomic measurements with Local Binary Pattern texture descriptors. We validated RimSet using simulated QSM images and an in vivo dataset of 172 MS subjects with 177 rim+ and 3986 rim-lesions. Results: RimSeg achieved a 78.7% Dice score against the ground truth, with challenges in partial rim lesions. RimSet detected rim+ lesions with a partial ROC AUC of 0.808 and PR AUC of 0.737, surpassing existing methods. QSMRim-Net showed the lowest mean square error (0.85) and high correlation (0.91; 95% CI: 0.88, 0.93) with expert annotations at the subject level.

Method: Proposed SAMCIRT combines image reconstruction and affine motion estimation in single update step using analytic adjoints of motion operators, with exact partial derivatives for both reconstruction and motion parameters.

Result: Method outperforms state-of-the-art CT reconstruction with affine motion correction in computational feasibility and projection distance, enabling accurate reconstruction of nonstationary objects like diamonds.

Conclusion: SAMCIRT provides efficient and convergent 4DCT reconstruction with simultaneous motion estimation, demonstrating novel application for nonstationary objects with proven theoretical convergence guarantees.

Abstract: The majority of the recent iterative approaches in 4DCT not only rely on nested iterations, thereby increasing computational complexity and constraining potential acceleration, but also fail to provide a theoretical proof of convergence for their proposed iterative schemes. On the other hand, the latest MATLAB and Python image processing toolboxes lack the implementation of analytic adjoints of affine motion operators for 3D object volumes, which does not allow gradient methods using exact derivatives towards affine motion parameters. In this work, we propose the Simultaneous Affine Motion-Compensated Image Reconstruction Technique (SAMCIRT)- an efficient iterative reconstruction scheme that combines image reconstruction and affine motion estimation in a single update step, based on the analytic adjoints of the motion operators then exact partial derivatives with respect to both the reconstruction and the affine motion parameters. Moreover, we prove the separated Lipschitz continuity of the objective function and its associated functions, including the gradient, which supports the convergence of our proposed iterative scheme, despite the non-convexity of the objective function with respect to the affine motion parameters. Results from simulation and real experiments show that our method outperforms the state-of-the-art CT reconstruction with affine motion correction methods in computational feasibility and projection distance. In particular, this allows accurate reconstruction for a real, nonstationary diamond, showing a novel application of 4DCT.

Method: Transform WSIs into graph representations with nodes from tissue regions respecting natural structures, use adaptive graph coarsening guided by learned embeddings to merge homogeneous regions while preserving critical details, enrich nodes with interpretable clinical features, and apply graph attention network for diagnosis.

Result: Achieves strong performance on cancer staging and survival prediction tasks with >13x fewer parameters and >300x less data while remaining competitive with massive foundation models and offering full interpretability through feature attribution.

Conclusion: The proposed biologically-informed graph representation framework provides an efficient, interpretable alternative to patch-based methods for histopathological analysis, maintaining competitive diagnostic performance while significantly reducing computational requirements.

Abstract: The histopathological analysis of whole-slide images (WSIs) is fundamental to cancer diagnosis but is a time-consuming and expert-driven process. While deep learning methods show promising results, dominant patch-based methods artificially fragment tissue, ignore biological boundaries, and produce black-box predictions. We overcome these limitations with a novel framework that transforms gigapixel WSIs into biologically-informed graph representations and is interpretable by design. Our approach builds graph nodes from tissue regions that respect natural structures, not arbitrary grids. We introduce an adaptive graph coarsening technique, guided by learned embeddings, to efficiently merge homogeneous regions while preserving diagnostically critical details in heterogeneous areas. Each node is enriched with a compact, interpretable feature set capturing clinically-motivated priors. A graph attention network then performs diagnosis on this compact representation. We demonstrate strong performance on challenging cancer staging and survival prediction tasks. Crucially, our resource-efficient model ($>$13x fewer parameters and $>$300x less data) achieves results competitive with a massive foundation model, while offering full interpretability through feature attribution. Our code is publicly available at https://github.com/HistoGraph31/pix2pathology.

Method: Multi-task deep learning network for simultaneous lumbar decomposition and landmark detection, followed by landmark-weighted 2D-3D registration with higher weights for complex posterior structures during optimization.

Result: Achieved sub-millimeter accuracy validated against CT segmentation gold standard, with full reconstruction and measurement workflow completed in under 20 seconds.

Conclusion: The method provides a fast, low-dose automated tool for diagnosing lumbar conditions in functional weight-bearing state, setting a new benchmark for precision and speed.

Abstract: To meet the clinical demand for accurate 3D lumbar spine assessment in a weight-bearing position, this study presents a novel, fully automatic framework for high-precision 3D reconstruction from biplanar X-ray images, overcoming the limitations of existing methods. The core of this method involves a novel multi-task deep learning network that simultaneously performs lumbar decomposition and landmark detection on the original biplanar radiographs. The decomposition effectively eliminates interference from surrounding tissues, simplifying subsequent image registration, while the landmark detection provides an initial pose estimation for the Statistical Shape Model (SSM), enhancing the efficiency and robustness of the registration process. Building on this, we introduce a landmark-weighted 2D-3D registration strategy. By assigning higher weights to complex posterior structures like the transverse and spinous processes during optimization, this strategy significantly enhances the reconstruction accuracy of the posterior arch. Our method was validated against a gold standard derived from registering CT segmentations to the biplanar X-rays. It sets a new benchmark by achieving sub-millimeter accuracy and completes the full reconstruction and measurement workflow in under 20 seconds, establishing a state-of-the-art combination of precision and speed. This fast and low-dose pipeline provides a powerful automated tool for diagnosing lumbar conditions such as spondylolisthesis and scoliosis in their functional, weight-bearing state.

Method: Used hyperspectral images (400-1000nm) from four grapevine cultivars across different vineyards and growth stages over two seasons. Applied feature selection methods to identify optimal spectral bands, then trained Gradient Boosting and XGBoost models for N concentration prediction.

Result: Identified consistent spectral regions (500-525nm, 650-690nm, 750-800nm, 900-950nm) across both feature selection methods and dataset types. ML models achieved R² of 0.49 for canopy-level and 0.57 for leaf-level data.

Conclusion: Demonstrated potential of in-field hyperspectral imaging combined with feature selection and ML techniques for monitoring nitrogen status in vineyards.

Abstract: Nitrogen (N) is one of the most crucial nutrients in vineyards, affecting plant growth and subsequent products such as wine and juice. Because soil N has high spatial and temporal variability, it is desirable to accurately estimate the N concentration of grapevine leaves and manage fertilization at the individual plant level to optimally meet plant needs. In this study, we used in-field hyperspectral images with wavelengths ranging from $400 to 1000nm of four different grapevine cultivars collected from distinct vineyards and over two growth stages during two growing seasons to develop models for predicting N concentration at the leaf-level and canopy-level. After image processing, two feature selection methods were employed to identify the optimal set of spectral bands that were responsive to leaf N concentrations. The selected spectral bands were used to train and test two different Machine Learning (ML) models, Gradient Boosting and XGBoost, for predicting nitrogen concentrations. The comparison of selected bands for both leaf-level and canopy-level datasets showed that most of the spectral regions identified by the feature selection methods were across both methods and the dataset types (leaf- and canopy-level datasets), particularly in the key regions, 500-525nm, 650-690nm, 750-800nm, and 900-950nm. These findings indicated the robustness of these spectral regions for predicting nitrogen content. The results for N prediction demonstrated that the ML model achieved an R square of 0.49 for canopy-level data and an R square of 0.57 for leaf-level data, despite using different sets of selected spectral bands for each analysis level. The study demonstrated the potential of using in-field hyperspectral imaging and the use of spectral data in integrated feature selection and ML techniques to monitor N status in vineyards.

Method: Uses gray and white matter MRI slices as inputs to three pretrained CNNs (ResNet50, NASNet, MobileNet) fine-tuned end-to-end, with stacked ensemble learning and weighted averaging to combine models.

Result: Achieves 99.21% accuracy for AD vs MCI and 91.0% for MCI vs Normal Controls on ADNI dataset, outperforming conventional transfer learning and baseline ensemble methods.

Conclusion: The framework shows potential for robust and scalable clinical decision support in neurodegenerative disease diagnostics, with improved interpretability through Explainable AI techniques.

Abstract: Early and accurate diagnosis of Alzheimer Disease is critical for effective clinical intervention, particularly in distinguishing it from Mild Cognitive Impairment, a prodromal stage marked by subtle structural changes. In this study, we propose a hybrid deep learning ensemble framework for Alzheimer Disease classification using structural magnetic resonance imaging. Gray and white matter slices are used as inputs to three pretrained convolutional neural networks such as ResNet50, NASNet, and MobileNet, each fine tuned through an end to end process. To further enhance performance, we incorporate a stacked ensemble learning strategy with a meta learner and weighted averaging to optimally combine the base models. Evaluated on the Alzheimer Disease Neuroimaging Initiative dataset, the proposed method achieves state of the art accuracy of 99.21% for Alzheimer Disease vs. Mild Cognitive Impairment and 91.0% for Mild Cognitive Impairment vs. Normal Controls, outperforming conventional transfer learning and baseline ensemble methods. To improve interpretability in image based diagnostics, we integrate Explainable AI techniques by Gradient weighted Class Activation, which generates heatmaps and attribution maps that highlight critical regions in gray and white matter slices, revealing structural biomarkers that influence model decisions. These results highlight the frameworks potential for robust and scalable clinical decision support in neurodegenerative disease diagnostics.

Method: Evaluated SkinGPT-4 performance on SCIN dataset across skin tones, developed fine-tuned models for custom skin disease classification, and implemented bias mitigation strategies. Clinical evaluation by dermatologists assessed diagnostic accuracy, utility, and fairness metrics.

Result: SkinGPT-4 showed demographic parity of 0.10 across Fitzpatrick types with 0.10-0.15 differences between lightest and darkest tones. Custom models achieved F1=0.75, precision=0.78, AUROC=0.78 with improved fairness (average demographic parity=0.75).

Conclusion: Large language models like SkinGPT-4 exhibit performance biases across skin tones, but custom fine-tuned models can achieve robust fairness and accuracy for skin disease classification across all skin tones.

Abstract: SkinGPT-4, a large vision-language model, leverages annotated skin disease images to augment clinical workflows in underserved communities. However, its training dataset predominantly represents lighter skin tones, limiting diagnostic accuracy for darker tones. Here, we evaluated performance biases in SkinGPT-4 across skin tones on common skin diseases, including eczema, allergic-contact dermatitis, and psoriasis using the open-sourced SCIN dataset. We leveraged the SkinGPT-4 backbone to develop finetuned models for custom skin disease classification tasks and explored bias mitigation strategies. Clinical evaluation by board-certified dermatologists on six relevant skin diseases from 300 SCIN cases assessed images for diagnostic accuracy, informativity, physician utility, and patient utility. Model fairness metrics, including demographic parity and equalized odds, were calculated across skin tones. SkinGPT-4 achieved an average demographic parity of 0.10 across Fitzpatrick types, with notable differences of 0.10-0.15 between lightest and darkest tones across evaluation metrics. Model hallucinations in artifacts and anatomy occurred at a rate of 17.8. Our customized models achieved average F1, precision, and AUROC of 0.75, 0.78, and 0.78 across visually similar disease pairs. Fairness analysis showed an average demographic parity of 0.75, with a maximum disparity of 0.21 across skin tones. The best model achieved parity scores of 0.83, 0.83, 0.76, 0.89, 0.90, and 0.90 for Fitzpatrick I-VI, indicating robust fairness. Large language models such as SkinGPT-4 showed weaker performance on darker tones. Model biases exist across evaluation criteria, and hallucinations may affect diagnostic efficacy. These findings demonstrate the efficacy of training accurate, fair models using existing backbones for custom skin disease classification.