Method: Proposes Timing-Aware Pre-Quantization Fusion for Video-Enhanced Audio Tokenization. Key findings: 1) Fusion location in tokenizer architecture is crucial for preserving reconstruction quality, 2) Contrastive learning is unsuitable for discrete tokenizers, 3) Temporal-axis fusion guided by distinctive features works best. The method integrates visual information into audio tokenizer while maintaining reconstruction fidelity.

Result: The approach maintains high-fidelity audio reconstruction while achieving superior performance on downstream understanding tasks compared to audio-only tokenizers and established multimodal fusion baselines.

Conclusion: Successfully demonstrates that visual information can be integrated into audio tokenizers without degrading reconstruction quality, enabling better multimodal understanding while preserving audio generation capabilities.

Abstract: Audio tokenization has emerged as a critical component in end-to-end audio language models, enabling efficient discrete representation learning for both audio understanding and generation tasks. However, existing audio tokenizers face fundamental limitations in understanding tasks due to single-modality constraints, particularly when audio signals contain ambiguous or incomplete information. While incorporating additional modality information can significantly enhance audio understanding, current multimodal fusion approaches invariably degrade reconstruction quality. This degradation is unacceptable for end-to-end audio systems that require high-fidelity audio generation capabilities. In this work, we investigate the root causes of reconstruction quality degradation in video-enhanced audio tokenization and present three key findings. First, the location of fusion within the tokenizer architecture is crucial for preserving reconstruction quality. Second, we show that contrastive learning, though effective in continuous representation fusion, is unsuitable for discrete tokenizers as it fails to enhance downstream task performance. Third, while feature-dimension fusion approaches achieve moderate success, we discover that fusing along the temporal axis – guided by the concept of distinctive features – yields significantly better results. Building on these insights, we introduce the Timing-Aware Pre-Quantization Fusion for Video-Enhanced Audio Tokenization, the first approach to successfully integrate visual information into audio tokenizer architectures while preserving reconstruction fidelity. Our approach not only maintains high-fidelity reconstruction but also achieves superior performance on downstream understanding tasks compared with audio-only tokenizers and established multimodal fusion baselines.

Method: Proposes a flow matching-based movie dubbing framework with Cognitive Synchronous Diffusion Transformer (CoSync-DiT) that progressively guides noise-to-speech generation through acoustic style adaptation, fine-grained visual calibration, and time-aware context alignment. Includes Joint Semantic and Alignment Regularization (JSAR) to constrain frame-level temporal consistency and semantic consistency.

Result: Extensive experiments on standard benchmarks and challenging in-the-wild dubbing benchmarks demonstrate state-of-the-art performance across multiple metrics.

Conclusion: The proposed CoSync-DiT framework with JSAR effectively addresses lip-sync precision and naturalness issues in movie dubbing, outperforming existing methods in both controlled and real-world scenarios.

Abstract: Movie dubbing aims to synthesize speech that preserves the vocal identity of a reference audio while synchronizing with the lip movements in a target video. Existing methods fail to achieve precise lip-sync and lack naturalness due to explicit alignment at the duration level. While implicit alignment solutions have emerged, they remain susceptible to interference from the reference audio, triggering timbre and pronunciation degradation in in-the-wild scenarios. In this paper, we propose a novel flow matching-based movie dubbing framework driven by the Cognitive Synchronous Diffusion Transformer (CoSync-DiT), inspired by the cognitive process of professional actors. This architecture progressively guides the noise-to-speech generative trajectory by executing acoustic style adapting, fine-grained visual calibrating, and time-aware context aligning. Furthermore, we design the Joint Semantic and Alignment Regularization (JSAR) mechanism to simultaneously constrain frame-level temporal consistency on the contextual outputs and semantic consistency on the flow hidden states, ensuring robust alignment. Extensive experiments on both standard benchmarks and challenging in-the-wild dubbing benchmarks demonstrate that our method achieves the state-of-the-art performance across multiple metrics.

Method: SpotSound introduces a novel training objective specifically designed to suppress hallucinated timestamps for events absent from the input. The authors also present SpotSound-Bench, a challenging temporal grounding benchmark where target events occupy less than ~10% of each clip, creating a rigorous ’needle-in-a-haystack’ evaluation scenario.

Result: Experiments demonstrate that SpotSound achieves state-of-the-art results on temporal grounding benchmarks while maintaining robust performance across general downstream audio-language tasks.

Conclusion: SpotSound addresses key limitations in ALMs for temporal grounding through innovative training objectives and challenging benchmarks, advancing the field of audio event localization in complex acoustic environments.

Abstract: Large Audio-Language Models (ALMs) have recently demonstrated remarkable capabilities in holistic audio understanding, yet they remain unreliable for temporal grounding, i.e., the task of pinpointing exactly when an event occurs within long-form audio. This limitation stems from two factors: training data dominated by clip-level supervision lacking precise timestamps, and benchmarks that fail to simulate real-world scenarios where short events are obscured by dense background sounds. In this paper, we introduce SpotSound, an audio language model designed for grounding audio events. SpotSound incorporates a novel training objective, specifically designed to suppress hallucinated timestamps for events absent from the input. Additionally, we present SpotSound-Bench, a challenging temporal grounding benchmark where target events occupy less than ~10% of each clip, creating a rigorous `needle-in-a-haystack’ evaluation. Experiments demonstrate that SpotSound achieves state-of-the-art results on temporal grounding benchmarks while maintaining robust performance across general downstream audio-language tasks. Code, models and benchmark are released on https://loiesun.github.io/spotsound/

Method: Proposes reasoning score evaluating traces along dimensions like faithfulness, coherence, utility, and factuality. Introduces Filtered Reasoning Score (FRS) that computes reasoning quality using only top-K% most confident traces to avoid issues with averaging across many possible trajectories.

Result: Models indistinguishable under standard accuracy show significant differences in reasoning quality under FRS. Models with higher FRS on one benchmark tend to perform better on other reasoning benchmarks in both accuracy and reasoning quality.

Conclusion: FRS complements accuracy by capturing transferable reasoning capabilities, providing a more nuanced evaluation of LLM reasoning beyond simple outcome-based metrics.

Abstract: Should we trust Large Language Models (LLMs) with high accuracy? LLMs achieve high accuracy on reasoning benchmarks, but correctness alone does not reveal the quality of the reasoning used to produce it. This highlights a fundamental limitation of outcome-based evaluation: models may arrive at correct answers through flawed reasoning, and models with substantially different reasoning capabilities can nevertheless exhibit similar benchmark accuracy, for example due to memorization or over-optimization. In this paper, we ask: given existing benchmarks, can we move beyond outcome-based evaluation to assess the quality of reasoning itself? We seek metrics that (1) differentiate models with similar accuracy and (2) are robust to variations in input prompts and generation configurations. To this end, we propose a reasoning score that evaluates reasoning traces along dimensions such as faithfulness, coherence, utility, and factuality. A remaining question is how to aggregate this score across multiple sampled traces. Naively averaging them is undesirable, particularly in long-horizon settings, where the number of possible trajectories grows rapidly, and low-confidence correct traces are more likely to be coincidental. To address this, we introduce the Filtered Reasoning Score (FRS), which computes reasoning quality using only the top-K% most confident traces. Evaluating with FRS, models that are indistinguishable under standard accuracy exhibit significant differences in reasoning quality. Moreover, models with higher FRS on one benchmark tend to perform better on other reasoning benchmarks, in both accuracy and reasoning quality. Together, these findings suggest that FRS complements accuracy by capturing a model’s transferable reasoning capabilities. We open source our evaluation codebase: https://github.com/Manas2006/benchmark_reproducibility.

Method: SD-Zero trains one model in two roles: Generator produces initial responses, Reviser conditions on those responses and binary rewards to produce improved responses. On-policy self-distillation transfers reviser knowledge back to generator using token distributions as supervision.

Result: On math and code reasoning benchmarks with Qwen3-4B-Instruct and Olmo-3-7B-Instruct, SD-Zero improves performance by at least 10% over base models and outperforms RFT, GRPO, and SDFT under same training budgets.

Conclusion: SD-Zero effectively transforms binary rewards into dense token-level self-supervision, demonstrating token-level self-localization and iterative self-evolution capabilities for improved reasoning performance.

Abstract: Current post-training methods in verifiable settings fall into two categories. Reinforcement learning (RLVR) relies on binary rewards, which are broadly applicable and powerful, but provide only sparse supervision during training. Distillation provides dense token-level supervision, typically obtained from an external teacher or using high-quality demonstrations. Collecting such supervision can be costly or unavailable. We propose Self-Distillation Zero (SD-Zero), a method that is substantially more training sample-efficient than RL and does not require an external teacher or high-quality demonstrations. SD-Zero trains a single model to play two roles: a Generator, which produces an initial response, and a Reviser, which conditions on that response and its binary reward to produce an improved response. We then perform on-policy self-distillation to distill the reviser into the generator, using the reviser’s token distributions conditioned on the generator’s response and its reward as supervision. In effect, SD-Zero trains the model to transform binary rewards into dense token-level self-supervision. On math and code reasoning benchmarks with Qwen3-4B-Instruct and Olmo-3-7B-Instruct, SD-Zero improves performance by at least 10% over the base models and outperforms strong baselines, including Rejection Fine-Tuning (RFT), GRPO, and Self-Distillation Fine-Tuning (SDFT), under the same question set and training sample budget. Extensive ablation studies show two novel characteristics of our proposed algorithm: (a) token-level self-localization, where the reviser can identify the key tokens that need to be revised in the generator’s response based on reward, and (b) iterative self-evolution, where the improving ability to revise answers can be distilled back into generation performance with regular teacher synchronization.

Method: The study evaluates various models including GPT-4o, BERT, and RoBERTa under zero-shot, one-shot, and few-shot settings on the ReCAM task (cloze-style passages with abstract concept questions). A bidirectional attention classifier is proposed, inspired by human cognitive strategies, which dynamically attends to both passages and options to enhance comprehension.

Result: Most LLMs (including GPT-4o) struggle with abstract meaning comprehension, while fine-tuned models like BERT and RoBERTa perform better. The proposed bidirectional attention classifier improves accuracy by 4.06% on Task 1 and 3.41% on Task 2 compared to baseline fine-tuned models.

Conclusion: Abstract meaning comprehension remains challenging for language models, with fine-tuned models outperforming LLMs. The bidirectional attention approach shows promise for enhancing abstract concept understanding by mimicking human cognitive strategies of dynamic attention between passages and options.

Abstract: Understanding abstract meanings is crucial for advanced language comprehension. Despite extensive research, abstract words remain challenging due to their non-concrete, high-level semantics. SemEval-2021 Task 4 (ReCAM) evaluates models’ ability to interpret abstract concepts by presenting passages with questions and five abstract options in a cloze-style format. Key findings include: (1) Most large language models (LLMs), including GPT-4o, struggle with abstract meaning comprehension under zero-shot, one-shot, and few-shot settings, while fine-tuned models like BERT and RoBERTa perform better. (2) A proposed bidirectional attention classifier, inspired by human cognitive strategies, enhances fine-tuned models by dynamically attending to passages and options. This approach improves accuracy by 4.06 percent on Task 1 and 3.41 percent on Task 2, demonstrating its potential for abstract meaning comprehension.

Method: Three main contributions: 1) A dynamic data curation pipeline that filters for genuinely retrieval-dependent samples to maintain benchmark difficulty over time; 2) VLM-DeflectionBench with 2,775 samples spanning diverse multimodal retrieval settings; 3) A fine-grained evaluation protocol with four scenarios that disentangle parametric memorization from retrieval robustness.

Result: Experiments across 20 state-of-the-art LVLMs show that models usually fail to deflect appropriately when faced with noisy or misleading evidence. The benchmark reveals significant gaps in how models handle uncertainty and conflicting multimodal information.

Conclusion: The work highlights the need to evaluate not just what models know, but how they behave when they don’t know. VLM-DeflectionBench serves as a reusable and extensible benchmark for reliable KB-VQA evaluation, addressing critical gaps in current multimodal retrieval assessment.

Abstract: Large Vision-Language Models (LVLMs) increasingly rely on retrieval to answer knowledge-intensive multimodal questions. Existing benchmarks overlook conflicts between visual and textual evidence and the importance of generating deflections (e.g., Sorry, I cannot answer…) when retrieved knowledge is incomplete. These benchmarks also suffer from rapid obsolescence, as growing LVLM training sets allow models to answer many questions without retrieval. We address these gaps with three contributions. First, we propose a dynamic data curation pipeline that preserves benchmark difficulty over time by filtering for genuinely retrieval-dependent samples. Second, we introduce VLM-DeflectionBench, a benchmark of 2,775 samples spanning diverse multimodal retrieval settings, designed to probe model behaviour under conflicting or insufficient evidence. Third, we define a fine-grained evaluation protocol with four scenarios that disentangle parametric memorization from retrieval robustness. Experiments across 20 state-of-the-art LVLMs indicate that models usually fail to deflect in the presence of noisy or misleading evidence. Our results highlight the need to evaluate not only what models know, but how they behave when they do not, and serve as a reusable and extensible benchmark for reliable KB-VQA evaluation. All resources will be publicly available upon publication.

Method: Introduces Claim-Aware Reasoning Protocol that structures outputs into atomic claims paired with explicit confidence estimates, then uses multi-stage training pipeline to align model confidence with claim correctness and optimize for factuality.

Result: Improves claim-level accuracy by up to 39.9% on Biography generation, with 16.0% increase in AUROC on FactBench, consistently outperforming supervised and RL baselines on four long-form factuality benchmarks.

Conclusion: CURE effectively improves long-form factuality by teaching LLMs to reason about uncertainty at the claim level, enabling better calibration and selective prediction for uncertain claims.

Abstract: Large language models (LLMs) often hallucinate in long-form generation. Existing approaches mainly improve factuality through post-hoc revision or reinforcement learning (RL) with correctness-based rewards, but they do not teach the model to estimate which parts of its generation are reliable. As a result, models may still state incorrect claims confidently in their responses. Recent advances in reasoning have significantly improved LLM performance, and have been leveraged to estimate confidence by incorporating calibration into RL objectives. However, existing approaches remain limited to a single scalar confidence for the entire response, which is insufficient for long-form generation where uncertainty varies across individual claims. To mitigate this problem, we propose CURE, a framework that improves long-form factuality by teaching LLMs to reason about uncertainty at the claim level. We first introduce a Claim-Aware Reasoning Protocol, which structures outputs into atomic claims paired with explicit confidence estimates. We then develop a multi-stage training pipeline that aligns model confidence with claims’ correctness and then optimizes on factuality. The resulting calibrated confidence further enables selective prediction, allowing the model to abstain from uncertain claims at inference time. Experiments on four long-form factuality benchmarks show that CURE consistently improves factual accuracy over competitive supervised and RL baselines, while maintaining factual recall. In particular, it improves claim-level accuracy by up to 39.9% on Biography generation. These gains are accompanied by improved calibration, as reflected by a 16.0% increase in AUROC on FactBench.

Method: Focuses on Question Answering as a specific language understanding task, using two benchmarks from the financial domain (including newly created TableQuest). Systematically examines multiple PDF parsers and chunking strategies with varied overlap, analyzing their synergies in preserving document structure and ensuring answer correctness.

Result: Results provide practical guidelines for building robust RAG pipelines for PDF understanding, showing how different parser and chunking strategy combinations affect performance on financial document QA tasks.

Conclusion: The study offers systematic insights into RAG system design for PDF processing, with practical recommendations for component selection and configuration to optimize PDF understanding performance.

Abstract: PDF files are primarily intended for human reading rather than automated processing. In addition, the heterogeneous content of PDFs, such as text, tables, and images, poses significant challenges for parsing and information extraction. To address these difficulties, both practitioners and researchers are increasingly developing new methods, including the promising Retrieval-Augmented Generation (RAG) systems to automated PDF processing. However, there is no comprehensive study investigating how different components and design choices affect the performance of a RAG system for understanding PDFs. In this paper, we propose such a study (1) by focusing on Question Answering, a specific language understanding task, and (2) by leveraging two benchmarks from the financial domain, including TableQuest, our newly generated, publicly available benchmark. We systematically examine multiple PDF parsers and chunking strategies (with varied overlap), along with their potential synergies in preserving document structure and ensuring answer correctness. Overall, our results offer practical guidelines for building robust RAG pipelines for PDF understanding.

Method: Two-phased validation framework: 1) organizes text into hierarchical classification structure (Themes, Stories, Clusters), 2) uses Signal-to-Noise Ratio to prioritize high-value semantic features, and incorporates scoring into Summary-of-Summaries architecture.

Result: wSSAS significantly improves clustering integrity and categorization accuracy across diverse datasets (Google Business reviews, Amazon Product reviews, Goodreads Book reviews), reduces categorization entropy, and provides reproducible LLM-based summaries.

Conclusion: wSSAS offers a high-precision, deterministic process for large-scale text categorization that addresses LLM limitations in enterprise analytics applications.

Abstract: The use of Large Language Models (LLMs) for reliable, enterprise-grade analytics such as text categorization is often hindered by the stochastic nature of attention mechanisms and sensitivity to noise that compromise their analytical precision and reproducibility. To address these technical frictions, this paper introduces the Weighted Syntactic and Semantic Context Assessment Summary (wSSAS), a deterministic framework designed to enforce data integrity on large-scale, chaotic datasets. We propose a two-phased validation framework that first organizes raw text into a hierarchical classification structure containing Themes, Stories, and Clusters. It then leverages a Signal-to-Noise Ratio (SNR) to prioritize high-value semantic features, ensuring the model’s attention remains focused on the most representative data points. By incorporating this scoring mechanism into a Summary-of-Summaries (SoS) architecture, the framework effectively isolates essential information and mitigates background noise during data aggregation. Experimental results using Gemini 2.0 Flash Lite across diverse datasets - including Google Business reviews, Amazon Product reviews, and Goodreads Book reviews - demonstrate that wSSAS significantly improves clustering integrity and categorization accuracy. Our findings indicate that wSSAS reduces categorization entropy and provides a reproducible pathway for improving LLM based summaries based on a high-precision, deterministic process for large-scale text categorization.

Method: LOSA (Locality-aware Sparse Attention) leverages the observation that between consecutive denoising steps, only a small fraction of active tokens exhibit significant hidden-state changes while most stable tokens remain nearly constant. It reuses cached prefix-attention results for stable tokens and applies sparse attention only to active tokens.

Result: LOSA preserves near-dense accuracy while significantly improving efficiency, achieving up to +9 points in average accuracy at aggressive sparsity levels while maintaining 1.54x lower attention density. It achieves up to 4.14x attention speedup on RTX A6000 GPUs.

Conclusion: LOSA effectively addresses the KV Inflation problem in block-wise DLMs through locality-aware sparse attention, substantially reducing the number of KV indices that must be loaded while maintaining accuracy and improving computational efficiency.

Abstract: Block-wise diffusion language models (DLMs) generate multiple tokens in any order, offering a promising alternative to the autoregressive decoding pipeline. However, they still remain bottlenecked by memory-bound attention in long-context scenarios. Naive sparse attention fails on DLMs due to a KV Inflation problem, where different queries select different prefix positions, making the union of accessed KV pages large. To address this, we observe that between consecutive denoising steps, only a small fraction of active tokens exhibit significant hidden-state changes, while the majority of stable tokens remain nearly constant. Based on this insight, we propose LOSA (Locality-aware Sparse Attention), which reuses cached prefix-attention results for stable tokens and applies sparse attention only to active tokens. This substantially shrinks the number of KV indices that must be loaded, yielding both higher speedup and higher accuracy. Across multiple block-wise DLMs and benchmarks, LOSA preserves near-dense accuracy while significantly improving efficiency, achieving up to +9 points in average accuracy at aggressive sparsity levels while maintaining 1.54x lower attention density. It also achieves up to 4.14x attention speedup on RTX A6000 GPUs, demonstrating the effectiveness of the proposed method.

Method: Proposes a unified black-box robustness evaluation framework using leave-one-out occlusion for token-level explanations. Measures explanation robustness with top-token flip rate under realistic perturbations (swap, deletion, shuffling, back-translation) at multiple severity levels. Conducts systematic cross-architecture comparison across 3 benchmark datasets and 6 models spanning encoder and decoder families (BERT, RoBERTa, Qwen 7B/14B, Llama 8B/70B; 64,800 cases).

Result: Decoder LLMs produce substantially more stable explanations than encoder baselines (73% lower flip rates on average). Stability improves with model scale (44% gain from 7B to 70B). Relates robustness improvements to inference cost, yielding a practical cost-robustness tradeoff curve for model selection.

Conclusion: Decoder LLMs offer more robust explanations than encoder models, with stability improving with scale. The framework provides practical guidance for model and explanation selection in compliance-sensitive applications, balancing robustness with inference costs.

Abstract: Robust explanations are increasingly required for user trust in enterprise NLP, yet pre-deployment validation is difficult in the common case of black-box deployment (API-only access) where representation-based explainers are infeasible and existing studies provide limited guidance on whether explanations remain stable under real user noise, especially when organizations migrate from encoder classifiers to decoder LLMs. To close this gap, we propose a unified black-box robustness evaluation framework for token-level explanations based on leave-one-out occlusion, and operationalize explanation robustness with top-token flip rate under realistic perturbations (swap, deletion, shuffling, and back-translation) at multiple severity levels. Using this protocol, we conduct a systematic cross-architecture comparison across three benchmark datasets and six models spanning encoder and decoder families (BERT, RoBERTa, Qwen 7B/14B, Llama 8B/70B; 64,800 cases). We find that decoder LLMs produce substantially more stable explanations than encoder baselines (73% lower flip rates on average), and that stability improves with model scale (44% gain from 7B to 70B). Finally, we relate robustness improvements to inference cost, yielding a practical cost-robustness tradeoff curve that supports model and explanation selection prior to deployment in compliance-sensitive applications.

Method: Generated synthetic datasets of simple interactions between expert and novice agents in a spatial navigation task. Used transformer models trained on different datasets (pedagogical interactions vs. expert-only demonstrations). Controlled paradigm allowed precise operationalization of key distinctions between interaction and solo expert behavior.

Result: Models trained on pedagogical interactions were more robust across various scenarios than models trained only on expert demonstrations. The ability to represent epistemically distinct agents led to expert-like behavior even when expert behavior was rarely observed.

Conclusion: Interaction data provides benefits beyond just more training data; pedagogical interactions and agent representation enable better generalization and robustness. The findings have implications for how multimodal AI systems should be trained with interaction data rather than just expert demonstrations.

Abstract: Work in cognitive science and artificial intelligence has suggested that exposing learning agents to traces of interaction between multiple individuals can improve performance in a variety of settings, yet it remains unknown which features of interactions contribute to this improvement. We examined the factors that support the effectiveness of interaction data, using a controlled paradigm that allowed us to precisely operationalize key distinctions between interaction and an expert acting alone. We generated synthetic datasets of simple interactions between an expert and a novice in a spatial navigation task, and then trained transformer models on those datasets, evaluating performance after exposure to different datasets. Our experiments showed that models trained on pedagogical interactions were more robust across a variety of scenarios compared to models trained only on expert demonstrations, and that having the ability to represent epistemically distinct agents led to expert-like behavior even when expert behavior was rarely observed.

Method: Large-scale empirical investigation across 51,955 API trials with 16 frontier models from 9 organizations, using 10 experiments that port and extend canonical IVE paradigms from human psychology research, analyzing effects of alignment training and prompting strategies.

Result: IVE is prevalent in LLMs with pooled effect size (d=0.223) approximately twice the human single-victim baseline; instruction-tuned models show extreme IVE while reasoning-specialized models invert it; standard CoT prompting nearly triples the effect size while only utilitarian CoT eliminates it.

Conclusion: LLMs exhibit amplified Identifiable Victim Effect compared to humans, with alignment training and prompting strategies significantly modulating this bias, raising concerns for ethical AI deployment in humanitarian decision-making contexts.

Abstract: The Identifiable Victim Effect (IVE) $-$ the tendency to allocate greater resources to a specific, narratively described victim than to a statistically characterized group facing equivalent hardship $-$ is one of the most robust findings in moral psychology and behavioural economics. As large language models (LLMs) assume consequential roles in humanitarian triage, automated grant evaluation, and content moderation, a critical question arises: do these systems inherit the affective irrationalities present in human moral reasoning? We present the first systematic, large-scale empirical investigation of the IVE in LLMs, comprising N=51,955 validated API trials across 16 frontier models spanning nine organizational lineages (Google, Anthropic, OpenAI, Meta, DeepSeek, xAI, Alibaba, IBM, and Moonshot). Using a suite of ten experiments $-$ porting and extending canonical paradigms from Small et al. (2007) and Kogut and Ritov (2005) $-$ we find that the IVE is prevalent but strongly modulated by alignment training. Instruction-tuned models exhibit extreme IVE (Cohen’s d up to 1.56), while reasoning-specialized models invert the effect (down to d=-0.85). The pooled effect (d=0.223, p=2e-6) is approximately twice the single-victim human meta-analytic baseline (d$\approx$0.10) reported by Lee and Feeley (2016) $-$ and likely exceeds the overall human pooled effect by a larger margin, given that the group-victim human effect is near zero. Standard Chain-of-Thought (CoT) prompting $-$ contrary to its role as a deliberative corrective $-$ nearly triples the IVE effect size (from d=0.15 to d=0.41), while only utilitarian CoT reliably eliminates it. We further document psychophysical numbing, perfect quantity neglect, and marginal in-group/out-group cultural bias, with implications for AI deployment in humanitarian and ethical decision-making contexts.

Method: Constructed longitudinal dataset of 412 human authors and 6,086 documents (2012-2024) across academic abstracts, blogs, and news. Compared to trajectories generated by three representative LLMs under standard and history-conditioned settings. Used drift and variance-based metrics over semantic, lexical, and cognitive-emotional representations.

Result: LLMs exhibit temporal flattening: greater lexical diversity but substantially reduced semantic and cognitive-emotional drift relative to humans. Temporal variability patterns alone achieve 94% accuracy and 98% ROC-AUC in distinguishing human from LLM trajectories. Flattening persists regardless of whether LLMs generate independently or with incremental history.

Conclusion: Current LLM deployment paradigms fundamentally lack the ability to reproduce authentic temporal structure in human writing, with implications for synthetic training data and longitudinal text modeling applications.

Abstract: Large language models (LLMs) are increasingly used in daily applications, from content generation to code writing, where each interaction treats the model as stateless, generating responses independently without memory. Yet human writing is inherently longitudinal: authors’ styles and cognitive states evolve across months and years. This raises a central question: can LLMs reproduce such temporal structure across extended time periods? We construct and publicly release a longitudinal dataset of 412 human authors and 6,086 documents spanning 2012–2024 across three domains (academic abstracts, blogs, news) and compare them to trajectories generated by three representative LLMs under standard and history-conditioned generation settings. Using drift and variance-based metrics over semantic, lexical, and cognitive-emotional representations, we find temporal flattening in LLM-generated text. LLMs produce greater lexical diversity but exhibit substantially reduced semantic and cognitive-emotional drift relative to humans. These differences are highly predictive: temporal variability patterns alone achieve 94% accuracy and 98% ROC-AUC in distinguishing human from LLM trajectories. Our results demonstrate that temporal flattening persists regardless of whether LLMs generate independently or with access to incremental history, revealing a fundamental property of current deployment paradigms. This gap has direct implications for applications requiring authentic temporal structure, such as synthetic training data and longitudinal text modeling.

Method: Introduces StableToken with a multi-branch architecture that processes audio in parallel, then merges representations through a powerful bit-wise voting mechanism to form a single, stable token sequence. Uses consensus-driven approach for stability.

Result: Sets new SOTA in token stability, drastically reducing Unit Edit Distance (UED) under diverse noise conditions. Foundational stability translates to downstream benefits, significantly improving robustness of SpeechLLMs on various tasks.

Conclusion: StableToken addresses fundamental flaws in semantic speech tokenizers by introducing consensus-driven stability, enabling more robust speech understanding systems and improving downstream LLM performance.

Abstract: Prevalent semantic speech tokenizers, designed to capture linguistic content, are surprisingly fragile. We find they are not robust to meaning-irrelevant acoustic perturbations; even at high Signal-to-Noise Ratios (SNRs) where speech is perfectly intelligible, their output token sequences can change drastically, increasing the learning burden for downstream LLMs. This instability stems from two flaws: a brittle single-path quantization architecture and a distant training signal indifferent to intermediate token stability. To address this, we introduce StableToken, a tokenizer that achieves stability through a consensus-driven mechanism. Its multi-branch architecture processes audio in parallel, and these representations are merged via a powerful bit-wise voting mechanism to form a single, stable token sequence. StableToken sets a new state-of-the-art in token stability, drastically reducing Unit Edit Distance (UED) under diverse noise conditions. This foundational stability translates directly to downstream benefits, significantly improving the robustness of SpeechLLMs on a variety of tasks. Our code and model are publicly available at https://github.com/Tencent/StableToken.

Method: Analyzed 4 models (Qwen3-VL-8B, Llama-3.2-11B, Llama-3.3-70B, Gemma-2-9B) with 300 prompts in a 14-level hierarchy at 3 temperatures. Measured 106 scalar metrics across up to 7 analysis passes, focusing on attention effective rank, variance kurtosis, and layer-wise SVD analysis.

Result: Self-reference alone isn’t destabilizing; grounded self-reference is stable. NCTR prompts cause significant disruption: elevated attention effective rank (d=3.14-3.52), 281/397 metric-model combinations differentiate NCTR from stable self-reference, per-layer SVD shows disruption at every layer, classifier achieves AUC 0.81-0.90, and NCTR prompts produce 34-56% more contradictory outputs.

Conclusion: Non-closing truth recursion forces transformers into dynamical regimes where classical matrix-semigroup problems concentrate, providing insights into self-referential failure modes with practical relevance for model reliability.

Abstract: We investigate how self-referential inputs alter the internal matrix dynamics of large language models. Measuring 106 scalar metrics across up to 7 analysis passes on four models from three architecture families – Qwen3-VL-8B, Llama-3.2-11B, Llama-3.3-70B, and Gemma-2-9B – over 300 prompts in a 14-level hierarchy at three temperatures ($T \in {0.0, 0.3, 0.7}$), we find that self-reference alone is not destabilizing: grounded self-referential statements and meta-cognitive prompts are markedly more stable than paradoxical self-reference on key collapse-related metrics, and on several such metrics can be as stable as factual controls. Instability concentrates in prompts inducing non-closing truth recursion (NCTR) – truth-value computations with no finite-depth resolution. NCTR prompts produce anomalously elevated attention effective rank – indicating attention reorganization with global dispersion rather than simple concentration collapse – and key metrics reach Cohen’s $d = 3.14$ (attention effective rank) to $3.52$ (variance kurtosis) vs. stable self-reference in the 70B model; 281/397 metric-model combinations differentiate NCTR from stable self-reference after FDR correction ($q < 0.05$), 198 with $|d| > 0.8$. Per-layer SVD confirms disruption at every sampled layer ($d > +1.0$ in all three models analyzed), ruling out aggregation artifacts. A classifier achieves AUC $0.81$-$0.90$; 30 minimal pairs yield 42/387 significant combinations; 43/106 metrics replicate across all four models. We connect these observations to three classical matrix-semigroup problems and propose, as a conjecture, that NCTR forces finite-depth transformers toward dynamical regimes where these problems concentrate. NCTR prompts also produce elevated contradictory output ($+34$-$56$ percentage points vs. controls), suggesting practical relevance for understanding self-referential failure modes.

Method: Proposes Unified Audio Schema (UAS), a holistic supervision framework that structures audio information into three explicit components (Transcription, Paralinguistics, Non-linguistic Events) in a unified JSON format to achieve comprehensive acoustic coverage while maintaining audio-text alignment.

Result: UAS-Audio boosts fine-grained perception by 10.9% on MMSU over same-size state-of-the-art models while preserving robust reasoning capabilities, validated on MMSU, MMAR, and MMAU benchmarks.

Conclusion: The structured supervision approach effectively addresses AudioLLMs’ perception-reasoning gap by providing explicit acoustic component modeling while maintaining reasoning capabilities through unified audio-text alignment.

Abstract: Recent Audio Large Language Models (AudioLLMs) exhibit a striking performance inversion: while excelling at complex reasoning tasks, they consistently underperform on fine-grained acoustic perception. We attribute this gap to a fundamental limitation of ASR-centric training, which provides precise linguistic targets but implicitly teaches models to suppress paralinguistic cues and acoustic events as noise. To address this, we propose Unified Audio Schema (UAS), a holistic and structured supervision framework that organizes audio information into three explicit components – Transcription, Paralinguistics, and Non-linguistic Events – within a unified JSON format. This design achieves comprehensive acoustic coverage without sacrificing the tight audio-text alignment that enables reasoning. We validate the effectiveness of this supervision strategy by applying it to both discrete and continuous AudioLLM architectures. Extensive experiments on MMSU, MMAR, and MMAU demonstrate that UAS-Audio yields consistent improvements, boosting fine-grained perception by 10.9% on MMSU over the same-size state-of-the-art models while preserving robust reasoning capabilities. Our code and model are publicly available at https://github.com/Tencent/Unified_Audio_Schema.

Method: Created AlphaEval benchmark with 94 tasks from 7 companies across 6 O*NET domains, evaluating complete agent products (not just models) using multiple paradigms: LLM-as-a-Judge, reference-driven metrics, formal verification, rubric-based assessment, and automated UI testing.

Result: Developed a production-grounded benchmark that captures performance variations invisible to model-level evaluation, along with a systematic requirement-to-benchmark construction framework for converting authentic production requirements into executable evaluation tasks.

Conclusion: AlphaEval provides a production-grounded evaluation framework that better reflects real-world AI agent deployment, with a reproducible methodology for organizations to create their own domain-specific benchmarks.

Abstract: The rapid deployment of AI agents in commercial settings has outpaced the development of evaluation methodologies that reflect production realities. Existing benchmarks measure agent capabilities through retrospectively curated tasks with well-specified requirements and deterministic metrics – conditions that diverge fundamentally from production environments where requirements contain implicit constraints, inputs are heterogeneous multi-modal documents with information fragmented across sources, tasks demand undeclared domain expertise, outputs are long-horizon professional deliverables, and success is judged by domain experts whose standards evolve over time. We present AlphaEval, a production-grounded benchmark of 94 tasks sourced from seven companies deploying AI agents in their core business, spanning six O*NET (Occupational Information Network) domains. Unlike model-centric benchmarks, AlphaEval evaluates complete agent products – Claude Code, Codex, etc. – as commercial systems, capturing performance variations invisible to model-level evaluation. Our evaluation framework covers multiple paradigms (LLM-as-a-Judge, reference-driven metrics, formal verification, rubric-based assessment, automated UI testing, etc.), with individual domains composing multiple paradigms. Beyond the benchmark itself, we contribute a requirement-to-benchmark construction framework – a systematic methodology that transforms authentic production requirements into executable evaluation tasks in minimal time. This framework standardizes the entire pipeline from requirement to evaluation, providing a reproducible, modular process that any organization can adopt to construct production-grounded benchmarks for their own domains.

Method: A modular agent-based framework using LLM agents to extract knowledge graphs, identify topics, build speaker personas, and simulate topic-guided conversations from unstructured conversations. Includes a QA module to generate memory-grounded Question Answer pairs from conversational histories.

Result: Created TopicGuidedChat (TGC) dataset with long-term memory encoded as speaker-specific knowledge graphs and short-term memory as topic-guided conversations. Framework yields higher conversational quality, and LLMs fine-tuned on TGC achieve improved performance on memory-grounded QA tasks.

Conclusion: AgenticAI-DialogGen provides an effective unsupervised approach to generate high-quality conversational datasets with proper memory encoding, addressing limitations in current conversational AI evaluation and training.

Abstract: Recent advancements in Large Language Models (LLMs) have improved their ability to process extended conversational contexts, yet fine-tuning and evaluating short- and long-term memories remain difficult due to the absence of datasets that encode both short- and long-term conversational history. Existing conversational datasets lack memory grounding, overlook topic continuity, or rely on costly human annotation. To address these gaps, we introduce AgenticAI-DialogGen, a modular agent-based framework that generates persona-grounded and topic-guided conversations without human supervision. The framework uses LLM agents to extract knowledge graphs, identify topics, build speaker personas, and simulate topic-guided conversations from unstructured conversations. A QA module generates memory-grounded Question Answer (QA) pairs drawn from short- and long-term conversational histories. We also generated a new dataset entitled, TopicGuidedChat (TGC), where long-term memory is encoded as speaker-specific knowledge graphs and short-term memory as newly generated topic-guided conversations. Evaluations depict that AgenticAI-DialogGen yields higher conversational quality and LLMs fine-tuned on TGC dataset achieve improved performance on memory-grounded QA tasks.

Method: Proposes MoshiRAG - modular approach combining compact full-duplex interface with selective retrieval. Uses asynchronous framework to identify knowledge-demanding queries and ground responses in external information. Leverages natural temporal gap between response onset and core information delivery to complete retrieval while maintaining conversation flow.

Result: Achieves factuality comparable to best publicly released non-duplex speech language models while preserving full-duplex interactivity. Flexible design supports plug-and-play retrieval without retraining and shows strong performance on out-of-domain mathematical reasoning tasks.

Conclusion: MoshiRAG successfully bridges the gap between real-time conversational interactivity and factual accuracy in speech-to-speech models through modular retrieval augmentation, enabling efficient knowledge access without compromising natural conversation flow.

Abstract: Speech-to-speech language models have recently emerged to enhance the naturalness of conversational AI. In particular, full-duplex models are distinguished by their real-time interactivity, including handling of pauses, interruptions, and backchannels. However, improving their factuality remains an open challenge. While scaling the model size could address this gap, it would make real-time inference prohibitively expensive. In this work, we propose MoshiRAG, a modular approach that combines a compact full-duplex interface with selective retrieval to access more powerful knowledge sources. Our asynchronous framework enables the model to identify knowledge-demanding queries and ground its responses in external information. By leveraging the natural temporal gap between response onset and the delivery of core information, the retrieval process can be completed while maintaining a natural conversation flow. With this approach, MoshiRAG achieves factuality comparable to the best publicly released non-duplex speech language models while preserving the interactivity inherent to full-duplex systems. Moreover, our flexible design supports plug-and-play retrieval methods without retraining and demonstrates strong performance on out-of-domain mathematical reasoning tasks.

Method: Proposes Order-Aware Knowledge Hypergraph RAG (OKH-RAG) that represents knowledge as higher-order interactions within a hypergraph augmented with precedence structure. Reformulates retrieval as sequence inference over hyperedges using a learned transition model that infers precedence from data without explicit temporal supervision.

Result: OKH-RAG consistently outperforms permutation-invariant baselines on order-sensitive question answering and explanation tasks (tropical cyclone and port operation scenarios). Ablations confirm gains specifically arise from modeling interaction order.

Conclusion: Effective reasoning requires not only retrieving relevant evidence but organizing it into structured sequences. Order-aware retrieval addresses a key limitation of set-based approaches and enables more coherent reasoning processes.

Abstract: Retrieval-augmented generation (RAG) enhances large language models by grounding outputs in retrieved knowledge. However, existing RAG methods including graph- and hypergraph-based approaches treat retrieved evidence as an unordered set, implicitly assuming permutation invariance. This assumption is misaligned with many real-world reasoning tasks, where outcomes depend not only on which interactions occur, but also on the order in which they unfold. We propose Order-Aware Knowledge Hypergraph RAG (OKH-RAG), which treats order as a first-class structural property. OKH-RAG represents knowledge as higher-order interactions within a hypergraph augmented with precedence structure, and reformulates retrieval as sequence inference over hyperedges. Instead of selecting independent facts, it recovers coherent interaction trajectories that reflect underlying reasoning processes. A learned transition model infers precedence directly from data without requiring explicit temporal supervision. We evaluate OKH-RAG on order-sensitive question answering and explanation tasks, including tropical cyclone and port operation scenarios. OKH-RAG consistently outperforms permutation-invariant baselines, and ablations show that these gains arise specifically from modeling interaction order. These results highlight a key limitation of set-based retrieval: effective reasoning requires not only retrieving relevant evidence, but organizing it into structured sequences.

Method: Cannot determine method due to missing abstract

Result: Cannot determine results due to missing abstract

Conclusion: Cannot determine conclusion due to missing abstract

Abstract: Failed to fetch summary for 2506.01256: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2506.01256&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: RCS computes a weighted Fréchet mean (semantic center) of answer embeddings and ranks candidates by their radial distance to this center. It supports multiple weighting schemes (uniform, frequency-based, probability-based) and works in black-box settings.

Result: Extensive experiments across seven benchmarks covering short-form QA and long-form reasoning tasks with five open-weight models show RCS variants consistently outperform strong baselines, with gains increasing as sampling budget grows. RCS also works well as a drop-in replacement for majority voting in multi-agent debate.

Conclusion: Geometric consensus provides a scalable and broadly applicable principle for reliable answer selection, extending beyond majority voting to more expressive and robust aggregation in LLM inference.

Abstract: Large language models (LLMs) frequently generate multiple candidate responses for a given prompt, yet selecting the most reliable one remains challenging, especially when correctness diverges from surface-level majority agreement. Existing approaches, such as self-consistency, rely on discrete voting, while probability-based methods often fail to capture relationships among candidate answers or tend to underweight high-quality but less frequent responses, and do not fully leverage the geometric structure of answer representations. To address these limitations, we introduce Radial Consensus Score (RCS), a simple, efficient, and training-free method for best-of-N selection. RCS models semantic consensus by computing a weighted Fréchet mean (semantic center) of answer embeddings and ranking candidates by their radial distance to this center. Importantly, RCS provides a general framework that supports multiple weighting schemes, including uniform, frequency-based, and probability-based variants, enabling flexible integration of agreement signals and model confidence while remaining fully applicable in black-box settings. Extensive experiments across seven benchmarks covering short-form QA and long-form reasoning tasks, and five open-weight models, demonstrate that RCS variants consistently outperform strong baselines, with gains becoming more pronounced as the sampling budget increases. RCS also serves as an effective drop-in replacement for majority voting in multi-agent debate and exhibits strong robustness in black-box scenarios. Overall, these results highlight geometric consensus as a scalable and broadly applicable principle for reliable answer selection, extending beyond majority voting to more expressive and robust aggregation in LLM inference.

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2507.09318: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.09318&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Semantic bootstrapping framework: LLM generates sub-intents for class labels, guides synthetic data creation through three-stage curriculum (seed, core, enriched). Non-Negated TM learns from examples to extract high-confidence literals as interpretable semantic cues, which are then injected into real data to align TM clause logic with LLM-inferred semantics.

Result: Improves interpretability and accuracy over vanilla TM across multiple text classification tasks, achieving performance comparable to BERT while remaining fully symbolic and efficient. No embeddings or runtime LLM calls required.

Conclusion: Successfully equips symbolic models with pretrained semantic priors, combining interpretability of symbolic models with semantic capacity of LLMs.

Abstract: Pretrained language models (PLMs) like BERT provide strong semantic representations but are costly and opaque, while symbolic models such as the Tsetlin Machine (TM) offer transparency but lack semantic generalization. We propose a semantic bootstrapping framework that transfers LLM knowledge into symbolic form, combining interpretability with semantic capacity. Given a class label, an LLM generates sub-intents that guide synthetic data creation through a three-stage curriculum (seed, core, enriched), expanding semantic diversity. A Non-Negated TM (NTM) learns from these examples to extract high-confidence literals as interpretable semantic cues. Injecting these cues into real data enables a TM to align clause logic with LLM-inferred semantics. Our method requires no embeddings or runtime LLM calls, yet equips symbolic models with pretrained semantic priors. Across multiple text classification tasks, it improves interpretability and accuracy over vanilla TM, achieving performance comparable to BERT while remaining fully symbolic and efficient.

Method: Cannot determine method due to missing paper content

Result: Cannot determine results due to missing paper content

Conclusion: Cannot determine conclusion due to missing paper content

Abstract: Failed to fetch summary for 2602.18899: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.18899&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: The algorithm lets LLMs leverage intermediate responses (thoughts) from past queries, filters meaningless/redundant thoughts, organizes them in thought memory, and retrieves relevant thoughts for new queries, creating a self-evolving long-term memory system.

Result: Thought-Retriever outperforms state-of-the-art baselines by at least 7.6% in F1 score and 16% in win rate across tasks, demonstrates self-evolution capability, and learns to leverage deeper thoughts for abstract queries.

Conclusion: The approach effectively equips LLM-based agents with scalable long-term memory that grows more capable through continuous interaction, overcoming context length limitations for external knowledge integration.

Abstract: Large language models (LLMs) have transformed AI research thanks to their powerful internal capabilities and knowledge. However, existing LLMs still fail to effectively incorporate the massive external knowledge when interacting with the world. Although retrieval-augmented LLMs are proposed to mitigate the issue, they are still fundamentally constrained by the context length of LLMs, as they can only retrieve top-K raw data chunks from the external knowledge base which often consists of millions of data chunks. Here we propose Thought-Retriever, a novel model-agnostic algorithm that helps LLMs generate output conditioned on arbitrarily long external data, without being constrained by the context length or number of retrieved data chunks. Our key insight is to let an LLM fully leverage its intermediate responses generated when solving past user queries (thoughts), filtering meaningless and redundant thoughts, organizing them in thought memory, and retrieving the relevant thoughts when addressing new queries. This effectively equips LLM-based agents with a self-evolving long-term memory that grows more capable through continuous interaction. Besides algorithmic innovation, we further meticulously prepare a novel benchmark, AcademicEval, which requires an LLM to faithfully leverage ultra-long context to answer queries based on real-world academic papers. Extensive experiments on AcademicEval and two other public datasets validate that Thought-Retriever remarkably outperforms state-of-the-art baselines, achieving an average increase of at least 7.6% in F1 score and 16% in win rate across various tasks. More importantly, we further demonstrate two exciting findings: (1) Thought-Retriever can indeed help LLM self-evolve after solving more user queries; (2) Thought-Retriever learns to leverage deeper thoughts to answer more abstract user queries.

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.09121: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.09121&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2502.11271: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2502.11271&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Continuous Knowledge Metabolism (CKM) processes literature through sliding time windows, incrementally updating a structured knowledge base. CKM-Lite is an efficient variant for incremental accumulation, while CKM-Full categorizes findings as novel/confirming/contradicting, detects knowledge change signals, and conditions hypothesis generation on evolution trajectories.

Result: CKM-Lite outperforms batch processing on hit rate (+2.8%), hypothesis yield (+3.6), and best-match alignment (+0.43) while reducing token cost by 92%. Analysis of 892 hypotheses shows: 1) incremental processing beats batch baselines; 2) change-aware instrumentation increases novelty but reduces coverage; 3) field trajectory stability correlates with hypothesis success; 4) knowledge convergence signals yield 5x higher hit rate than contradictions.

Conclusion: Hypothesis generation quality depends on both how much literature is processed and how it’s processed. Evaluation frameworks must account for quality-coverage trade-offs rather than optimizing single metrics. The character of generated hypotheses is shaped by knowledge evolution patterns.

Abstract: Scientific hypothesis generation requires tracking how knowledge evolves, not just what is currently known. We introduce Continuous Knowledge Metabolism (CKM), a framework that processes scientific literature through sliding time windows and incrementally updates a structured knowledge base as new findings arrive. We present CKM-Lite, an efficient variant that achieves strong predictive coverage through incremental accumulation, outperforming batch processing on hit rate (+2.8%, p=0.006), hypothesis yield (+3.6, p<0.001), and best-match alignment (+0.43, p<0.001) while reducing token cost by 92%. To understand what drives these differences, we develop CKM-Full, an instrumented variant that categorizes each new finding as novel, confirming, or contradicting, detects knowledge change signals, and conditions hypothesis generation on the full evolution trajectory. Analyzing 892 hypotheses generated by CKM-Full across 50 research topics, alongside parallel runs of the other variants, we report four empirical observations: (1) incremental processing outperforms batch baseline across predictive and efficiency metrics; (2) change-aware instrumentation is associated with higher LLM-judged novelty (Cohen’s d=3.46) but lower predictive coverage, revealing a quality-coverage trade-off; (3) a field’s trajectory stability is associated with hypothesis success (r=-0.28, p=0.051), suggesting boundary conditions for literature-based prediction; (4) knowledge convergence signals are associated with nearly 5x higher hit rate than contradiction signals, pointing to differential predictability across change types. These findings suggest that the character of generated hypotheses is shaped not only by how much literature is processed, but also by how it is processed. They further indicate that evaluation frameworks must account for the quality-coverage trade-off rather than optimize for a single metric.

Method: Proposes a self-draft framework that suppresses spurious confidence via layer-wise temperature annealing in early-exit decisions and adaptively bounds speculation length based on token-wise decoding difficulty. Uses reprocessing of hidden states in a unified parallel pass through deep layers.

Result: Achieves up to 2.33x wall-time speedup over standard autoregressive decoding across diverse long-form generation tasks and multiple model architectures, with no modifications to base LLM parameters.

Conclusion: The proposed method effectively addresses limitations of existing self-draft speculative decoding approaches, providing significant inference acceleration while maintaining exact output equivalence with original models.

Abstract: Speculative decoding has emerged as a promising approach to accelerate autoregressive inference in large language models (LLMs). Self-draft methods, which leverage the base LLM itself for speculation, avoid the overhead of auxiliary draft models but face limitations: shallow layers often produce overconfident yet incorrect token predictions, and the presence of difficult tokens in a draft sequence forces redundant computation through deeper layers, undermining both draft acceptance and overall speedup. To address these issues, we propose a novel self-draft framework that suppresses spurious confidence via layer-wise temperature annealing in early-exit decision and adaptively bounds speculation length based on token-wise decoding difficulty. By reprocessing the hidden states of draft tokens in a unified parallel pass through deep layers, our method maintains exact output equivalence with the original model while maximizing computational efficiency. It requires no modifications to the base LLM parameters and achieves up to 2.33x wall-time speedup over standard autoregressive decoding across diverse long-form generation tasks and multiple model architectures.

Method: Integrates LLMs with modular tools via Model Context Protocol (MCP), keeping databases secure on MCP server. Automates full pipeline: research planning, literature search, cohort construction, IRB documentation, Vibe ML, and report generation with human-in-the-loop refinement.

Result: Evaluated on three heterogeneous datasets: research plans and IRB documents finalized in 3-4 iterations; supported Vibe ML by exploring feature-model combinations and ranking top models; reports achieved 96% LLM evaluation coverage and 82% human evaluation coverage based on TRIPOD+AI framework.

Conclusion: CARIS demonstrates agentic AI can transform clinical hypotheses into executable research workflows across heterogeneous datasets, lowering barriers by eliminating coding and direct data access needs.

Abstract: Clinical research involves labor-intensive processes such as study design, cohort construction, model development, and documentation, requiring domain expertise, programming skills, and access to sensitive patient data. These demands create barriers for clinicians and external researchers conducting data-driven studies. To overcome these limitations, we developed a Clinical Agentic Research Intelligence System (CARIS) that automates the clinical research workflow while preserving data privacy, enabling comprehensive studies without direct access to raw data. CARIS integrates Large Language Models (LLMs) with modular tools via the Model Context Protocol (MCP), enabling natural language-driven orchestration of appropriate tools. Databases remain securely within the MCP server, and users access only the outputs and final research reports. Based on user intent, CARIS automatically executes the full pipeline: research planning, literature search, cohort construction, Institutional Review Board (IRB) documentation, Vibe Machine Learning (ML), and report generation, with iterative human-in-the-loop refinement. We evaluated CARIS on three heterogeneous datasets with distinct clinical tasks. Research plans and IRB documents were finalized within three to four iterations, using evidence from literature and data. The system supported Vibe ML by exploring feature-model combinations, ranking the top ten models, and generating performance visualizations. Final reports showed high completeness based on a checklist derived from the TRIPOD+AI framework, achieving 96% coverage in LLM evaluation and 82% in human evaluation. CARIS demonstrates that agentic AI can transform clinical hypotheses into executable research workflows across heterogeneous datasets. By eliminating the need for coding and direct data access, the system lowers barriers and bridges public and private clinical data environments.

Method: CascadeDebate inserts multi-agent deliberation directly at each tier’s escalation boundary. It uses confidence-based routers to activate lightweight agent ensembles only for uncertain cases, enabling consensus-driven resolution of ambiguities internally. The unified architecture alternates single-model inference with selective multi-agent deliberation across model scales, with human experts as the final fallback.

Result: Across five benchmarks spanning science, medicine, and general knowledge, CascadeDebate outperforms strong single-model cascades and standalone multi-agent systems by up to 26.75%. An online threshold optimizer boosts accuracy by 20.98 to 52.33% relative improvement over fixed policies and enables elastic adaptation to real-world distributions.

Conclusion: CascadeDebate effectively addresses the limitations of traditional cascaded LLM systems by introducing multi-agent deliberation at escalation boundaries, enabling dynamic scaling of test-time compute according to query difficulty while maintaining accuracy and cost efficiency.

Abstract: Cascaded LLM systems coordinate models of varying sizes with human experts to balance accuracy, cost, and abstention under uncertainty. However, single-model tiers at each stage often struggle with ambiguous queries, triggering premature escalations to costlier models or experts due to under-confidence and inefficient compute scaling. CascadeDebate addresses this gap by inserting multi-agent deliberation directly at each tier’s escalation boundary. Confidence-based routers activate lightweight agent ensembles only for uncertain cases, enabling consensus-driven resolution of ambiguities internally without invoking higher-cost upgrades. Our unified architecture alternates single-model inference with selective multi-agent deliberation across model scales, culminating in human experts as the final fallback. This design scales test-time compute dynamically according to query difficulty. Across five benchmarks spanning science, medicine, and general knowledge, CascadeDebate outperforms strong single-model cascades and standalone multi-agent systems by up to 26.75 percent. An online threshold optimizer proves essential, boosting accuracy by 20.98 to 52.33 percent relative improvement over fixed policies and enabling elastic adaptation to real-world distributions.

Method: Two-stage multi-agent framework: 1) Construct structural sketch and row/column summaries through incremental interpretation of localized regions using multiple modalities (code execution results, images, LaTeX tables), 2) Task-driven reasoning over intermediate representation with verification module for error reduction.

Result: Achieves 38.16% on Spreadsheet Bench with GPT-OSS-120B, outperforming ChatGPT Agent baseline (35.27%) by 2.89 absolute points. Demonstrates effectiveness on two spreadsheet datasets.

Conclusion: SpreadsheetAgent advances robust and scalable spreadsheet understanding by addressing layout awareness and scale limitations through multimodal incremental processing and verification mechanisms.

Abstract: Spreadsheets are central to real-world applications such as enterprise reporting, auditing, and scientific data management. Despite their ubiquity, existing large language model based approaches typically treat tables as plain text, overlooking critical layout cues and visual semantics. Moreover, real-world spreadsheets are often massive in scale, exceeding the input length that LLMs can efficiently process. To address these challenges, we propose SpreadsheetAgent, a two-stage multi-agent framework for spreadsheet understanding that adopts a step-by-step reading and reasoning paradigm. Instead of loading the entire spreadsheet at once, SpreadsheetAgent incrementally interprets localized regions through multiple modalities, including code execution results, images, and LaTeX tables. The method first constructs a structural sketch and row/column summaries, and then performs task-driven reasoning over this intermediate representation in the Solving Stage. To further enhance reliability, we design a verification module that validates extracted structures via targeted inspections, reducing error propagation and ensuring trustworthy inputs for downstream reasoning. Extensive experiments on two spreadsheet datasets demonstrate the effectiveness of our approach. With GPT-OSS-120B, SpreadsheetAgent achieves 38.16% on Spreadsheet Bench, outperforming the ChatGPT Agent baseline (35.27%) by 2.89 absolute points. These results highlight the potential of SpreadsheetAgent to advance robust and scalable spreadsheet understanding in real-world applications. Code is available at https://github.com/renhouxing/SpreadsheetAgent.git.

Method: ContextLens framework leverages LLMs to ground input context in legal domain, explicitly identifies known/unknown factors. Instead of direct safety assessment, uses crafted questions spanning applicability, general principles, and detailed provisions to assess compliance with pre-defined priorities/rules.

Result: Significantly improves LLMs’ compliance assessment on GDPR and EU AI Act benchmarks, surpasses existing baselines without training. Can identify ambiguous and missing factors in contexts.

Conclusion: ContextLens provides effective framework for contextualized safety/privacy compliance assessment by addressing real-world ambiguity and incompleteness through legal grounding and structured questioning.

Abstract: Individuals’ concerns about data privacy and AI safety are highly contextualized and extend beyond sensitive patterns. Addressing these issues requires reasoning about the context to identify and mitigate potential risks. Though researchers have widely explored using large language models (LLMs) as evaluators for contextualized safety and privacy assessments, these efforts typically assume the availability of complete and clear context, whereas real-world contexts tend to be ambiguous and incomplete. In this paper, we propose ContextLens, a semi-rule-based framework that leverages LLMs to ground the input context in the legal domain and explicitly identify both known and unknown factors for legal compliance. Instead of directly assessing safety outcomes, our ContextLens instructs LLMs to answer a set of crafted questions that span over applicability, general principles and detailed provisions to assess compliance with pre-defined priorities and rules. We conduct extensive experiments on existing compliance benchmarks that cover the General Data Protection Regulation (GDPR) and the EU AI Act. The results suggest that our ContextLens can significantly improve LLMs’ compliance assessment and surpass existing baselines without any training. Additionally, our ContextLens can further identify the ambiguous and missing factors.

Method: Developed CompliBench benchmark with scalable automated data generation pipeline that simulates user-agent interactions. Uses controllable flaw injection for precise ground-truth labels of violated guidelines and exact conversation turns, plus adversarial search to ensure challenging perturbations.

Result: Current state-of-the-art proprietary LLMs struggle significantly with detecting policy violations. A small-scale judge model fine-tuned on synthesized data outperforms leading LLMs and generalizes well to unseen business domains.

Conclusion: The automated pipeline provides effective foundation for training robust generative reward models, addressing the critical need for reliable LLM judges in enterprise policy compliance evaluation.

Abstract: As Large Language Models (LLMs) are increasingly deployed as task-oriented agents in enterprise environments, ensuring their strict adherence to complex, domain-specific operational guidelines is critical. While utilizing an LLM-as-a-Judge is a promising solution for scalable evaluation, the reliability of these judges in detecting specific policy violations remains largely unexplored. This gap is primarily due to the lack of a systematic data generation method, which has been hindered by the extensive cost of fine-grained human annotation and the difficulty of synthesizing realistic agent violations. In this paper, we introduce CompliBench, a novel benchmark designed to evaluate the ability of LLM judges to detect and localize guideline violations in multi-turn dialogues. To overcome data scarcity, we develop a scalable, automated data generation pipeline that simulates user-agent interactions. Our controllable flaw injection process automatically yields precise ground-truth labels for the violated guideline and the exact conversation turn, while an adversarial search method ensures these introduced perturbations are highly challenging. Our comprehensive evaluation reveals that current state-of-the-art proprietary LLMs struggle significantly with this task. In addition, we demonstrate that a small-scale judge model fine-tuned on our synthesized data outperforms leading LLMs and generalizes well to unseen business domains, highlighting our pipeline as an effective foundation for training robust generative reward models.

Method: Three-component approach: (1) CuSA refines encoder-derived saliency cues into fine-grained toxic spans with lightweight LLM guidance; (2) GCLoss uses gradient-constrained objective to concentrate token-level saliency on toxic evidence while suppressing irrelevant activations; (3) ARCL constructs sample-specific contrastive reasoning pairs to sharpen semantic boundaries between toxic and non-toxic content.

Result: ToxiTrace improves classification accuracy and toxic span extraction while preserving efficient encoder-based inference and producing more coherent, human-readable explanations.

Conclusion: The proposed method successfully addresses the explainability gap in Chinese toxic content detection by providing readable toxic evidence spans while maintaining classification performance.

Abstract: Existing Chinese toxic content detection methods mainly target sentence-level classification but often fail to provide readable and contiguous toxic evidence spans. We propose \textbf{ToxiTrace}, an explainability-oriented method for BERT-style encoders with three components: (1) \textbf{CuSA}, which refines encoder-derived saliency cues into fine-grained toxic spans with lightweight LLM guidance; (2) \textbf{GCLoss}, a gradient-constrained objective that concentrates token-level saliency on toxic evidence while suppressing irrelevant activations; and (3) \textbf{ARCL}, which constructs sample-specific contrastive reasoning pairs to sharpen the semantic boundary between toxic and non-toxic content. Experiments show that ToxiTrace improves classification accuracy and toxic span extraction while preserving efficient encoder-based inference and producing more coherent, human-readable explanations. We have released the model at https://huggingface.co/ArdLi/ToxiTrace.

Method: Trained correctness classifiers on question representations from both a model’s own hidden states and external models, comparing self-probes vs peer-model probes. Evaluated on standard benchmarks and disagreement subsets where models produce conflicting predictions. Analyzed layer-wise localization of privileged knowledge.

Result: No advantage found in standard evaluation (self-probes comparable to peer-model probes). On disagreement subsets, discovered domain-specific privileged knowledge: self-representations consistently outperform peer representations in factual knowledge tasks but show no advantage in math reasoning. Factual advantage emerges progressively from early-to-mid layers onward.

Conclusion: LLMs possess domain-specific privileged knowledge about answer correctness, with factual knowledge showing self-representation advantages consistent with model-specific memory retrieval, while math reasoning lacks such privileged access.

Abstract: Humans use introspection to evaluate their understanding through private internal states inaccessible to external observers. We investigate whether large language models possess similar privileged knowledge about answer correctness, information unavailable through external observation. We train correctness classifiers on question representations from both a model’s own hidden states and external models, testing whether self-representations provide a performance advantage. On standard evaluation, we find no advantage: self-probes perform comparably to peer-model probes. We hypothesize this is due to high inter-model agreement of answer correctness. To isolate genuine privileged knowledge, we evaluate on disagreement subsets, where models produce conflicting predictions. Here, we discover domain-specific privileged knowledge: self-representations consistently outperform peer representations in factual knowledge tasks, but show no advantage in math reasoning. We further localize this domain asymmetry across model layers, finding that the factual advantage emerges progressively from early-to-mid layers onward, consistent with model-specific memory retrieval, while math reasoning shows no consistent advantage at any depth.

Method: Proposes cooperative paging: replace evicted conversation segments with minimal keyword bookmarks (~8-24 tokens), and provide the model with a recall() tool to retrieve full content when needed. Evaluates different paging strategies (boundary policies, eviction policies) and bookmark generation methods.

Result: Cooperative paging achieves highest answer quality on LoCoMo benchmark (10 real multi-session conversations, 300+ turns), outperforming truncation, BM25, word-overlap retrieval, search-tool baseline, and full context across four models (GPT-4o-mini, DeepSeek-v3.2, Claude Haiku, GLM-5). Key findings: coarse fixed-size pages work best (96.7%), content-aware topic_shift collapses (56.7%), eviction policy is data-dependent, and bookmark discrimination remains a bottleneck (57% correct page selection).

Conclusion: Cooperative paging is an effective approach for managing long conversations in LLMs, but bookmark specificity is crucial for accurate recall. The method demonstrates practical value for real-world multi-session dialogues.

Abstract: When LLM conversations grow beyond the context window, old content must be evicted – but how does the model recover it when needed? We propose cooperative paging: evicted segments are replaced with minimal keyword bookmarks ([pN:keywords], ~8-24 tokens each), and the model is given a recall() tool to retrieve full content on demand. On the LoCoMo benchmark (10 real multi-session conversations, 300+ turns), cooperative paging achieves the highest answer quality among six methods – outperforming truncation, BM25, word-overlap retrieval, a search-tool baseline, and full context – on four models (GPT-4o-mini, DeepSeek-v3.2, Claude Haiku, GLM-5), confirmed by four independent LLM judges ($p=0.017$, paired bootstrap). We then study the paging design space with a 5x4 ablation over boundary strategies and eviction policies (3,176 synthetic probes, 1,600 LoCoMo probes). Key findings: (1) coarse fixed-size pages (fixed_20) reach 96.7% while content-aware topic_shift collapses to 56.7%; (2) eviction policy choice is data-dependent (FIFO best on synthetic, LFU on LoCoMo); (3) two bookmark generation strategies improve over the heuristic baseline (+4.4 and +8.7 E2E points); (4) the remaining bottleneck is bookmark discrimination – the model triggers recall() 96% of the time but selects the correct page only 57% when bookmarks are insufficiently distinctive. Keyword specificity alone accounts for a 25 percentage point accuracy difference.

Method: SCRIPT is a model-agnostic module that enhances subword embeddings with structural granularity from Jamo units. It doesn’t require architectural changes or additional pre-training, but rather injects subcharacter compositional knowledge into existing Korean PLMs by decomposing characters into their constituent Jamo components.

Result: SCRIPT enhances all baselines across various Korean natural language understanding and generation tasks. Linguistic analyses show it reshapes the embedding space to better capture grammatical regularities and semantically cohesive variations.

Conclusion: The proposed SCRIPT module effectively injects subcharacter compositional knowledge into Korean language models, improving performance on Korean NLP tasks while better capturing the linguistic structure of Korean through its featural writing system.

Abstract: Korean is a morphologically rich language with a featural writing system in which each character is systematically composed of subcharacter units known as Jamo. These subcharacters not only determine the visual structure of Korean but also encode frequent and linguistically meaningful morphophonological processes. However, most current Korean language models (LMs) are based on subword tokenization schemes, which are not explicitly designed to capture the internal compositional structure of characters. To address this limitation, we propose SCRIPT, a model-agnostic module that injects subcharacter compositional knowledge into Korean PLMs. SCRIPT allows to enhance subword embeddings with structural granularity, without requiring architectural changes or additional pre-training. As a result, SCRIPT enhances all baselines across various Korean natural language understanding (NLU) and generation (NLG) tasks. Moreover, beyond performance gains, detailed linguistic analyses show that SCRIPT reshapes the embedding space in a way that better captures grammatical regularities and semantically cohesive variations. Our code is available at https://github.com/SungHo3268/SCRIPT.

Method: Three contributions: (1) ReasonXL corpus with over 2M aligned reasoning traces across 5 European languages; (2) Two-stage adaptation pipeline: supervised fine-tuning followed by reinforcement learning with verifiable rewards; (3) Representational analysis of adaptation across model depth.

Result: Models adapted with this approach match or exceed baseline performance, maintain general knowledge, preserve cross-lingual transfer, and show RLVR achieves greater behavioral divergence with smaller parameter updates than SFT.

Conclusion: LLMs can be effectively adapted to reason entirely in target languages using parallel reasoning traces and a two-stage adaptation pipeline, with RLVR providing more efficient representational rerouting than SFT alone.

Abstract: Despite advances in multilingual capabilities, most large language models (LLMs) remain English-centric in their training and, crucially, in their production of reasoning traces. Even when tasked with non-English problems, these models predominantly reason in English, creating a fundamental mismatch for non-English usage scenarios. We address this disparity directly with three contributions. (i) We introduce ReasonXL, the first large-scale parallel corpus of cross-domain reasoning traces spanning five European languages (English, German, French, Italian, and Spanish), with over two million aligned samples per language, each comprising prompts, reasoning traces, and final outputs, enabling direct supervision of language-specific reasoning. (ii) Using ReasonXL, we demonstrate that LLMs can be adapted to reason entirely in a desired target language, using a simple two-stage pipeline of supervised fine-tuning (SFT) followed by reinforcement learning with verifiable rewards (RLVR). The resulting models match or exceed baseline performance, with minimal loss in general knowledge and broadly preserved cross-lingual transfer. (iii) We conduct an extensive representational analysis of the adaptation and find a clear functional division across model depth: early layers contain an activation bottleneck that causally determines language identity, while upper layers concentrate the weight and activation changes driven by adaptation. We further find that RLVR achieves greater behavioral divergence from the base model with smaller parameter updates than SFT, suggesting a more efficient representational rerouting despite much smaller weight updates.

Method: Proposes DialRouter which: 1) uses Monte Carlo Tree Search (MCTS) to explore dialogue branches from different LLM selections and collect high-reward trajectories, 2) learns a lightweight routing policy from search data augmented with retrieval-based future state approximation, enabling multi-turn routing without online search.

Result: Experiments on open-domain and domain-specific dialogue tasks across diverse candidate sets (open-source and closed-source LLMs) show DialRouter significantly outperforms single LLMs and existing routing baselines in task success rate, achieving superior performance-cost trade-off with cost-aware rewards.

Conclusion: DialRouter successfully addresses the challenge of multi-turn LLM routing by moving from myopic selection to long-horizon sequential routing, demonstrating effectiveness across diverse dialogue tasks and LLM combinations.

Abstract: Multi-turn dialogue is the predominant form of interaction with large language models (LLMs). While LLM routing is effective in single-turn settings, existing methods fail to maximize cumulative performance in multi-turn dialogue due to interaction dynamics and delayed rewards. To address this challenge, we move from myopic, single-turn selection to long-horizon sequential routing for multi-turn dialogue. Accordingly, we propose DialRouter, which first performs MCTS to explore dialogue branches induced by different LLM selections and collect trajectories with high cumulative rewards. DialRouter then learns a lightweight routing policy from search-derived data, augmented with retrieval-based future state approximation, enabling multi-turn routing without online search. Experiments on both open-domain and domain-specific dialogue tasks across diverse candidate sets of both open-source and closed-source LLMs demonstrate that DialRouter significantly outperforms single LLMs and existing routing baselines in task success rate, while achieving a superior performance-cost trade-off when combined with a cost-aware reward.

Method: Knowledge Coordinate Conditioning (KoCo) maps each document into a three-dimensional semantic coordinate that represents its position in knowledge space. These coordinates are prepended as textual prefixes during pre-training to give the model explicit contextual awareness.

Result: KoCo significantly enhances performance across 10 downstream tasks, accelerates pre-training convergence by approximately 30%, and helps models distinguish stable facts from noise, effectively mitigating hallucination in generated outputs.

Conclusion: Explicitly modeling knowledge coordinates provides LLMs with better contextual awareness, leading to improved performance, faster training, and reduced hallucinations by helping models understand the structural relationships between documents.

Abstract: Standard Large Language Model (LLM) pre-training typically treats corpora as flattened token sequences, often overlooking the real-world context that humans naturally rely on to contextualize information. To bridge this gap, we introduce Knowledge Coordinate Conditioning (KoCo), a simple method that maps every document into a three-dimensional semantic coordinate. By prepending these coordinates as textual prefixes for pre-training, we aim to equip the model with explicit contextual awareness to learn the documents within the real-world knowledge structure. Experiment results demonstrate that KoCo significantly enhances performance across 10 downstream tasks and accelerates pre-training convergence by approximately 30%. Furthermore, our analysis indicates that explicitly modeling knowledge coordinates helps the model distinguish stable facts from noise, effectively mitigating hallucination in generated outputs.

Method: Proposes a decoupled, two-step agentic architecture that separates orchestration from data analysis, leveraging progressive data discovery infused with domain expert knowledge

Result: Top-tier LLMs achieve up to 86% accuracy with high robustness across evaluation runs, demonstrating framework viability

Conclusion: The proposed architecture effectively addresses limitations of existing approaches for value stream map analysis by enabling intelligent data source selection and multi-hop reasoning

Abstract: Extracting actionable insights from complex value stream map simulations can be challenging, time-consuming, and error-prone. Recent advances in large language models offer new avenues to support users with this task. While existing approaches excel at processing raw data to gain information, they are structurally unfit to pick up on subtle situational differences needed to distinguish similar data sources in this domain. To address this issue, we propose a decoupled, two-step agentic architecture. By separating orchestration from data analysis, the system leverages progressive data discovery infused with domain expert knowledge. This architecture allows the orchestration to intelligently select data sources and perform multi-hop reasoning across data structures while maintaining a slim internal context. Results from multiple state-of-the-art large language models demonstrate the framework’s viability: with top-tier models achieving accuracies of up to 86% and demonstrating high robustness across evaluation runs.

Method: DeP (Decoding by Perturbation) uses controlled textual interventions during decoding. It applies multi-level textual perturbations to elicit latent language priors, leverages attention variance to enhance stable evidence regions while suppressing suspicious noise, and constructs interpretable prior drift direction using logits statistics to counteract probability biases from textual co-occurrences.

Result: Extensive experiments confirm DeP effectively reduces hallucinations and achieves superior performance across multiple benchmarks.

Conclusion: DeP provides a novel training-free approach to mitigate multimodal hallucinations by addressing the hypersensitivity of visual grounding to textual phrasing during decoding, offering better performance than existing methods.

Abstract: Multimodal Large Language Models frequently suffer from inference hallucinations, partially stemming from language priors dominating visual evidence. Existing training-free mitigation methods either perturb the visual representation and deviate from the natural image distribution, or enforce intrusive manipulations that compromise the model’s inherent generative fluency. We introduce a novel perspective that multimodal hallucination manifests as the hypersensitivity of visual grounding to textual phrasing during the decoding phase. Building on this insight, we propose Decoding by Perturbation (DeP), a training-free framework mitigating prior-induced hallucinations via controlled textual interventions. DeP employs a dynamic probe applying multi-level textual perturbations to elicit latent language priors. Leveraging attention variance, it enhances stable evidence regions while suppressing suspicious noise in the feature space. Furthermore, it constructs an interpretable prior drift direction using logits statistics to counteract probability biases from textual co-occurrences. Extensive experiments confirm DeP effectively reduces hallucinations and achieves superior performance across multiple benchmarks.

Method: GLeMM is created automatically from Wiktionary articles using identical design across all languages. It includes automatic annotation of morphological features and semantic descriptions for a significant subset of entries.

Result: A large-scale derivational resource covering German, English, Spanish, French, Italian, Polish, and Russian, enabling researchers to address questions about form-meaning relations and test computational methods for derivational morphology.

Conclusion: GLeMM provides a valuable resource for data-driven morphological research, addressing reproducibility issues and enabling systematic study of derivational morphology across multiple languages.

Abstract: In derivational morphology, what mechanisms govern the variation in form-meaning relations between words? The answers to this type of questions are typically based on intuition and on observations drawn from limited data, even when a wide range of languages is considered. Many of these studies are difficult to replicate and generalize. To address this issue, we present GLeMM, a new derivational resource designed for experimentation and data-driven description in morphology. GLeMM is characterized by (i) its large size, (ii) its extensive coverage (currently amounting to seven European languages, i.e., German, English, Spanish, French, Italian, Polish, Russian, (iii) its fully automated design, identical across all languages, (iv) the automatic annotation of morphological features on each entry, as well as (v) the encoding of semantic descriptions for a significant subset of these entries. It enables researchers to address difficult questions, such as the role of form and meaning in word-formation, and to develop and experimentally test computational methods that identify the structures of derivational morphology. The article describes how GLeMM is created using Wiktionary articles and presents various case studies illustrating possible applications of the resource.

Method: Proposes Latent-Condensed Attention (LCA) which directly condenses context within MLA’s latent space. The representation is disentangled into semantic latent vectors and positional keys, with semantic vectors aggregated via query-aware pooling and positional keys preserved via anchor selection.

Result: LCA achieves up to 2.5× prefilling speedup and 90% KV cache reduction at 128K context while maintaining competitive performance. The method is architecture-agnostic and extends to other attention mechanisms like GQA.

Conclusion: LCA provides an effective joint optimization approach for reducing both computational cost and KV cache in LLMs for long contexts, with theoretical guarantees and practical efficiency improvements.

Abstract: Large language models (LLMs) face significant challenges in processing long contexts due to the linear growth of the key-value (KV) cache and quadratic complexity of self-attention. Existing approaches address these bottlenecks separately: Multi-head Latent Attention (MLA) reduces the KV cache by projecting tokens into a low-dimensional latent space, while sparse attention reduces computation. However, sparse methods cannot operate natively on MLA’s compressed latent structure, missing opportunities for joint optimization. In this paper, we propose Latent-Condensed Attention (LCA), which directly condenses context within MLA’s latent space, where the representation is disentangled into semantic latent vectors and positional keys. LCA separately aggregates semantic vectors via query-aware pooling and preserves positional keys via anchor selection. This approach jointly reduces both computational cost and KV cache without adding parameters. Beyond MLA, LCA’s design is architecture-agnostic and readily extends to other attention mechanisms such as GQA. Theoretically, we prove a length-independent error bound. Experiments show LCA achieves up to 2.5$\times$ prefilling speedup and 90% KV cache reduction at 128K context while maintaining competitive performance.

Method: Systematically compare six elicitation task types across two commercial LLMs (GPT-4o Mini and Gemini 2.5 Flash) for Hausa and Fongbe, analyzing optimal prompting strategies for each language.

Result: GPT-4o Mini extracts 6-41 times more usable target-language words per API call than Gemini; optimal strategies differ by language: Hausa benefits from functional text and dialogue, while Fongbe requires constrained generation prompts.

Conclusion: Strategic prompting can effectively extract usable text data from commercial LLMs for low-resource languages, with performance varying significantly between models and requiring language-specific approaches.

Abstract: Large language models (LLMs) are trained on data contributed by low-resource language communities, yet the linguistic knowledge encoded in these models remains accessible only through commercial APIs. This paper investigates whether strategic prompting can extract usable text data from LLMs for two West African languages: Hausa (Afroasiatic, approximately 80 million speakers) and Fongbe (Niger-Congo, approximately 2 million speakers). We systematically compare six elicitation task types across two commercial LLMs (GPT-4o Mini and Gemini 2.5 Flash). GPT-4o Mini extracts 6-41 times more usable target-language words per API call than Gemini. Optimal strategies differ by language: Hausa benefits from functional text and dialogue, while Fongbe requires constrained generation prompts. We release all generated corpora and code.

Method: Introduces Preference-Paired Fine-Tuning (PFT) framework and Value Conflict Dilemma (VCD) dataset containing scenarios with conflicting human preferences for evaluation.

Result: PFT achieves 96.6% accuracy in multi-choice classification, highest open-ended generation score of 8.69, outperforms DPO, SFT and traditional methods, and shows 44.76% improvement in user-specific preference alignment with limited data.

Conclusion: PFT effectively handles contradictory and evolving individual preferences, enabling better personalization of LLMs to diverse user needs.

Abstract: Recent advances in large language models (LLMs) have significantly improved the alignment of models with general human preferences. However, a major challenge remains in adapting LLMs to individual preferences, which are not only diverse but also dynamic. In this paper, we introduce a novel framework, Preference-Paired Fine-Tuning (PFT), designed to align models with contradictory and evolving individual preferences. We present a new dataset, Value Conflict Dilemma (VCD), which includes scenarios that involve conflicting human preferences, facilitating the evaluation of our approach. Our experiments demonstrate that PFT outperforms single-preference training methods, achieving up to 96.6% accuracy in multi-choice classification tasks and the highest open-ended generation score of 8.69. PFT also shows significant improvements over DPO, SFT and some traditional training methods, especially when handling conflicting preferences. Additionally, with limited user history data, models can inferring preference vector rapidly, achieving a 44.76% improvement in user-specific preference alignment in comparison to single-preference models.

Method: KG-Reasoner uses Reinforcement Learning to train an LLM to internalize KG traversal, creating a unified “thinking” phase where the model dynamically explores reasoning paths and performs backtracking when needed, rather than using fixed pipeline steps.

Result: Experiments on eight multi-hop and knowledge-intensive reasoning benchmarks show KG-Reasoner achieves competitive or superior performance compared to state-of-the-art methods.

Conclusion: The end-to-end RL framework enables more flexible and coherent reasoning over KGs by integrating multi-step reasoning into a unified thinking process, addressing limitations of pipeline-based approaches.

Abstract: Large Language Models (LLMs) exhibit strong abilities in natural language understanding and generation, yet they struggle with knowledge-intensive reasoning. Structured Knowledge Graphs (KGs) provide an effective form of external knowledge representation and have been widely used to enhance performance in classical Knowledge Base Question Answering (KBQA) tasks. However, performing precise multi-hop reasoning over KGs for complex queries remains highly challenging. Most existing approaches decompose the reasoning process into a sequence of isolated steps executed through a fixed pipeline. While effective to some extent, such designs constrain reasoning flexibility and fragment the overall decision process, often leading to incoherence and the loss of critical intermediate information from earlier steps. In this paper, we introduce KG-Reasoner, an end-to-end framework that integrates multi-step reasoning into a unified “thinking” phase of a Reasoning LLM. Through Reinforcement Learning (RL), the LLM is trained to internalize the KG traversal process, enabling it to dynamically explore reasoning paths, and perform backtracking when necessary. Experiments on eight multi-hop and knowledge-intensive reasoning benchmarks demonstrate that KG-Reasoner achieves competitive or superior performance compared to the state-of-the-art methods. Codes are available at the repository: https://github.com/Wangshuaiia/KG-Reasoner.

Method: Compares five confidence estimation methods across five frontier LLMs and two tabular QA benchmarks. Proposes Multi-Format Agreement (MFA) which exploits deterministic serialization variations (Markdown, HTML, JSON, CSV) unique to structured data.

Result: All models show severe overconfidence (smooth ECE 0.35-0.64 vs 0.10-0.15 for textual QA). Self-evaluation methods achieve AUROC 0.42-0.76 while perturbation methods achieve 0.78-0.86. MFA reduces ECE by 44-63% and achieves mean AUROC 0.80.

Conclusion: LLMs are poorly calibrated for tabular QA. Perturbation methods outperform self-evaluation methods. MFA provides effective confidence estimation at lower cost and combines well with sampling methods.

Abstract: Large language models (LLMs) are increasingly deployed for tabular question answering, yet calibration on structured data is largely unstudied. This paper presents the first systematic comparison of five confidence estimation methods across five frontier LLMs and two tabular QA benchmarks. All models are severely overconfident (smooth ECE 0.35-0.64 versus 0.10-0.15 reported for textual QA). A consistent self-evaluation versus perturbation dichotomy replicates across both benchmarks and all four fully-covered models: self-evaluation methods (verbalized, P(True)) achieve AUROC 0.42-0.76, while perturbation methods (semantic entropy, self-consistency, and our Multi-Format Agreement) achieve AUROC 0.78-0.86. Per-model paired bootstrap tests reject the null at p<0.001 after Holm-Bonferroni correction, and a 3-seed check on GPT-4o-mini gives a per-seed standard deviation of only 0.006. The paper proposes Multi-Format Agreement (MFA), which exploits the lossless and deterministic serialization variation unique to structured data (Markdown, HTML, JSON, CSV) to estimate confidence at 20% lower API cost than sampling baselines. MFA reduces ECE by 44-63%, generalizes across all four models on TableBench (mean AUROC 0.80), and combines complementarily with sampling: an MFA + self-consistency ensemble lifts AUROC from 0.74 to 0.82. A secondary contribution, structure-aware recalibration, improves AUROC by +10 percentage points over standard post-hoc methods.

Method: Defined latent planning as internal representations that (1) cause specific future token generation and (2) shape preceding context. Studied Qwen-3 family (0.6B-14B) on simple planning tasks and rhyming couplets, analyzing feature representations and planning mechanisms.

Result: Latent planning ability increases with model scale. Larger models possess features representing planned words and cause appropriate context generation. On rhyming tasks, models often identify rhymes ahead of time but seldom plan far ahead. Planning can be elicited and increases with scale when steering toward planned words.

Conclusion: LLMs do exhibit latent planning capabilities that scale with model size, providing mechanistic evidence of how planning abilities develop in language models and offering a framework for measuring such planning.

Abstract: LLMs can perform seemingly planning-intensive tasks, like writing coherent stories or functioning code, without explicitly verbalizing a plan; however, the extent to which they implicitly plan is unknown. In this paper, we define latent planning as occurring when LLMs possess internal planning representations that (1) cause the generation of a specific future token or concept, and (2) shape preceding context to license said future token or concept. We study the Qwen-3 family (0.6B-14B) on simple planning tasks, finding that latent planning ability increases with scale. Models that plan possess features that represent a planned-for word like “accountant”, and cause them to output “an” rather than “a”; moreover, even the less-successful Qwen-3 4B-8B have nascent planning mechanisms. On the more complex task of completing rhyming couplets, we find that models often identify a rhyme ahead of time, but even large models seldom plan far ahead. However, we can elicit some planning that increases with scale when steering models towards planned words in prose. In sum, we offer a framework for measuring planning and mechanistic evidence of how models’ planning abilities grow with scale.

Method: Proposes a graph-based soft prompting framework that shifts reasoning from node-level path traversal to subgraph-level reasoning. Uses GNN to encode extracted structural subgraphs into soft prompts, enabling LLMs to reason over richer structural context. Introduces a two-stage paradigm: lightweight LLM first uses soft prompts to identify relevant entities/relations, then more powerful LLM generates evidence-aware answers.

Result: Achieves state-of-the-art performance on three out of four multi-hop KBQA benchmarks, demonstrating effectiveness in reducing sensitivity to missing edges.

Conclusion: The graph-based soft prompting framework effectively addresses KG incompleteness issues in multi-hop KBQA by enabling richer structural reasoning through GNN-encoded soft prompts.

Abstract: Large Language Models (LLMs) have shown remarkable capabilities across various tasks but remain prone to hallucinations in knowledge-intensive scenarios. Knowledge Base Question Answering (KBQA) mitigates this by grounding generation in Knowledge Graphs (KGs). However, most multi-hop KBQA methods rely on explicit edge traversal, making them fragile to KG incompleteness. In this paper, we proposed a novel graph-based soft prompting framework that shifts the reasoning paradigm from node-level path traversal to subgraph-level reasoning. Specifically, we employ a Graph Neural Network (GNN) to encode extracted structural subgraphs into soft prompts, enabling LLM to reason over richer structural context and identify relevant entities beyond immediate graph neighbors, thereby reducing sensitivity to missing edges. Furthermore, we introduce a two-stage paradigm that reduces computational cost while preserving good performance: a lightweight LLM first leverages the soft prompts to identify question-relevant entities and relations, followed by a more powerful LLM for evidence-aware answer generation. Experiments on four multi-hop KBQA benchmarks show that our approach achieves state-of-the-art performance on three of them, demonstrating its effectiveness. Code is available at the repository: https://github.com/Wangshuaiia/GraSP.

Method: Proposes Enhance-then-Balance Modality Collaboration (EBMC) framework with three components: 1) semantic disentanglement and cross-modal enhancement to strengthen weaker modalities, 2) Energy-guided Modality Coordination for implicit gradient rebalancing via differentiable equilibrium objective, and 3) Instance-aware Modality Trust Distillation for sample-level reliability estimation and adaptive fusion weight modulation.

Result: EBMC achieves state-of-the-art or competitive results on multimodal sentiment analysis benchmarks and maintains strong performance under missing-modality settings, demonstrating improved robustness.

Conclusion: The EBMC framework effectively addresses modality imbalance in multimodal sentiment analysis by enhancing weaker modalities and balancing contributions, leading to improved performance and robustness across various conditions.

Abstract: Multimodal sentiment analysis (MSA) integrates heterogeneous text, audio, and visual signals to infer human emotions. While recent approaches leverage cross-modal complementarity, they often struggle to fully utilize weaker modalities. In practice, dominant modalities tend to overshadow non-verbal ones, inducing modality competition and limiting overall contributions. This imbalance degrades fusion performance and robustness under noisy or missing modalities. To address this, we propose a novel model, Enhance-then-Balance Modality Collaboration framework (EBMC). EBMC improves representation quality via semantic disentanglement and cross-modal enhancement, strengthening weaker modalities. To prevent dominant modalities from overwhelming others, an Energy-guided Modality Coordination mechanism achieves implicit gradient rebalancing via a differentiable equilibrium objective. Furthermore, Instance-aware Modality Trust Distillation estimates sample-level reliability to adaptively modulate fusion weights, ensuring robustness. Extensive experiments demonstrate that EBMC achieves state-of-the-art or competitive results and maintains strong performance under missing-modality settings.

Method: Evaluated two data augmentation methods: LLM-based generation using Gemini 2.5 Flash and back-translation using NLLB-200. Tested on Hausa and Fongbe languages using MasakhaNER 2.0 and MasakhaPOS benchmarks for named entity recognition and part-of-speech tagging tasks.

Result: Augmentation effectiveness was task-dependent rather than language or LLM quality dependent. For NER, neither method improved over baseline (LLM reduced Hausa NER by 0.24% F1, Fongbe by 1.81% F1). For POS tagging, LLM improved Fongbe by 0.33% accuracy, back-translation improved Hausa by 0.17%, while back-translation reduced Fongbe POS by 0.35%.

Conclusion: Data augmentation should be treated as a task-specific intervention rather than a universally beneficial preprocessing step, challenging assumptions that LLM generation quality predicts augmentation success.

Abstract: Data scarcity limits NLP development for low-resource African languages. We evaluate two data augmentation methods – LLM-based generation (Gemini 2.5 Flash) and back-translation (NLLB-200) – for Hausa and Fongbe, two West African languages that differ substantially in LLM generation quality. We assess augmentation on named entity recognition (NER) and part-of-speech (POS) tagging using MasakhaNER 2.0 and MasakhaPOS benchmarks. Our results reveal that augmentation effectiveness depends on task type rather than language or LLM quality alone. For NER, neither method improves over baseline for either language; LLM augmentation reduces Hausa NER by 0.24% F1 and Fongbe NER by 1.81% F1. For POS tagging, LLM augmentation improves Fongbe by 0.33% accuracy, while back-translation improves Hausa by 0.17%; back-translation reduces Fongbe POS by 0.35% and has negligible effect on Hausa POS. The same LLM-generated synthetic data produces opposite effects across tasks for Fongbe – hurting NER while helping POS – suggesting task structure governs augmentation outcomes more than synthetic data quality. These findings challenge the assumption that LLM generation quality predicts augmentation success, and provide actionable guidance: data augmentation should be treated as a task-specific intervention rather than a universally beneficial preprocessing step.

Method: Two-stage hierarchical framework: discrete facts are anchored in shallow layers first, followed by minimal updates to deeper layers to produce coherent text, reflecting the unidirectional Transformer flow.

Result: FABLE substantially improves fine-grained question answering while maintaining state-of-the-art holistic editing performance, as evaluated on the UnFine benchmark with fact-level metrics.

Conclusion: The decoupling approach resolves the mismatch between holistic recall and fine-grained fact access, enabling more reliable model editing with better fact retrieval capabilities.

Abstract: Unstructured model editing aims to update models with real-world text, yet existing methods often memorize text holistically without reliable fine-grained fact access. To address this, we propose FABLE, a hierarchical framework that decouples fine-grained fact injection from holistic text generation. FABLE follows a two-stage, fact-first strategy: discrete facts are anchored in shallow layers, followed by minimal updates to deeper layers to produce coherent text. This decoupling resolves the mismatch between holistic recall and fine-grained fact access, reflecting the unidirectional Transformer flow in which surface-form generation amplifies rather than corrects underlying fact representations. We also introduce UnFine, a diagnostic benchmark with fine-grained question-answer pairs and fact-level metrics for systematic evaluation. Experiments show that FABLE substantially improves fine-grained question answering while maintaining state-of-the-art holistic editing performance. Our code is publicly available at https://github.com/caskcsg/FABLE.

Method: Transforms external knowledge into standardized structured triplets (Condition, Proof, Conclusion) using lightweight prompt-based adaptation with frozen model parameters. Uses triplet head Condition as semantic anchor for retrieval and matching, enabling precise identification of query-relevant knowledge without concatenating lengthy raw texts.

Result: Significantly improves retrieval quality and reasoning efficiency across multiple benchmark datasets, produces more stable generation behavior, and enables more efficient resource utilization in complex reasoning scenarios.

Conclusion: Tri-RAG achieves favorable balance between retrieval accuracy and context token efficiency through structured triplet-based retrieval and reasoning-aligned context construction.

Abstract: Retrieval-Augmented Generation (RAG) mitigates hallucination in large language models (LLMs) by incorporating external knowledge during generation. However, the effectiveness of RAG depends not only on the design of the retriever and the capacity of the underlying model, but also on how retrieved evidence is structured and aligned with the query. Existing RAG approaches typically retrieve and concatenate unstructured text fragments as context, which often introduces redundant or weakly relevant information. This practice leads to excessive context accumulation, reduced semantic alignment, and fragmented reasoning chains, thereby degrading generation quality while increasing token consumption. To address these challenges, we propose Tri-RAG, a structured triplet-based retrieval framework that improves retrieval efficiency through reasoning-aligned context construction. Tri-RAG automatically transforms external knowledge from natural language into standardized structured triplets consisting of Condition, Proof, and Conclusion, explicitly capturing logical relations among knowledge fragments using lightweight prompt-based adaptation with frozen model parameters. Building on this representation, the triplet head Condition is treated as an explicit semantic anchor for retrieval and matching, enabling precise identification of query-relevant knowledge units without directly concatenating lengthy raw texts. As a result, Tri-RAG achieves a favorable balance between retrieval accuracy and context token efficiency. Experimental results across multiple benchmark datasets demonstrate that Tri-RAG significantly improves retrieval quality and reasoning efficiency, while producing more stable generation behavior and more efficient resource utilization in complex reasoning scenarios.

Method: Constructed synthetic training corpus of over 1M multi-label samples across 23 languages using culturally-adapted generation and programmatic quality filtering. Trained and compared six multilingual transformer encoders from DistilBERT to XLM-R-Large under identical conditions.

Result: XLM-R-Large achieved 0.868 F1-micro and 0.987 AUC-micro on in-domain test set. Zero-shot evaluation on human-annotated datasets showed XLM-R-Large matches or exceeds English-only specialist models on threshold-free ranking metrics while natively supporting all 23 languages.

Conclusion: Successfully created large-scale multilingual emotion classification system that performs competitively with English-only models while supporting 23 languages, addressing data scarcity in this domain.

Abstract: Emotion classification in multilingual settings remains constrained by the scarcity of annotated data: existing corpora are predominantly English, single-label, and cover few languages. We address this gap by constructing a large-scale synthetic training corpus of over 1M multi-label samples (50k per language) across 23 languages: Arabic, Bengali, Dutch, English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Mandarin, Polish, Portuguese, Punjabi, Russian, Spanish, Swahili, Tamil, Turkish, Ukrainian, Urdu, and Vietnamese, covering 11 emotion categories using culturally-adapted generation and programmatic quality filtering. We train and compare six multilingual transformer encoders, from DistilBERT (135M parameters) to XLM-R-Large (560M parameters), under identical conditions. On our in-domain test set, XLM-R-Large achieves 0.868 F1-micro and 0.987 AUC-micro. To validate against human-annotated data, we evaluate all models zero-shot on GoEmotions (English) and SemEval-2018 Task 1 E-c (English, Arabic, Spanish). On threshold-free ranking metrics, XLM-R-Large matches or exceeds English-only specialist models, tying on AP-micro (0.636) and LRAP (0.804) while surpassing on AUC-micro (0.810 vs. 0.787), while natively supporting all 23 languages. The best base-sized model is publicly available at https://huggingface.co/tabularisai/multilingual-emotion-classification

Method: RALP reformulates link prediction as a prompt learning problem, learning string-based chain-of-thought prompts as scoring functions for triples. It uses Bayesian Optimization through the MIPRO algorithm to identify effective prompts from fewer than 30 training examples without gradient access. At inference, it predicts missing entities, relations, or whole triples and assigns confidence scores based on learned prompts.

Result: RALP improves state-of-the-art KGE models by over 5% MRR across datasets and enhances generalization via high-quality inferred triples. On OWL reasoning tasks with complex class expressions, it achieves over 88% Jaccard similarity.

Conclusion: Prompt-based LLM reasoning serves as a flexible alternative to embedding-based methods for knowledge graph completion, demonstrating strong performance on both transductive tasks and complex OWL reasoning with minimal training data.

Abstract: Knowledge graph embedding (KGE) models perform well on link prediction but struggle with unseen entities, relations, and especially literals, limiting their use in dynamic, heterogeneous graphs. In contrast, pretrained large language models (LLMs) generalize effectively through prompting. We reformulate link prediction as a prompt learning problem and introduce RALP, which learns string-based chain-of-thought (CoT) prompts as scoring functions for triples. Using Bayesian Optimization through MIPRO algorithm, RALP identifies effective prompts from fewer than 30 training examples without gradient access. At inference, RALP predicts missing entities, relations or whole triples and assigns confidence scores based on the learned prompt. We evaluate on transductive, numerical, and OWL instance retrieval benchmarks. RALP improves state-of-the-art KGE models by over 5% MRR across datasets and enhances generalization via high-quality inferred triples. On OWL reasoning tasks with complex class expressions (e.g., $\exists hasChild.Female$, $\geq 5 ; hasChild.Female$), it achieves over 88% Jaccard similarity. These results highlight prompt-based LLM reasoning as a flexible alternative to embedding-based methods. We release our implementation, training, and evaluation pipeline as open source: https://github.com/dice-group/RALP .

Method: LLM-based approach that generates 5P-aligned causal graphs from patient-therapist dialogues. Evaluated using 46 psychotherapy intake transcripts annotated by clinical experts, comparing LLM-generated graphs against human formulations using structural (NetSimile), semantic (embedding similarity), and expert-rated clinical criteria.

Result: Generated graphs show structural similarity comparable to inter-annotator agreement and high semantic alignment with human graphs. Expert evaluations rate outputs as moderately complete, consistent, and clinically useful. LLM graphs tend to form more interconnected structures compared to chain-like human patterns, but overall complexity and content coverage are similar.

Conclusion: LLMs can produce clinically meaningful case formulation graphs within the natural variability of expert practice. InsightFlow highlights potential for automated causal modeling to augment clinical workflows, with future work needed to improve temporal reasoning and reduce redundancy.

Abstract: Clinical case formulation organizes patient symptoms and psychosocial factors into causal models, often using the 5P framework. However, constructing such graphs from therapy transcripts is time consuming and varies across clinicians. We present InsightFlow, an LLM based approach that automatically generates 5P aligned causal graphs from patient-therapist dialogues. Using 46 psychotherapy intake transcripts annotated by clinical experts, we evaluate LLM generated graphs against human formulations using structural (NetSimile), semantic (embedding similarity), and expert rated clinical criteria. The generated graphs show structural similarity comparable to inter annotator agreement and high semantic alignment with human graphs. Expert evaluations rate the outputs as moderately complete, consistent, and clinically useful. While LLM graphs tend to form more interconnected structures compared to the chain like patterns of human graphs, overall complexity and content coverage are similar. These results suggest that LLMs can produce clinically meaningful case formulation graphs within the natural variability of expert practice. InsightFlow highlights the potential of automated causal modeling to augment clinical workflows, with future work needed to improve temporal reasoning and reduce redundancy.

Method: TEPO uses (1) sequence-level likelihood to link group-level rewards with individual tokens via token-level aggregation, and (2) a token-level KL-Divergence mask constraint targeting tokens with positive advantages and decreasing entropy to mitigate abrupt policy updates.

Result: TEPO achieves state-of-the-art performance on mathematical reasoning benchmarks and enhances training stability, reducing convergence time by 50% compared with GRPO/DAPO.

Conclusion: TEPO effectively addresses token-level sparse reward challenges in CoT reasoning, improving both performance and training efficiency for mathematical reasoning tasks.

Abstract: Group Relative Policy Optimization (GRPO) has significantly advanced the reasoning ability of large language models (LLMs), particularly in their mathemat ical reasoning performance. However, GRPO and related entropy regularization methods still struggle with token-level sparse-rewards, which is an inherent chal lenge in chain-of-thought (CoT) reasoning. These approaches often rely on undifferen tiated token-level entropy regularization, which easily leads to entropy collapse or model degradation under sparse token rewards. In this work, we propose TEPO, a novel token-level framework that (1) leverages sequence-level likelihood to link group-level rewards with individual tokens via token-level aggregation, and (2) introduces a token-level KL-Divergence mask constraint that targets tokens with positive advantages and decreasing entropy to mitigate abrupt policy updates. Experiments demonstrate that TEPO not only achieves state-of-the-art performance on mathematical reasoning benchmarks but also markedly enhances training stability, reducing convergence time by 50% compared with GRPO/DAPO.

Method: Uses general tagset and thorough annotation guidelines to collect standardized, cross-lingual named entity span annotations through community collaboration (organizers, annotators, collaborators).

Result: Released UNER v1 in 2024, with continued expansion and active community involvement in building multilingual NER benchmarks.

Conclusion: The UNER project addresses the critical need for standardized multilingual evaluation benchmarks to properly assess and improve multilingual language models across diverse languages.

Abstract: While multilingual language models promise to bring the benefits of LLMs to speakers of many languages, gold-standard evaluation benchmarks in most languages to interrogate these assumptions remain scarce. The Universal NER project, now entering its fourth year, is dedicated to building gold-standard multilingual Named Entity Recognition (NER) benchmark datasets. Inspired by existing massively multilingual efforts for other core NLP tasks (e.g., Universal Dependencies), the project uses a general tagset and thorough annotation guidelines to collect standardized, cross-lingual annotations of named entity spans. The first installment (UNER v1) was released in 2024, and the project has continued and expanded since then, with various organizers, annotators, and collaborators in an active community.

Method: 1) Investigate criteria for effective CoT traces to mitigate causal hallucination, 2) Design a pipeline to generate CoT traces meeting these criteria, 3) Introduce Causal Hallucination Rate (CHR) metric to quantify hallucination and guide CoT criteria formulation.

Result: Fine-tuning with generated CoT traces substantially reduces causal hallucination in smaller LLMs while improving mean accuracy. Models show strong cross-dataset/difficulty generalization and robustness under misleading intervention prompts.

Conclusion: The proposed pipeline effectively generates CoT traces that mitigate causal hallucination in smaller LLMs for ECI tasks, with the CHR metric providing valuable quantification and guidance for this problem.

Abstract: Although large language models (LLMs) excel in complex reasoning tasks, they suffer from severe causal hallucination in event causality identification (ECI), particularly in smaller models ($\leq$1.5B parameters). A promising approach to address this issue is to fine-tune them with Chain-of-Thought (CoT) traces. However, there is currently a lack of CoT trace dataset available for ECI. In this paper, we first investigate the essential criteria that effective CoT traces should possess to mitigate causal hallucination in smaller models. We then design a pipeline to generate CoT traces that meet these criteria. Moreover, since there is currently no metric for quantifying causal hallucination, we also introduce a new metric, the Causal Hallucination Rate (CHR), to quantify causal hallucination, guide the formulation of effective CoT trace criteria, and validate the effectiveness of our pipeline. Our experiments show that fine-tuning with the CoT traces generated by our pipeline not only substantially reduces causal hallucination in smaller LLMs but also improves mean accuracy. Moreover, the fine-tuned models exhibit strong cross-dataset and cross-difficulty generalization, as well as robustness under misleading intervention prompts.

Method: 1) Structures knowledge documents hierarchically preserving semantic relationships from coarse-grained topics to fine-grained details. 2) Uses LLM agent to actively navigate records, iteratively identifying information gaps and retrieving relevant content from appropriate granularity levels.

Result: Extensive experiments on long-document QA benchmarks show NaviRAG consistently improves both retrieval recall and end-to-end answer performance over conventional RAG baselines. Ablation studies confirm gains from multi-granular evidence localization and dynamic retrieval planning.

Conclusion: NaviRAG makes RAG systems more intelligent and autonomous by shifting from passive segment retrieval to active knowledge navigation across hierarchical structures.

Abstract: Retrieval-augmented generation (RAG) typically relies on a flat retrieval paradigm that maps queries directly to static, isolated text segments. This approach struggles with more complex tasks that require the conditional retrieval and dynamic synthesis of information across different levels of granularity (e.g., from broad concepts to specific evidence). To bridge this gap, we introduce NaviRAG, a novel framework that shifts from passive segment retrieval to active knowledge navigation. NaviRAG first structures the knowledge documents into a hierarchical form, preserving semantic relationships from coarse-grained topics to fine-grained details. Leveraging this reorganized knowledge records, a large language model (LLM) agent actively navigates the records, iteratively identifying information gaps and retrieving relevant content from the most appropriate granularity level. Extensive experiments on long-document QA benchmarks show that NaviRAG consistently improves both retrieval recall and end-to-end answer performance over conventional RAG baselines. Ablation studies confirm performance gains stem from our method’s capacity for multi-granular evidence localization and dynamic retrieval planning. We further discuss efficiency, applicable scenario, and future directions of our method, hoping to make RAG systems more intelligent and autonomous.

Method: Reinforcement learning approach using group relative policy optimization with a multi-component reward function that jointly optimizes edit-level semantic similarity, fluency, pattern conformity, and argument-level appropriateness.

Result: Outperforms competitive baselines and state-of-the-art in human-like editing, with multi-round editing achieving appropriateness close to full rewriting in both automatic and human evaluations.

Conclusion: The approach successfully teaches LLMs human-like editing strategies, producing self-contained sentence-level edit suggestions that can be accepted or rejected independently, significantly improving argument appropriateness.

Abstract: Editing human-written text has become a standard use case of large language models (LLMs), for example, to make one’s arguments more appropriate for a discussion. Comparing human to LLM-generated edits, however, we observe a mismatch in editing strategies: While LLMs often perform multiple scattered edits and tend to change meaning notably, humans rather encapsulate dependent changes in self-contained, meaning-preserving edits. In this paper, we present a reinforcement learning approach that teaches LLMs human-like editing to improve the appropriateness of arguments. Our approach produces self-contained sentence-level edit suggestions that can be accepted or rejected independently. We train the approach using group relative policy optimization with a multi-component reward function that jointly optimizes edit-level semantic similarity, fluency, and pattern conformity as well as argument-level appropriateness. In automatic and human evaluation, it outperforms competitive baselines and the state of the art in human-like editing, with multi-round editing achieving appropriateness close to full rewriting.

Method: Proposes EvoSpark framework with three key components: 1) Stratified Narrative Memory using Role Socio-Evolutionary Base for dynamic experience metabolism, 2) Generative Mise-en-Scène mechanism for Role-Location-Plot alignment, and 3) Unified Narrative Operation Engine with Emergent Character Grounding Protocol to transform stochastic sparking into persistent characters.

Result: EvoSpark significantly outperforms baselines across diverse paradigms, enabling sustained generation of expressive and coherent narrative experiences in long-horizon simulations.

Conclusion: EvoSpark successfully bridges the gap in maintaining logically coherent long-horizon narratives within Endogenous Interactive Agent Societies by addressing fundamental challenges of stochastic generative emergence through integrated memory, spatial, and operational mechanisms.

Abstract: Realizing endogenous narrative evolution in LLM-based multi-agent systems is hindered by the inherent stochasticity of generative emergence. In particular, long-horizon simulations suffer from social memory stacking, where conflicting relational states accumulate without resolution, and narrative-spatial dissonance, where spatial logic detaches from the evolving plot. To bridge this gap, we propose EvoSpark, a framework specifically designed to sustain logically coherent long-horizon narratives within Endogenous Interactive Agent Societies. To ensure consistency, the Stratified Narrative Memory employs a Role Socio-Evolutionary Base as living cognition, dynamically metabolizing experiences to resolve historical conflicts. Complementarily, Generative Mise-en-Scène mechanism enforces Role-Location-Plot alignment, synchronizing character presence with the narrative flow. Underpinning these is the Unified Narrative Operation Engine, which integrates an Emergent Character Grounding Protocol to transform stochastic sparking into persistent characters. This engine establishes a substrate that expands a minimal premise into an open-ended, evolving story world. Experiments demonstrate that EvoSpark significantly outperforms baselines across diverse paradigms, enabling the sustained generation of expressive and coherent narrative experiences.

Method: Model System 1 and System 2 thinking as semantic memory networks with distinct structures built from comparable datasets generated by both humans and LLMs, then use network-based evaluation metrics to analyze relationships between semantic memory structure and implicit gender bias.

Result: Semantic memory structures are irreducible only in humans (not LLMs), and semantic memory structure relates consistently to implicit bias only in humans, with lower bias levels in System 2 structures. LLMs lack certain types of human-like conceptual knowledge that contribute to bias regulation.

Conclusion: Fundamental differences exist between human and machine cognition regarding bias regulation, with humans possessing irreducible semantic structures that help mitigate bias while LLMs lack this capability, highlighting limitations in current LLM architectures.

Abstract: Implicit biases in both humans and large language models (LLMs) pose significant societal risks. Dual process theories propose that biases arise primarily from associative System 1 thinking, while deliberative System 2 thinking mitigates bias, but the cognitive mechanisms that give rise to this phenomenon remain poorly understood. To better understand what underlies this duality in humans, and possibly in LLMs, we model System 1 and System 2 thinking as semantic memory networks with distinct structures, built from comparable datasets generated by both humans and LLMs. We then investigate how these distinct semantic memory structures relate to implicit gender bias using network-based evaluation metrics. We find that semantic memory structures are irreducible only in humans, suggesting that LLMs lack certain types of human-like conceptual knowledge. Moreover, semantic memory structure relates consistently to implicit bias only in humans, with lower levels of bias in System~2 structures. These findings suggest that certain types of conceptual knowledge contribute to bias regulation in humans, but not in LLMs, highlighting fundamental differences between human and machine cognition.

Method: Proposes a framework based on multidimensional Item Response Theory (IRT) that uses anchor items to calibrate new benchmarks to existing evaluation suites. The approach holds previously calibrated item parameters fixed while using a fixed anchor set for each dataset, allowing results from different evaluation periods to be compared directly.

Result: In experiments with over 400 models, the framework predicts full-evaluation performance within 2-3 percentage points using only 100 anchor questions per dataset, with Spearman ρ ≥ 0.9 for ranking preservation. This demonstrates the ability to extend benchmark suites over time while preserving score comparability at constant evaluation cost per new dataset.

Conclusion: The proposed IRT-based framework with anchor items enables efficient and comparable evaluation of language models across different time periods and benchmark expansions, solving the growing pains of benchmark evaluation in NLP.

Abstract: The rapid release of both language models and benchmarks makes it increasingly costly to evaluate every model on every dataset. In practice, models are often evaluated on different samples, making scores difficult to compare across studies. To address this, we propose a framework based on multidimensional Item Response Theory (IRT) that uses anchor items to calibrate new benchmarks to the evaluation suite while holding previously calibrated item parameters fixed. Our approach supports a realistic evaluation setting in which datasets are introduced over time and models are evaluated only on the datasets available at the time of evaluation, while a fixed anchor set for each dataset is used so that results from different evaluation periods can be compared directly. In large-scale experiments on more than $400$ models, our framework predicts full-evaluation performance within 2-3 percentage points using only $100$ anchor questions per dataset, with Spearman $ρ\geq 0.9$ for ranking preservation, showing that it is possible to extend benchmark suites over time while preserving score comparability, at a constant evaluation cost per new dataset. Code available at https://github.com/eliyahabba/growing-pains

Method: Proposes round-trip translation evaluation: translate text from source to target language and back, then measure semantic gaps between original and result. Introduces Lost in Translation (LiT) benchmark using this method across widely spoken languages.

Result: Round-trip translation correlates almost perfectly (ρ = 0.94) with user ratings on real-world multilingual tasks (LMArena), requires no human references, and doesn’t need more capable multilingual judges than tested models.

Conclusion: Round-trip translation is a superior method for evaluating true multilingual generation capabilities of frontier models, and the LiT benchmark provides realistic multilingual evaluation.

Abstract: Multilingual benchmarks guide the development of frontier models. Yet multilingual evaluations reported by frontier models are structured similar to popular reasoning and knowledge benchmarks, but across many languages. We show such benchmarks, and consequently multilingual evaluations, measure mathematical reasoning and factual recall, not multilingual proficiency. For example, thinking variants dramatically outperform instruct variants on these benchmarks, yet often perform worse on real-world multilingual tasks, such as LMArena. We propose a simple alternative: evaluate multilingual capability via round-trip translation. Given text in a source language, translate it to a target language and back; semantic gaps between the original and result expose failures in multilingual generation capabilities. Round-trip translation correlates almost perfectly (\r{ho} = 0.94) with user ratings on LMArena with our benchmark, requires no human reference translations, and does not require a more capable multilingual judge than tested models. Lastly, we introduce Lost in Translation (LiT), a challenging round-trip translation benchmark spanning widely spoken languages worldwide, for realistic evaluation of multilingual frontier models.

Method: Created a framework to transform literal sentences into three figurative variants (metonymic, metaphoric, hybrid). Built MetFuse dataset with 1,000 human-verified quadruplets (4,000 sentences total). Conducted extrinsic experiments on eight benchmarks and analyzed human and LLM performance.

Result: Augmenting training data with MetFuse consistently improves both metonymy and metaphor classification, with hybrid examples yielding largest gains on metonymy tasks. Both humans and LLMs better identify metonymy in hybrid sentences than in metonymy-only sentences.

Conclusion: MetFuse enables study of figurative language fusion and shows that metaphor presence makes metonymic nouns more explicit. The dataset provides valuable resource for figurative language understanding research.

Abstract: Metonymy and metaphor often co-occur in natural language, yet computational work has studied them largely in isolation. We introduce a framework that transforms a literal sentence into three figurative variants: metonymic, metaphoric, and hybrid. Using this framework, we construct MetFuse, the first dedicated dataset of figurative fusion between metonymy and metaphor, containing 1,000 human-verified meaning-aligned quadruplets totaling 4,000 sentences. Extrinsic experiments on eight existing benchmarks show that augmenting training data with MetFuse consistently improves both metonymy and metaphor classification, with hybrid examples yielding the largest gains on metonymy tasks. Using this dataset, we also analyze how the presence of one figurative type impacts another. Our findings show that both human annotators and large language models better identify metonymy in hybrid sentences than in metonymy-only sentences, demonstrating that the presence of a metaphor makes a metonymic noun more explicit. Our dataset is publicly available at: https://github.com/cincynlp/MetFuse.

Method: Created GlotOCR Bench with clean and degraded image variants rendered from real multilingual texts using fonts from Google Fonts repository, shaped with HarfBuzz and rasterized with FreeType. Supports both LTR and RTL scripts with manual verification of correct rendering. Evaluated a broad suite of open-weight and proprietary vision-language models.

Result: Most models perform well on fewer than ten scripts, and even the strongest frontier models fail to generalize beyond thirty scripts. Performance tracks script-level pretraining coverage, indicating reliance on language model pretraining over visual recognition. Models either produce random noise or hallucinate characters from similar known scripts when confronted with unfamiliar scripts.

Conclusion: Current OCR systems rely heavily on language model pretraining rather than visual recognition capabilities. There’s significant room for improvement in cross-script generalization for vision-language models in OCR tasks.

Abstract: Optical character recognition (OCR) has advanced rapidly with the rise of vision-language models, yet evaluation has remained concentrated on a small cluster of high- and mid-resource scripts. We introduce GlotOCR Bench, a comprehensive benchmark evaluating OCR generalization across 100+ Unicode scripts. Our benchmark comprises clean and degraded image variants rendered from real multilingual texts. Images are rendered using fonts from the Google Fonts repository, shaped with HarfBuzz and rasterized with FreeType, supporting both LTR and RTL scripts. Samples of rendered images were manually reviewed to verify correct rendering across all scripts. We evaluate a broad suite of open-weight and proprietary vision-language models and find that most perform well on fewer than ten scripts, and even the strongest frontier models fail to generalize beyond thirty scripts. Performance broadly tracks script-level pretraining coverage, suggesting that current OCR systems rely on language model pretraining as much as on visual recognition. Models confronted with unfamiliar scripts either produce random noise or hallucinate characters from similar scripts they already know. We release the benchmark and pipeline for reproducibility. Pipeline Code: https://github.com/cisnlp/glotocr-bench, Benchmark: https://hf.co/datasets/cis-lmu/glotocr-bench.

Method: DDTree constructs a draft tree directly from per-position distributions of a block diffusion drafter. Under a fixed node budget, it uses a best-first heap algorithm to select continuations most likely to match the target model according to a surrogate metric. The resulting tree is verified efficiently in a single target model forward pass using an ancestor-only attention mask.

Result: DDTree achieves state-of-the-art speculative decoding performance, outperforming strong autoregressive drafters like EAGLE-3 and improving upon vanilla DFlash by enabling verification of multiple draft trajectories.

Conclusion: DDTree represents a leading approach to speculative decoding by combining the efficiency of block diffusion drafters with tree-based verification, enabling higher acceptance rates and better performance through parallel verification of multiple draft trajectories.

Abstract: Speculative decoding accelerates autoregressive language models by using a lightweight drafter to propose multiple future tokens, which the target model then verifies in parallel. DFlash shows that a block diffusion drafter can generate an entire draft block in a single forward pass and achieve state-of-the-art speculative decoding performance, outperforming strong autoregressive drafters such as EAGLE-3. Vanilla DFlash, however, still verifies only a single drafted trajectory per round, potentially limiting its acceptance length. We introduce DDTree (Diffusion Draft Tree), a method that constructs a draft tree directly from the per-position distributions of a block diffusion drafter. Under a fixed node budget, DDTree uses a simple best-first heap algorithm to select the continuations that are most likely to match the target model according to a surrogate defined by the draft model’s output. The resulting tree is verified efficiently in a single target model forward pass using an ancestor-only attention mask. Because DDTree builds on DFlash, a leading draft model for speculative decoding, these gains place DDTree among the leading approaches to speculative decoding.

Method: Created PolicyBench - first large-scale cross-system benchmark (US-China) with 21K cases across policy areas; assessed three capabilities following Bloom’s taxonomy: Memorization, Understanding, and Application; proposed PolicyMoE - domain-specialized Mixture-of-Experts model with expert modules aligned to each cognitive level.

Result: PolicyMoE models demonstrate stronger performance on application-oriented policy tasks than on memorization or conceptual understanding, and yield highest accuracy on structured reasoning tasks; reveals key limitations of current LLMs in policy understanding.

Conclusion: The work identifies limitations in current LLMs for policy comprehension and suggests paths toward more reliable, policy-focused models through specialized architectures like MoE.

Abstract: Large Language Models (LLMs) are increasingly integrated into real-world decision-making, including in the domain of public policy. Yet, their ability to comprehend and reason about policy-related content remains underexplored. To fill this gap, we present \textbf{\textit{PolicyBench}}, the first large-scale cross-system benchmark (US-China) evaluating policy comprehension, comprising 21K cases across a broad spectrum of policy areas, capturing the diversity and complexity of real-world governance. Following Bloom’s taxonomy, the benchmark assesses three core capabilities: (1) \textbf{Memorization}: factual recall of policy knowledge, (2) \textbf{Understanding}: conceptual and contextual reasoning, and (3) \textbf{Application}: problem-solving in real-life policy scenarios. Building on this benchmark, we further propose \textbf{\textit{PolicyMoE}}, a domain-specialized Mixture-of-Experts (MoE) model with expert modules aligned to each cognitive level. The proposed models demonstrate stronger performance on application-oriented policy tasks than on memorization or conceptual understanding, and yields the highest accuracy on structured reasoning tasks. Our results reveal key limitations of current LLMs in policy understanding and suggest paths toward more reliable, policy-focused models.

Method: Tested three open-weight model families and one closed-weight model (GPT-4o-mini) with simple lexical constraints. Used pairwise evaluation with GPT-4o-mini and GPT-4o as judges. Conducted mechanistic analysis including two-pass generation (free generation followed by constrained rewriting) and linear probes on prompt representations. Compared instruction-tuned models with base models under identical constraints.

Result: Instruction-tuned LLMs collapse their responses under trivial constraints, losing 14-48% comprehensiveness. GPT-4o-mini suffers 31% loss with 99% baseline win rate. Base models show no systematic collapse. Two-pass generation recovers 59-96% of response length. Linear probes predict response length with R²=0.51-0.93 for instruction-tuned models but negative R² for base models. Standard LLM-as-judge evaluation detects only 3.5% quality drop vs 23% in pairwise evaluation.

Conclusion: Instruction tuning creates fragility by coupling task competence to narrow surface-form templates, making models vulnerable to trivial constraints. This reveals a methodological blind spot in constrained generation assessment and shows that commercially deployed models share this vulnerability.

Abstract: Instruction-tuned large language models produce helpful, structured responses, but how robust is this helpfulness when trivially constrained? We show that simple lexical constraints (banning a single punctuation character or common word) cause instruction-tuned LLMs to collapse their responses, losing 14–48% of comprehensiveness in pairwise evaluation across three open-weight model families and one closed-weight model (GPT-4o-mini). The baseline response is preferred in 77–100% of 1,920 pairwise comparisons judged by GPT-4o-mini and GPT-4o. Notably, GPT-4o-mini suffers 31% comprehensiveness loss (99% baseline win rate), demonstrating that the fragility extends to commercially deployed closed-weight models, contrary to prior findings on format-level constraints. Through mechanistic analysis, we identify this as a planning failure: two-pass generation (free generation followed by constrained rewriting) recovers 59–96% of response length, and linear probes on prompt representations predict response length with $R^2 = 0.51$–$0.93$ before generation begins, with $R^2$ tracking collapse severity across models. The same probes yield negative $R^2$ on base models, confirming that instruction tuning creates the representational structure encoding the collapse decision. Crucially, base models show no systematic collapse under identical constraints, with effects that are small, noisy, and bidirectional, demonstrating that instruction tuning creates this fragility by coupling task competence to narrow surface-form templates. The effect replicates on MT-Bench across all eight task categories. We further show that standard independent LLM-as-judge evaluation detects only a 3.5% average quality drop where pairwise evaluation reveals 23%, exposing a methodological blind spot in how constrained generation is assessed.

Method: AiScientist combines hierarchical orchestration with a permission-scoped File-as-Bus workspace: a top-level Orchestrator maintains stage-level control through summaries and workspace maps, while specialized agents repeatedly re-ground on durable artifacts (analyses, plans, code, experimental evidence) rather than relying on conversational handoffs.

Result: AiScientist improves PaperBench score by 10.54 points on average over the best matched baseline and achieves 81.82 Any Medal% on MLE-Bench Lite. Ablation studies show File-as-Bus protocol is key, reducing PaperBench by 6.41 points and MLE-Bench Lite by 31.82 points when removed.

Conclusion: Long-horizon ML research engineering is a systems problem of coordinating specialized work over durable project state, rather than a purely local reasoning problem. The File-as-Bus approach with durable state continuity is crucial for sustained performance.

Abstract: Autonomous AI research has advanced rapidly, but long-horizon ML research engineering remains difficult: agents must sustain coherent progress across task comprehension, environment setup, implementation, experimentation, and debugging over hours or days. We introduce AiScientist, a system for autonomous long-horizon engineering for ML research built on a simple principle: strong long-horizon performance requires both structured orchestration and durable state continuity. To this end, AiScientist combines hierarchical orchestration with a permission-scoped File-as-Bus workspace: a top-level Orchestrator maintains stage-level control through concise summaries and a workspace map, while specialized agents repeatedly re-ground on durable artifacts such as analyses, plans, code, and experimental evidence rather than relying primarily on conversational handoffs, yielding thin control over thick state. Across two complementary benchmarks, AiScientist improves PaperBench score by 10.54 points on average over the best matched baseline and achieves 81.82 Any Medal% on MLE-Bench Lite. Ablation studies further show that File-as-Bus protocol is a key driver of performance, reducing PaperBench by 6.41 points and MLE-Bench Lite by 31.82 points when removed. These results suggest that long-horizon ML research engineering is a systems problem of coordinating specialized work over durable project state, rather than a purely local reasoning problem.

Method: Proposes a minimalist framework with two key components: Turn Isolation Retrieval (TIR) that uses max-activation strategy to capture turn-level signals instead of global aggregation, and Query-Driven Pruning (QDP) that removes redundant sessions and conversational filler to create compact, high-density evidence sets.

Result: Extensive experiments on multiple benchmarks show the method achieves robust performance across diverse settings, consistently outperforming strong baselines while maintaining high efficiency in tokens and latency.

Conclusion: The framework establishes a new minimalist baseline for conversational memory by addressing signal sparsity through simple yet effective retrieval and pruning mechanisms, demonstrating that complex hierarchical summarization or reinforcement learning may not be necessary for effective long-term dialogue management.

Abstract: Existing conversational memory systems rely on complex hierarchical summarization or reinforcement learning to manage long-term dialogue history, yet remain vulnerable to context dilution as conversations grow. In this work, we offer a different perspective: the primary bottleneck may lie not in memory architecture, but in the \textit{Signal Sparsity Effect} within the latent knowledge manifold. Through controlled experiments, we identify two key phenomena: \textit{Decisive Evidence Sparsity}, where relevant signals become increasingly isolated with longer sessions, leading to sharp degradation in aggregation-based methods; and \textit{Dual-Level Redundancy}, where both inter-session interference and intra-session conversational filler introduce large amounts of non-informative content, hindering effective generation. Motivated by these insights, we propose \method, a minimalist framework that brings conversational memory back to basics, relying solely on retrieval and generation via Turn Isolation Retrieval (TIR) and Query-Driven Pruning (QDP). TIR replaces global aggregation with a max-activation strategy to capture turn-level signals, while QDP removes redundant sessions and conversational filler to construct a compact, high-density evidence set. Extensive experiments on multiple benchmarks demonstrate that \method achieves robust performance across diverse settings, consistently outperforming strong baselines while maintaining high efficiency in tokens and latency, establishing a new minimalist baseline for conversational memory.

Method: Unable to determine method due to access limitations

Result: Unable to determine results due to access limitations

Conclusion: Unable to determine conclusion due to access limitations

Abstract: Failed to fetch summary for 2409.06679: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2409.06679&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2410.23728: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2410.23728&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to paper fetch failure

Result: Unable to determine results due to paper fetch failure

Conclusion: Unable to determine conclusion due to paper fetch failure

Abstract: Failed to fetch summary for 2412.07238: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2412.07238&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as abstract content is unavailable due to API rate limiting

Result: Cannot determine results as abstract content is unavailable due to API rate limiting

Conclusion: Cannot draw conclusions as abstract content is unavailable due to API rate limiting

Abstract: Failed to fetch summary for 2501.16154: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2501.16154&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper abstract

Result: Cannot determine results without access to the paper abstract

Conclusion: Cannot draw conclusions without access to the paper abstract

Abstract: Failed to fetch summary for 2503.10676: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2503.10676&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2505.15467: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2505.15467&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to technical error in fetching paper content

Result: Unable to determine results due to technical error in fetching paper content

Conclusion: Unable to draw conclusions due to technical error in fetching paper content

Abstract: Failed to fetch summary for 2505.17086: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2505.17086&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content could not be retrieved

Result: Cannot determine results as paper content could not be retrieved

Conclusion: Cannot draw conclusions about paper content due to retrieval failure

Abstract: Failed to fetch summary for 2507.06448: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.06448&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2509.07177: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.07177&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot determine conclusion without access to paper content

Abstract: Failed to fetch summary for 2509.17995: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.17995&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to missing paper content

Result: Cannot determine results due to missing paper content

Conclusion: Cannot determine conclusion due to missing paper content

Abstract: Failed to fetch summary for 2509.19695: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.19695&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: N/A - Paper content not accessible

Result: N/A - Paper content not accessible

Conclusion: N/A - Paper content not accessible

Abstract: Failed to fetch summary for 2509.19742: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.19742&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable due to API rate limiting

Result: Cannot determine results as paper content is unavailable due to API rate limiting

Conclusion: Cannot determine conclusion as paper content is unavailable due to API rate limiting

Abstract: Failed to fetch summary for 2510.00919: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.00919&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2510.03174: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.03174&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper retrieval

Result: Unable to determine results due to failed paper retrieval

Conclusion: Unable to determine conclusion due to failed paper retrieval

Abstract: Failed to fetch summary for 2510.03323: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.03323&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2510.09536: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.09536&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about the paper due to access limitations

Abstract: Failed to fetch summary for 2510.14420: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.14420&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as abstract retrieval failed

Result: Cannot determine results as abstract retrieval failed

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2510.15552: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.15552&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2510.19644: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.19644&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to inability to access paper content

Result: Cannot determine results due to inability to access paper content

Conclusion: Cannot draw conclusions due to inability to access paper content

Abstract: Failed to fetch summary for 2510.20635: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.20635&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as abstract is unavailable due to rate limiting error

Result: Cannot determine results as abstract is unavailable due to rate limiting error

Conclusion: Cannot draw conclusions as abstract is unavailable due to rate limiting error

Abstract: Failed to fetch summary for 2511.09803: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.09803&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2511.10453: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.10453&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2512.13961: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.13961&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to abstract fetch failure

Result: Unable to determine results due to abstract fetch failure

Conclusion: Unable to determine conclusion due to abstract fetch failure

Abstract: Failed to fetch summary for 2601.02589: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.02589&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot determine conclusion without access to paper content

Abstract: Failed to fetch summary for 2601.05524: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.05524&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to missing paper content

Result: Unable to determine results due to missing paper content

Conclusion: Unable to determine conclusion due to missing paper content

Abstract: Failed to fetch summary for 2601.08209: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.08209&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2601.09313: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.09313&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to unavailability of paper content

Result: Cannot determine results due to unavailability of paper content

Conclusion: Cannot determine conclusion due to unavailability of paper content

Abstract: Failed to fetch summary for 2601.11047: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.11047&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to inability to access paper content

Result: Cannot determine results due to inability to access paper content

Conclusion: Cannot determine conclusion due to inability to access paper content

Abstract: Failed to fetch summary for 2601.14004: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.14004&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to determine conclusion due to failed paper fetch

Abstract: Failed to fetch summary for 2601.19917: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.19917&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API rate limiting preventing access to paper content

Result: Unable to determine results due to API rate limiting preventing access to paper content

Conclusion: Unable to determine conclusion due to API rate limiting preventing access to paper content

Abstract: Failed to fetch summary for 2602.05971: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.05971&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2603.03249: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.03249&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2603.06552: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.06552&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot analyze method as paper content is unavailable due to HTTP 429 error

Result: No results available - arXiv API returned HTTP 429 (Too Many Requests) error

Conclusion: Cannot provide analysis due to technical limitations in accessing the paper content

Abstract: Failed to fetch summary for 2603.18203: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.18203&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2603.20640: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.20640&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API access error

Result: Unable to determine results due to API access error

Conclusion: Unable to analyze paper due to technical issues with arXiv API access

Abstract: Failed to fetch summary for 2507.22359: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.22359&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to draw conclusions due to fetch failure

Abstract: Failed to fetch summary for 2603.21440: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.21440&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: No method information available since the paper content could not be retrieved.

Result: No results can be analyzed without access to the paper content.

Conclusion: Cannot provide conclusions about a paper whose content is unavailable due to API rate limiting.

Abstract: Failed to fetch summary for 2603.23998: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.23998&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot determine conclusion without access to paper content

Abstract: Failed to fetch summary for 2604.02830: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.02830&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to missing paper content

Result: Cannot determine results due to missing paper content

Conclusion: Cannot determine conclusion due to missing paper content

Abstract: Failed to fetch summary for 2604.03136: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.03136&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper content. The technical approach is unknown due to the fetch failure.

Result: Cannot determine results without access to the paper content. No experimental findings or evaluations can be analyzed.

Conclusion: Cannot draw conclusions about the paper’s contributions without access to its content. The analysis is blocked by technical limitations.

Abstract: Failed to fetch summary for 2604.05546: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.05546&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper abstract

Result: Cannot determine results without access to paper abstract

Conclusion: Cannot draw conclusions without access to paper abstract

Abstract: Failed to fetch summary for 2604.05643: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.05643&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.05795: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.05795&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2510.23538: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.23538&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.05821: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.05821&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2604.06812: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.06812&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes Many-Tier Instruction Hierarchy (ManyIH) paradigm for resolving instruction conflicts with arbitrarily many privilege levels. Introduces ManyIH-Bench benchmark with up to 12 privilege levels across 853 agentic tasks (coding and instruction-following) spanning 46 real-world agents.

Result: Experiments show current frontier models perform poorly (~40% accuracy) when instruction conflict scales, highlighting the urgent need for methods targeting fine-grained, scalable instruction conflict resolution.

Conclusion: The work underscores the need for explicit methods targeting fine-grained, scalable instruction conflict resolution in agentic settings, as current models struggle with complex privilege hierarchies.

Abstract: Large language model agents receive instructions from many sources-system messages, user prompts, tool outputs, other agents, and more-each carrying different levels of trust and authority. When these instructions conflict, agents must reliably follow the highest-privilege instruction to remain safe and effective. The dominant paradigm, instruction hierarchy (IH), assumes a fixed, small set of privilege levels (typically fewer than five) defined by rigid role labels (e.g., system > user). This is inadequate for real-world agentic settings, where conflicts can arise across far more sources and contexts. In this work, we propose Many-Tier Instruction Hierarchy (ManyIH), a paradigm for resolving instruction conflicts among instructions with arbitrarily many privilege levels. We introduce ManyIH-Bench, the first benchmark for ManyIH. ManyIH-Bench requires models to navigate up to 12 levels of conflicting instructions with varying privileges, comprising 853 agentic tasks (427 coding and 426 instruction-following). ManyIH-Bench composes constraints developed by LLMs and verified by humans to create realistic and difficult test cases spanning 46 real-world agents. Our experiments show that even the current frontier models perform poorly (~40% accuracy) when instruction conflict scales. This work underscores the urgent need for methods that explicitly target fine-grained, scalable instruction conflict resolution in agentic settings.

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about the paper due to lack of access to its content

Abstract: Failed to fetch summary for 2604.10151: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10151&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content.

Result: Cannot determine results without access to paper content.

Conclusion: Cannot draw conclusions without access to paper content.

Abstract: Failed to fetch summary for 2604.11201: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11201&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed summary fetch

Result: Unable to determine results due to failed summary fetch

Conclusion: Unable to determine conclusion due to failed summary fetch

Abstract: Failed to fetch summary for 2604.11246: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11246&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper abstract

Result: Cannot determine results without access to paper abstract

Conclusion: Cannot draw conclusions without access to paper abstract

Abstract: Failed to fetch summary for 2604.11427: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11427&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot analyze method as paper content is unavailable due to technical limitations

Result: No results available - arXiv API returned HTTP 429 error indicating too many requests

Conclusion: Cannot provide analysis due to technical limitations in accessing the paper content

Abstract: Failed to fetch summary for 2604.11554: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11554&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about paper content due to access limitations

Abstract: Failed to fetch summary for 2604.11748: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11748&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot analyze method as paper content is unavailable due to HTTP 429 error

Result: No results available - paper summary retrieval failed

Conclusion: Cannot provide analysis due to technical limitations in accessing the paper

Abstract: Failed to fetch summary for 2502.17403: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2502.17403&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed API request

Result: Unable to determine results due to failed API request

Conclusion: Unable to draw conclusions due to failed API request

Abstract: Failed to fetch summary for 2503.21708: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2503.21708&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2505.14264: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2505.14264&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper abstract

Result: Cannot determine results without access to paper abstract

Conclusion: Cannot draw conclusions without access to paper abstract

Abstract: Failed to fetch summary for 2505.18646: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2505.18646&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to technical error in accessing paper content

Result: Unable to determine results due to technical error in accessing paper content

Conclusion: Unable to draw conclusions due to technical error in accessing paper content

Abstract: Failed to fetch summary for 2509.18127: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.18127&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2509.26404: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.26404&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to inability to access paper content

Result: Cannot determine results due to inability to access paper content

Conclusion: Cannot determine conclusion due to inability to access paper content

Abstract: Failed to fetch summary for 2511.19820: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.19820&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2512.14754: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.14754&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable due to API rate limiting

Result: Cannot determine results as paper content is unavailable due to API rate limiting

Conclusion: Cannot determine conclusion as paper content is unavailable due to API rate limiting

Abstract: Failed to fetch summary for 2604.07595: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.07595&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2602.11236: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.11236&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method as paper content could not be retrieved

Result: Unable to determine results as paper content could not be retrieved

Conclusion: Unable to draw conclusions as paper content could not be retrieved

Abstract: Failed to fetch summary for 2602.23061: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.23061&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to analyze paper content due to technical fetch error

Abstract: Failed to fetch summary for 2603.17361: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.17361&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.00136: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.00136&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content.

Result: Cannot determine results without access to paper content.

Conclusion: Cannot determine conclusion without access to paper content.

Abstract: Failed to fetch summary for 2604.05818: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.05818&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes a framework with three components: Adaptive Semantic Grouping (ASG) for dynamic codebook partitioning based on semantic similarity and secret key; Block-wise Multi-bit Encoding (BME) for reliable message transmission with error correction; Unified Token-Replacement Interface (UTRI) to support different AR paradigms (next-token and next-scale prediction).

Result: Achieves state-of-the-art performance in image quality (FID), watermark detection accuracy, and multi-bit message extraction. Maintains robustness against various attacks including cropping, JPEG compression, noise, blur, color jitter, and random erasing.

Conclusion: The proposed framework addresses key limitations of existing AR watermarking methods by enabling multi-bit message embedding, improving security through adaptive partitioning, and providing generalization across different AR architectures.

Abstract: Invisible watermarking for autoregressive (AR) image generation has recently gained attention as a means of protecting image ownership and tracing AI-generated content. However, existing approaches suffer from three key limitations: (1) they embed only zero-bit watermarks for binary verification, lacking the ability to convey multi-bit messages; (2) they rely on static codebook partitioning strategies that are vulnerable to security attacks once the partition is exposed; and (3) they are designed for specific AR architectures, failing to generalize across diverse AR paradigms. We propose \method{}, a training-free, unified watermarking framework for autoregressive image generators that addresses all three limitations. \method{} introduces three core components: \textbf{Adaptive Semantic Grouping (ASG)}, which dynamically partitions codebook entries based on semantic similarity and a secret key, ensuring both image quality preservation and security; \textbf{Block-wise Multi-bit Encoding (BME)}, which divides the token sequence into blocks and encodes different bits across blocks with error-correcting codes for reliable message transmission; and \textbf{a Unified Token-Replacement Interface (UTRI)} that abstracts the watermark embedding process to support both next-token prediction (e.g., LlamaGen) and next-scale prediction (e.g., VAR) paradigms. We provide theoretical analysis on detection error rates and embedding capacity. Extensive experiments on three AR models demonstrate that \method{} achieves state-of-the-art performance in image quality (FID), watermark detection accuracy, and multi-bit message extraction, while maintaining robustness against cropping, JPEG compression, Gaussian noise, blur, color jitter, and random erasing attacks.

Method: Built upon GTPBD, GTPBD-MM integrates high-resolution optical imagery, structured text descriptions, and DEM data, supporting evaluation under Image-only, Image+Text, and Image+Text+DEM settings. Also proposes ETTerra, a multimodal baseline network for terraced parcel delineation.

Result: Extensive experiments show that textual semantics and terrain geometry provide complementary cues beyond visual appearance alone, yielding more accurate, coherent, and structurally consistent delineation results in complex terraced scenes.

Conclusion: The multimodal approach combining visual, textual, and elevation data significantly improves terraced parcel extraction in challenging mountainous environments, addressing limitations of existing unimodal methods.

Abstract: Agricultural parcel extraction plays an important role in remote sensing-based agricultural monitoring, supporting parcel surveying, precision management, and ecological assessment. However, existing public benchmarks mainly focus on regular and relatively flat farmland scenes. In contrast, terraced parcels in mountainous regions exhibit stepped terrain, pronounced elevation variation, irregular boundaries, and strong cross-regional heterogeneity, making parcel extraction a more challenging problem that jointly requires visual recognition, semantic discrimination, and terrain-aware geometric understanding. Although recent studies have advanced visual parcel benchmarks and image-text farmland understanding, a unified benchmark for complex terraced parcel extraction under aligned image-text-DEM settings remains absent. To fill this gap, we present GTPBD-MM, the first multimodal benchmark for global terraced parcel extraction. Built upon GTPBD, GTPBD-MM integrates high-resolution optical imagery, structured text descriptions, and DEM data, and supports systematic evaluation under Image-only, Image+Text, and Image+Text+DEM settings. We further propose Elevation and Text guided Terraced parcel network (ETTerra), a multimodal baseline for terraced parcel delineation. Extensive experiments demonstrate that textual semantics and terrain geometry provide complementary cues beyond visual appearance alone, yielding more accurate, coherent, and structurally consistent delineation results in complex terraced scenes.

Method: Identifies sparse neuron-level concept activations from pretrained VLM representations, translates them into pseudo-report-style summaries, and uses a quantitative semantic verification protocol with an independent medical LLM as evaluator to assess concept alignment with radiology reports.

Result: Introduces three concept scores (Aligned, Unaligned, Uncertain) to quantify semantic support, contradiction, or ambiguity relative to radiology reports, providing a quantitative baseline for assessing interpretability in medical VLMs.

Conclusion: MedConcept enables physician-level inspection of internal model reasoning and provides a framework for trustworthy clinical deployment of pretrained medical VLMs through concept-based interpretability.

Abstract: While medical Vision-Language models (VLMs) achieve strong performance on tasks such as tumor or organ segmentation and diagnosis prediction, their opaque latent representations limit clinical trust and the ability to explain predictions. Interpretability of these multimodal representations are therefore essential for the trustworthy clinical deployment of pretrained medical VLMs. However, current interpretability methods, such as gradient- or attention-based visualizations, are often limited to specific tasks such as classification. Moreover, they do not provide concept-level explanations derived from shared pretrained representations that can be reused across downstream tasks. We introduce MedConcept, a framework that uncovers latent medical concepts in a fully unsupervised manner and grounds them in clinically verifiable textual semantics. MedConcept identifies sparse neuron-level concept activations from pretrained VLM representations and translates them into pseudo-report-style summaries, enabling physician-level inspection of internal model reasoning. To address the lack of quantitative evaluation in concept-based interpretability, we introduce a quantitative semantic verification protocol that leverages an independent pretrained medical LLM as a frozen external evaluator to assess concept alignment with radiology reports. We define three concept scores, Aligned, Unaligned, and Uncertain, to quantify semantic support, contradiction, or ambiguity relative to radiology reports and use them exclusively for post hoc evaluation. These scores provide a quantitative baseline for assessing interpretability in medical VLMs. All codes, prompt and data to be released on acceptance. Ke

Method: Created EgoEsportsQA benchmark with 1,745 QA pairs from professional esports matches across 3 FPS games using a six-stage pipeline. Structured questions into two-dimensional taxonomy: 11 cognitive capability sub-tasks (perception/reasoning) and 6 esports knowledge sub-tasks.

Result: State-of-the-art Video-LLMs achieve only 71.58% accuracy, showing significant gaps: better at basic visual perception than deep tactical reasoning, and better at macro-progression than micro-operations. Ablation experiments reveal architectural weaknesses.

Conclusion: The benchmark reveals limitations of current Video-LLMs in virtual environments, connects real-world and virtual domains, and provides guidance for optimizing esports applications and advancing Video-LLMs in egocentric environments.

Abstract: While video large language models (Video-LLMs) excel in understanding slow-paced, real-world egocentric videos, their capabilities in high-velocity, information-dense virtual environments remain under-explored. Existing benchmarks focus on daily activities, yet lack a rigorous testbed for evaluating fast, rule-bound reasoning in virtual scenarios. To fill this gap, we introduce EgoEsportsQA, a pioneering video question-answering (QA) benchmark for grounding perception and reasoning in expert esports knowledge. We curate 1,745 high-quality QA pairs from professional matches across 3 first-person shooter games via a scalable six-stage pipeline. These questions are structured into a two-dimensional decoupled taxonomy: 11 sub-tasks in the cognitive capability dimension (covering perception and reasoning levels) and 6 sub-tasks in the esports knowledge dimension. Comprehensive evaluations of state-of-the-art Video-LLMs reveal that current models still fail to achieve satisfactory performance, with the best model only 71.58%. The results expose notable gaps across both axes: models exhibit stronger capabilities in basic visual perception than in deep tactical reasoning, and they grasp overall macro-progression better than fine-grained micro-operations. Extensive ablation experiments demonstrate the intrinsic weaknesses of current Video-LLM architectures. Further analysis suggests that our dataset not only reveals the connections between real-world and virtual egocentric domains, but also offers guidance for optimizing downstream esports applications, thereby fostering the future advancement of Video-LLMs in various egocentric environments.

Method: Proposes V-Nutri, a staged framework combining Nutrition5K-pretrained visual backbones with a lightweight fusion module that aggregates features from final dish frame and cooking process keyframes. Includes a cooking keyframes selection module using VideoMamba-based event-detection model targeting ingredient-addition moments.

Result: Experiments on HD-EPIC dataset show process cues provide complementary nutritional evidence, improving nutrition estimation under controlled conditions. Benefit of process keyframes depends strongly on backbone representation capacity and event detection quality.

Conclusion: Cooking process information from egocentric videos can contribute to dish-level nutrition estimation, establishing first benchmark for video-based nutrition estimation with V-Nutri framework.

Abstract: Nutrition estimation of meals from visual data is an important problem for dietary monitoring and computational health, but existing approaches largely rely on single images of the finally completed dish. This setting is fundamentally limited because many nutritionally relevant ingredients and transformations, such as oils, sauces, and mixed components, become visually ambiguous after cooking, making accurate calorie and macronutrient estimation difficult. In this paper, we investigate whether the cooking process information from egocentric cooking videos can contribute to dish-level nutrition estimation. First, we further manually annotated the HD-EPIC dataset and established the first benchmark for video-based nutrition estimation. Most importantly, we propose V-Nutri, a staged framework that combines Nutrition5K-pretrained visual backbones with a lightweight fusion module that aggregates features from the final dish frame and cooking process keyframes extracted from the egocentric videos. V-Nutri also includes a cooking keyframes selection module, a VideoMamba-based event-detection model that targets ingredient-addition moments. Experiments on the HD-EPIC dataset show that process cues can provide complementary nutritional evidence, improving nutrition estimation under controlled conditions. Our results further indicate that the benefit of process keyframes depends strongly on backbone representation capacity and event detection quality. Our code and annotated dataset is available at https://github.com/K624-YCK/V-Nutri.

Method: Proposed framework enables users to outline a rough ROI enclosing dense tissue on the central slice and select a segmentation threshold. The algorithm projects the ROI to remaining slices and iteratively adjusts slice-specific thresholds to maintain consistent dense tissue delineation across the DBT volume.

Result: Evaluation on 44 DBT volumes showed high inter-reader agreement (median Dice = 0.84) and method accuracy (median Dice = 0.83) compared to manual segmentations on 176 slices from 20th and 80th percentile slices.

Conclusion: The framework substantially reduces annotation time and labor while maintaining high segmentation accuracy, addressing the critical need for annotated DBT data for algorithm development in breast cancer screening.

Abstract: Digital breast tomosynthesis (DBT) is now the standard of care for breast cancer screening in the USA. Accurate segmentation of fibroglandular tissue in DBT images is essential for personalized risk estimation, but algorithm development is limited by scarce human-delineated training data. In this study we introduce a time- and labor-saving framework to generate a human-annotated binary segmentation mask for dense tissue in DBT. Our framework enables a user to outline a rough region of interest (ROI) enclosing dense tissue on the central reconstructed slice of a DBT volume and select a segmentation threshold to generate the dense tissue mask. The algorithm then projects the ROI to the remaining slices and iteratively adjusts slice-specific thresholds to maintain consistent dense tissue delineation across the DBT volume. By requiring annotation only on the central slice, the framework substantially reduces annotation time and labor. We used 44 DBT volumes from the DBTex dataset for evaluation. Inter-reader agreement was assessed by computing patient-wise Dice similarity coefficients between segmentation masks produced by two radiologists, yielding a median of 0.84. Accuracy of the proposed method was evaluated by having a radiologist manually segment the 20th and 80th percentile slices from each volume (CC and MLO views; 176 slices total) and calculate Dice scores between the manual and proposed segmentations, yielding a median of 0.83.

Method: Proposes MANet (Motion-aware and Audio-guided Network) that captures subtle motion inconsistencies in listener videos while leveraging speaker’s audio semantics to guide cross-modal fusion. Creates ListenForge dataset using 5 Listening Head Generation methods.

Result: Existing Speaking Deepfake Detection models perform poorly in listening scenarios. MANet achieves significantly superior performance on ListenForge dataset, demonstrating effectiveness of the proposed approach.

Conclusion: Highlights need to rethink deepfake detection beyond traditional speaking-centric paradigm, opens new directions for multimodal forgery analysis in interactive communication settings.

Abstract: Existing deepfake detection research has primarily focused on scenarios where the manipulated subject is actively speaking, i.e., generating fabricated content by altering the speaker’s appearance or voice. However, in realistic interaction settings, attackers often alternate between falsifying speaking and listening states to mislead their targets, thereby enhancing the realism and persuasiveness of the scenario. Although the detection of ’listening deepfakes’ remains largely unexplored and is hindered by a scarcity of both datasets and methodologies, the relatively limited quality of synthesized listening reactions presents an excellent breakthrough opportunity for current deepfake detection efforts. In this paper, we present the task of Listening Deepfake Detection (LDD). We introduce ListenForge, the first dataset specifically designed for this task, constructed using five Listening Head Generation (LHG) methods. To address the distinctive characteristics of listening forgeries, we propose MANet, a Motion-aware and Audio-guided Network that captures subtle motion inconsistencies in listener videos while leveraging speaker’s audio semantics to guide cross-modal fusion. Extensive experiments demonstrate that existing Speaking Deepfake Detection (SDD) models perform poorly in listening scenarios. In contrast, MANet achieves significantly superior performance on ListenForge. Our work highlights the necessity of rethinking deepfake detection beyond the traditional speaking-centric paradigm and opens new directions for multimodal forgery analysis in interactive communication settings. The dataset and code are available at https://anonymous.4open.science/r/LDD-B4CB.

Method: 1) Four-method parallel plate detector for robust plate reading across lighting conditions. 2) Three-stage state identification engine combining OCR text matching, color scoring, and lightweight neural network. 3) Hybrid depth fusion with inverse-variance weighting, online scale alignment, and 1D constant-velocity Kalman filter for smoothed distance, velocity, and time-to-collision estimation.

Result: 2.3% coefficient of variation in character height measurements, 36% reduction in distance-estimate variance compared to prior plate-width methods. Mean absolute error of 2.3% at 10m, continuous output during brief occlusions, outperforming deep learning baselines by 5x in relative error.

Conclusion: The framework successfully resolves scale ambiguity in monocular distance estimation using license plates as passive fiducial markers, providing accurate, certifiable distance measurements without training data or active illumination.

Abstract: Accurate inter-vehicle distance estimation is a cornerstone of Advanced Driver Assistance Systems (ADAS) and autonomous driving. While LiDAR and radar provide high precision, their high cost prohibits widespread adoption in mass-market vehicles. Monocular camera-based estimation offers a low-cost alternative but suffers from fundamental scale ambiguity. Recent deep learning methods for monocular depth achieve impressive results yet require expensive supervised training, suffer from domain shift, and produce predictions that are difficult to certify for safety-critical deployment. This paper presents a framework that exploits the standardized typography of United States license plates as passive fiducial markers for metric ranging, resolving scale ambiguity through explicit geometric priors without any training data or active illumination. First, a four-method parallel plate detector achieves robust plate reading across the full automotive lighting range. Second, a three-stage state identification engine fusing OCR text matching, multi-design color scoring, and a lightweight neural network classifier provides robust identification across all ambient conditions. Third, hybrid depth fusion with inverse-variance weighting and online scale alignment, combined with a one-dimensional constant-velocity Kalman filter, delivers smoothed distance, relative velocity, and time-to-collision for collision warning. Baseline validation reproduces a 2.3% coefficient of variation in character height measurements and a 36% reduction in distance-estimate variance compared with plate-width methods from prior work. Extensive outdoor experiments confirm a mean absolute error of 2.3% at 10 m and continuous distance output during brief plate occlusions, outperforming deep learning baselines by a factor of five in relative error.

Method: EigenCoin manifold with Bhattacharyya distance consisting of three steps: manifold construction, mapping test data, and classification.

Result: EigenCoin outperformed other observed algorithms with accuracy improvements from 9.45% up to 21.75%, while demonstrating capability to handle over-fitting problems.

Conclusion: The EigenCoin approach is effective for imbalanced coin classification tasks, showing superiority over other methods in both accuracy and overfitting resistance.

Abstract: Solving pattern recognition problems using imbalanced databases is a hot topic, which entices researchers to bring it into focus. Therefore, we consider this problem in the application of Sassanid coins classification. Our focus is not only on proposing EigenCoin manifold with Bhattacharyya distance for the classification task, but also on testing the influence of the holistic and feature-based approaches. EigenCoin consists of three main steps namely manifold construction, mapping test data, and classification. Conducted experiments show EigenCoin outperformed other observed algorithms and achieved the accuracy from 9.45% up to 21.75%, while it has the capability of handling the over-fitting problem.

Method: DPC-VQA decouples perception and calibration: uses a frozen MLLM to provide base quality estimates and perceptual priors, and employs a lightweight calibration branch to predict residual corrections for target-scenario adaptation. This avoids costly end-to-end retraining.

Result: DPC-VQA achieves competitive performance on both UGC and AIGC benchmarks, using less than 2% of trainable parameters compared to conventional MLLM-based VQA methods, and remains effective with only 20% of MOS labels.

Conclusion: The decoupling approach enables efficient adaptation of MLLMs for VQA tasks, significantly reducing training costs and data requirements while maintaining reliable performance across different content types.

Abstract: Recent multimodal large language models (MLLMs) have shown promising performance on video quality assessment (VQA) tasks. However, adapting them to new scenarios remains expensive due to large-scale retraining and costly mean opinion score (MOS) annotations. In this paper, we argue that a pretrained MLLM already provides a useful perceptual prior for VQA, and that the main challenge is to efficiently calibrate this prior to the target MOS space. Based on this insight, we propose DPC-VQA, a decoupling perception and calibration framework for video quality assessment. Specifically, DPC-VQA uses a frozen MLLM to provide a base quality estimate and perceptual prior, and employs a lightweight calibration branch to predict a residual correction for target-scenario adaptation. This design avoids costly end-to-end retraining while maintaining reliable performance with lower training and data costs. Extensive experiments on both user-generated content (UGC) and AI-generated content (AIGC) benchmarks show that DPC-VQA achieves competitive performance against representative baselines, while using less than 2% of the trainable parameters of conventional MLLM-based VQA methods and remaining effective with only 20% of MOS labels. The code will be released upon publication.

Method: Uses a 3D Human Mesh Recovery model to extract gait parameters from video recordings of older adults completing the Timed Up and Go test. Parameters include step time, sit-to-stand duration, and step length. Validation against IMU-based insole measurements and analysis using linear mixed effects models.

Result: Video-derived step time significantly correlated with IMU measurements. Shorter, more variable step lengths and longer sit-to-stand durations were associated with higher self-rated fall risk and fear of falling.

Conclusion: The pipeline enables accessible, ecologically valid gait analysis in community settings, providing clinically relevant insights into fall risk through video-based assessment.

Abstract: Gait assessment is a key clinical indicator of fall risk and overall health in older adults. However, standard clinical practice is largely limited to stopwatch-measured gait speed. We present a pipeline that leverages a 3D Human Mesh Recovery (HMR) model to extract gait parameters from recordings of older adults completing the Timed Up and Go (TUG) test. From videos recorded across different community centers, we extract and analyze spatiotemporal gait parameters, including step time, sit-to-stand duration, and step length. We found that video-derived step time was significantly correlated with IMU-based insole measurements. Using linear mixed effects models, we confirmed that shorter, more variable step lengths and longer sit-to-stand durations were predicted by higher self-rated fall risk and fear of falling. These findings demonstrate that our pipeline can enable accessible and ecologically valid gait analysis in community settings.

Method: Unable to determine method as paper content could not be retrieved

Result: Unable to determine results as paper content could not be retrieved

Conclusion: Unable to determine conclusion as paper content could not be retrieved

Abstract: Failed to fetch summary for 2604.06063: Page request resulted in HTTP 406 (https://export.arxiv.org/api/query?search_query=&id_list=2604.06063&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2511.06424: Page request resulted in HTTP 406 (https://export.arxiv.org/api/query?search_query=&id_list=2511.06424&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Created a dataset of 1,593 document images with tables in three visual styles, paired with question-answer sets in four languages (Bahasa Indonesia, English, Hindi, Arabic). Benchmarked leading VLMs and fine-tuned models with LoRA, testing spatial priors via table region coordinates.

Result: Found substantial performance gaps in VLMs, especially on complex tables and low-resource languages. Fine-tuning improved accuracy by 11.6-17.8%, and adding spatial priors boosted performance by 4-7%.

Conclusion: Targeted fine-tuning and spatial priors significantly enhance VLM performance on specialized document understanding. INDOTABVQA provides valuable resources for advancing cross-lingual, structure-aware document understanding research.

Abstract: We introduce INDOTABVQA, a benchmark for evaluating cross-lingual Table Visual Question Answering (VQA) on real-world document images in Bahasa Indonesia. The dataset comprises 1,593 document images across three visual styles (bordered, borderless, and colorful) with one or more than one tables, and 1,593 question-answer sets in four languages: Bahasa Indonesia, English, Hindi, and Arabic. This enables evaluation of Vision-Language Models (VLMs) in both monolingual (Bahasa documents with Bahasa questions) and cross-lingual settings (Bahasa documents with questions in other languages). We benchmark leading open-source VLMs (Qwen2.5-VL, Gemma-3, LLaMA-3.2) and GPT-4o and reveal substantial performance gaps, particularly on structurally complex tables and in low-resource languages. Fine-tuning a compact 3B and LoRA-finetuned 7B model on our dataset yields 11.6% and 17.8% improvements in accuracy. Providing explicit table region coordinates as additional input further improves performance by 4-7%, demonstrating the value of Spatial priors for table-based reasoning. Our findings underscore the importance of language-diverse, domain-specific datasets and demonstrate that targeted fine-tuning can significantly enhance VLM performance on specialized document understanding tasks. INDOTABVQA provides a valuable resource for advancing research in cross-lingual, structure-aware document understanding, especially in underrepresented regions of the world. Full dataset can be accessed in huggingface at: https://huggingface.co/datasets/NusaBharat/INDOTABVQA}

Method: Unable to determine method due to paper access failure

Result: Unable to determine results due to paper access failure

Conclusion: Paper content unavailable for analysis

Abstract: Failed to fetch summary for 2604.10655: Page request resulted in HTTP 406 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10655&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Correlation-aware training (CAT): end-to-end optimization of a trainable correlated-photon illumination source and a Transformer backend. The system learns to benefit from photon correlations using ≤100 shots, with the Transformer backend processing the correlated illumination patterns for object recognition.

Result: Achieved up to 15 percentage points improvement in classification accuracy over conventional uncorrelated-illumination methods in ultra-low-light and noisy conditions. Also outperformed untrained correlated-photon illumination approaches.

Conclusion: Specializing photon correlation patterns for specific computer vision tasks (object recognition) and jointly training illumination with digital backends enables superior accuracy in photon-budget-constrained scenarios beyond traditional image reconstruction methods.

Abstract: Illumination using correlated photon sources has been established as an approach to allowing high-fidelity images to be reconstructed from noisy camera frames by taking advantage of the knowledge that signal photons are spatially correlated whereas detector clicks due to noise are uncorrelated. However, in computer-vision tasks, the goal is often not ultimately to reconstruct an image, but to make inferences about a scene – such as what object is present. Here we show how correlated-photon illumination can be used to gain an advantage in a hybrid optical-electronic computer-vision pipeline for object recognition. We demonstrate correlation-aware training (CAT): end-to-end optimization of a trainable correlated-photon illumination source and a Transformer backend in a way that the Transformer can learn to benefit from the correlations, using a small number (<= 100) of shots. We show a classification accuracy enhancement of up to 15 percentage points over conventional, uncorrelated-illumination-based computer vision in ultra-low-light and noisy imaging conditions, as well as an improvement over using untrained correlated-photon illumination. Our work illustrates how specializing to a computer-vision task – object recognition – and training the pattern of photon correlations in conjunction with a digital backend allows us to push the limits of accuracy in highly photon-budget-constrained scenarios beyond existing methods focused on image reconstruction.

Method: ARGOS reformulates person search as interactive reasoning where agents receive vague witness statements and must decide what to ask, when to invoke spatial/temporal tools, and interpret ambiguous responses. It uses a Spatio-Temporal Topology Graph encoding camera connectivity and transition times. The benchmark includes 2,691 tasks across 14 real-world scenarios in three tracks: semantic perception, spatial reasoning, and temporal reasoning.

Result: Experiments with four LLM backbones show the benchmark is far from solved (best TWS: 0.383 on Track 2, 0.590 on Track 3). Ablations confirm removing domain-specific tools drops accuracy by up to 49.6 percentage points.

Conclusion: ARGOS presents a challenging benchmark for interactive reasoning in multi-camera person search, highlighting the importance of domain-specific tools and the current limitations of LLMs in handling information asymmetry and spatial-temporal reasoning.

Abstract: We introduce ARGOS, the first benchmark and framework that reformulates multi-camera person search as an interactive reasoning problem requiring an agent to plan, question, and eliminate candidates under information asymmetry. An ARGOS agent receives a vague witness statement and must decide what to ask, when to invoke spatial or temporal tools, and how to interpret ambiguous responses, all within a limited turn budget. Reasoning is grounded in a Spatio-Temporal Topology Graph (STTG) encoding camera connectivity and empirically validated transition times. The benchmark comprises 2,691 tasks across 14 real-world scenarios in three progressive tracks: semantic perception (Who), spatial reasoning (Where), and temporal reasoning (When). Experiments with four LLM backbones show the benchmark is far from solved (best TWS: 0.383 on Track 2, 0.590 on Track 3), and ablations confirm that removing domain-specific tools drops accuracy by up to 49.6 percentage points.

Method: The paper describes a challenge framework with open-source and closed-source tracks where participants submitted various approaches for CD-FSOD. The report analyzes submitted methods including innovative strategies that push performance frontiers in cross-domain few-shot detection.

Result: Strong community interest with 128 registered participants, 696 submissions, 31 active teams, and 19 teams submitting valid final results. The challenge successfully evaluated diverse approaches and identified methods that advance CD-FSOD performance.

Conclusion: The NTIRE 2026 CD-FSOD Challenge successfully promoted progress in cross-domain few-shot object detection, with community participation demonstrating innovative approaches to address the challenging problem of detecting objects in unseen domains with limited annotations.

Abstract: Cross-domain few-shot object detection (CD-FSOD) remains a challenging problem for existing object detectors and few-shot learning approaches, particularly when generalizing across distinct domains. As part of NTIRE 2026, we hosted the second CD-FSOD Challenge to systematically evaluate and promote progress in detecting objects in unseen target domains under limited annotation conditions. The challenge received strong community interest, with 128 registered participants and a total of 696 submissions. Among them, 31 teams actively participated, and 19 teams submitted valid final results. Participants explored a wide range of strategies, introducing innovative methods that push the performance frontier under both open-source and closed-source tracks. This report presents a detailed overview of the NTIRE 2026 CD-FSOD Challenge, including a summary of the submitted approaches and an analysis of the final results across all participating teams. Challenge Codes: https://github.com/ohMargin/NTIRE2026_CDFSOD.

Method: Developed SkyScraper, an iterative multi-agent workflow that geocodes news articles and synthesizes captions for corresponding satellite image sequences using agentic feedback to surface new multi-temporal events.

Result: SkyScraper successfully finds 5x more events than traditional geocoding methods and was used to curate a new multi-temporal captioning dataset with 5,000 sequences from a large database of global news articles.

Conclusion: Agentic feedback is an effective strategy for surfacing new multi-temporal events in satellite imagery, and the framework supports journalism and reporting efforts by automatically identifying imagery related to news events.

Abstract: Changes in satellite imagery often occur over multiple time steps. Despite the emergence of bi-temporal change captioning datasets, there is a lack of multi-temporal event captioning datasets (at least two images per sequence) in remote sensing. This gap exists because (1) searching for visible events in satellite imagery and (2) labeling multi-temporal sequences require significant time and labor. To address these challenges, we present SkyScraper, an iterative multi-agent workflow that geocodes news articles and synthesizes captions for corresponding satellite image sequences. Our experiments show that SkyScraper successfully finds 5x more events than traditional geocoding methods, demonstrating that agentic feedback is an effective strategy for surfacing new multi-temporal events in satellite imagery. We apply our framework to a large database of global news articles, curating a new multi-temporal captioning dataset with 5,000 sequences. By automatically identifying imagery related to news events, our work also supports journalism and reporting efforts.

Method: Cannot determine method as abstract is unavailable due to arXiv API rate limiting

Result: Cannot determine results as abstract is unavailable due to arXiv API rate limiting

Conclusion: Cannot determine conclusion as abstract is unavailable due to arXiv API rate limiting

Abstract: Failed to fetch summary for 2512.09664: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.09664&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes iBOT++ objective where unmasked tokens also contribute to loss, patch-level distillation that surprisingly improves alignment beyond teacher models, modified exponential moving average setup, and caption sampling strategy for synthetic captions.

Result: TIPSv2 achieves strong performance on 9 tasks across 20 datasets, generally on par with or better than recent vision encoder models, with improved patch-text alignment capabilities.

Conclusion: The proposed techniques significantly enhance vision-language pretraining, particularly for dense patch-text alignment, resulting in versatile models suitable for diverse downstream applications.

Abstract: Recent progress in vision-language pretraining has enabled significant improvements to many downstream computer vision applications, such as classification, retrieval, segmentation and depth prediction. However, a fundamental capability that these models still struggle with is aligning dense patch representations with text embeddings of corresponding concepts. In this work, we investigate this critical issue and propose novel techniques to enhance this capability in foundational vision-language models. First, we reveal that a patch-level distillation procedure significantly boosts dense patch-text alignment – surprisingly, the patch-text alignment of the distilled student model strongly surpasses that of the teacher model. This observation inspires us to consider modifications to pretraining recipes, leading us to propose iBOT++, an upgrade to the commonly-used iBOT masked image objective, where unmasked tokens also contribute directly to the loss. This dramatically enhances patch-text alignment of pretrained models. Additionally, to improve vision-language pretraining efficiency and effectiveness, we modify the exponential moving average setup in the learning recipe, and introduce a caption sampling strategy to benefit from synthetic captions at different granularities. Combining these components, we develop TIPSv2, a new family of image-text encoder models suitable for a wide range of downstream applications. Through comprehensive experiments on 9 tasks and 20 datasets, we demonstrate strong performance, generally on par with or better than recent vision encoder models. Code and models are released via our project page at https://gdm-tipsv2.github.io/ .

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2512.14732: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.14732&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Introduces Curvelet-based detection with wedge-level attention and scale-aware spatial masking to emphasize discriminative frequency components. Refined frequency cues are reconstructed and passed to modified pretrained Xception network for classification.

Result: Achieves 98.48% accuracy and 99.96% AUC on FaceForensics++ low compression, maintaining strong performance under high compression.

Conclusion: Curvelet Transform is effective for deepfake detection, offering robustness to compression and interpretability through frequency-domain analysis.

Abstract: The proliferation of sophisticated generative models has significantly advanced the realism of synthetic facial content, known as deepfakes, raising serious concerns about digital trust. Although modern deep learning-based detectors perform well, many rely on spatial-domain features that degrade under compression. This limitation has prompted a shift toward integrating frequency-domain representations with deep learning to improve robustness. Prior research has explored frequency transforms such as Discrete Cosine Transform (DCT), Fast Fourier Transform (FFT), and Wavelet Transform, among others. However, to the best of our knowledge, the Curvelet Transform, despite its superior directional and multiscale properties, remains entirely unexplored in the context of deepfake detection. In this work, we introduce a novel Curvelet-based detection approach that enhances feature quality through wedge-level attention and scale-aware spatial masking, both trained to selectively emphasize discriminative frequency components. The refined frequency cues are reconstructed and passed to a modified pretrained Xception network for classification. Evaluated on two compression qualities in the challenging FaceForensics++ dataset, our method achieves 98.48% accuracy and 99.96% AUC on FF++ low compression, while maintaining strong performance under high compression, demonstrating the efficacy and interpretability of Curvelet-informed forgery detection.

Method: Evaluated LLaVA-1.5-7B on POPE and ScienceQA-IMG using Expected Calibration Error (ECE), Brier score, and AURC under various pruning strategies (SCOPE with different saliency weights, saliency-only pruning, FastV, random pruning) across multiple token budgets.

Result: On POPE, coverage-based SCOPE achieved substantially lower ECE than unpruned model while maintaining similar accuracy; reducing saliency weight improved calibration at all token budgets with minimal accuracy impact. Saliency-based pruning worsened calibration, and FastV caused severe degradation. On ScienceQA-IMG, pruning reduced ECE with stable/slightly improved accuracy.

Conclusion: Visual token pruning should be evaluated for confidence quality, not just accuracy, especially for multimodal systems needing reliable decisions; coverage-based pruning can improve calibration while maintaining efficiency.

Abstract: Visual token pruning is a widely used strategy for efficient inference in multimodal large language models (MLLMs), but existing work mainly evaluates it with task accuracy. In this paper, we study how visual token pruning affects model calibration, that is, whether predicted confidence matches actual correctness. Using LLaVA-1.5-7B on POPE and ScienceQA-IMG, we evaluate Expected Calibration Error (ECE), Brier score, and AURC under several pruning strategies, including SCOPE with different saliency weights, saliency-only pruning, FastV, and random pruning, across multiple token budgets. Our results show that pruning does not simply trade reliability for efficiency. On POPE, a pure-coverage setting in SCOPE achieves substantially lower ECE than the full unpruned model while maintaining similar accuracy. An internal alpha-sweep further shows a consistent trend: reducing the saliency weight improves calibration at all tested token budgets, while accuracy changes only slightly. In contrast, saliency-based pruning leads to worse calibration, and real FastV causes severe performance degradation in our setting. On ScienceQA-IMG, pruning also reduces ECE, with accuracy remaining stable or slightly improving. We additionally study the gap power exponent in coverage-based selection and find that its default setting is not always optimal. Overall, our results suggest that visual token pruning should be evaluated not only by accuracy, but also by confidence quality, especially for multimodal systems that need reliable decisions.

Method: Uses structureless localization methods with simple image obfuscation based on common image operations, such as replacing RGB images with semantic segmentations. Shows that existing structureless pipelines require no special adjustments as modern feature matchers can match obfuscated images directly.

Result: Achieves state-of-the-art pose accuracy for privacy-preserving approaches on multiple datasets. The method is easy to implement and ensures privacy of both query images and scene representations.

Conclusion: Simple image obfuscation combined with structureless localization provides an effective privacy-preserving solution that maintains high accuracy while being straightforward to implement.

Abstract: Visual localization is the task of estimating the camera pose of an image relative to a scene representation. In practice, visual localization systems are often cloud-based. Naturally, this raises privacy concerns in terms of revealing private details through the images sent to the server or through the representations stored on the server. Privacy-preserving localization aims to avoid such leakage of private details. However, the resulting localization approaches are significantly more complex, slower, and less accurate than their non-privacy-preserving counterparts. In this paper, we consider structureless localization methods in the context of privacy preservation. Structureless methods represent the scene through a set of reference images with known camera poses and intrinsics. In contrast to existing methods proposing representations that are as privacy-preserving as possible, we study a simple image obfuscation approach based on common image operations, e.g., replacing RGB images with (semantic) segmentations. We show that existing structureless pipelines do not need any special adjustments, as modern feature matchers can match obfuscated images out of the box. The results are easy-to-implement pipelines that can ensure both the privacy of the query images and the scene representations. Detailed experiments on multiple datasets show that the resulting methods achieve state-of-the-art pose accuracy for privacy-preserving approaches.

Method: Used Atlas H&E-TME, an AI-powered application built on pathology foundation models, to perform tissue quality control, tissue segmentation, cell detection and classification, and spatial neighborhood analysis on 3,634 whole-slide images from TCGA across five cancer types.

Result: Created OpenTME dataset with over 4,500 quantitative readouts per slide at cell-level resolution, available for non-commercial academic research on Hugging Face, covering bladder, breast, colorectal, liver, and lung cancers.

Conclusion: OpenTME serves as a valuable resource for biomarker discovery, spatial biology research, and computational method development for tumor microenvironment analysis, with plans for future expansion.

Abstract: The tumor microenvironment (TME) plays a central role in cancer progression, treatment response, and patient outcomes, yet large-scale, consistent, and quantitative TME characterization from routine hematoxylin and eosin (H&E)-stained histopathology remains scarce. We introduce OpenTME, an open-access dataset of pre-computed TME profiles derived from 3,634 H&E-stained whole-slide images across five cancer types (bladder, breast, colorectal, liver, and lung cancer) from The Cancer Genome Atlas (TCGA). All outputs were generated using Atlas H&E-TME, an AI-powered application built on the Atlas family of pathology foundation models, which performs tissue quality control, tissue segmentation, cell detection and classification, and spatial neighborhood analysis, yielding over 4,500 quantitative readouts per slide at cell-level resolution. OpenTME is available for non-commercial academic research on Hugging Face. We will continue to expand OpenTME over time and anticipate it will serve as a resource for biomarker discovery, spatial biology research, and the development of computational methods for TME analysis.

Method: Proposes INST-Align with a coordinate-based deformation network coupled with a shared Canonical Expression Field (implicit neural representation mapping spatial coordinates to expression embeddings). Uses two-phase training: first establishes stable canonical embedding space, then jointly optimizes deformation and spatial-feature matching with cross-slice parameter sharing.

Result: Achieves state-of-the-art performance across nine datasets with mean OT Accuracy (0.702), NN Accuracy (0.719), and Chamfer distance reductions up to 94.9% on large-deformation sections compared to baselines. Produces biologically meaningful spatial embeddings and coherent 3D tissue reconstruction.

Conclusion: INST-Align successfully addresses coupled alignment and integration challenges in spatial transcriptomics through joint optimization, enabling accurate 3D reconstruction and meaningful spatial analysis of tissue organization.

Abstract: Spatial transcriptomics (ST) measures mRNA expression while preserving spatial organization, but multi-slice analysis faces two coupled difficulties: large non-rigid deformations across slices and inter-slice batch effects when alignment and integration are treated independently. We present INST-Align, an unsupervised pairwise framework that couples a coordinate-based deformation network with a shared Canonical Expression Field, an implicit neural representation mapping spatial coordinates to expression embeddings, for joint alignment and reconstruction. A two-phase training strategy first establishes a stable canonical embedding space and then jointly optimizes deformation and spatial-feature matching, enabling mutually constrained alignment and representation learning. Cross-slice parameter sharing of the canonical field regularizes ambiguous correspondences and absorbs batch variation. Across nine datasets, INST-Align achieves state-of-the-art mean OT Accuracy (0.702), NN Accuracy (0.719), and Chamfer distance, with Chamfer reductions of up to 94.9% on large-deformation sections relative to the strongest baseline. The framework also yields biologically meaningful spatial embeddings and coherent 3D tissue reconstruction. The code will be released after review phase.

Method: PC-MIL treats spatial extent of supervision as a design dimension, using fixed 20x features while varying MIL bag extent in millimeter units. It anchors supervision at clinically motivated 2mm scale and progressively mixes slide- and region-level supervision in controlled proportions for explicit train-context x test-context analysis.

Result: On 1,476 prostate WSIs from five public datasets for binary cancer detection, modest regional supervision improves cross-context performance, and balanced multi-context training stabilizes accuracy across slide and regional evaluation without sacrificing global performance.

Conclusion: Anatomical context is an independent axis of generalization in MIL, orthogonal to feature resolution. Supervision extent shapes MIL inductive bias and supports anatomically grounded WSI generalization.

Abstract: Whole-slide image (WSI) classification in computational pathology is commonly formulated as slide-level Multiple Instance Learning (MIL) with a single global bag representation. However, slide-level MIL is fundamentally underconstrained: optimizing only global labels encourages models to aggregate features without learning anatomically meaningful localization. This creates a mismatch between the scale of supervision and the scale of clinical reasoning. Clinicians assess tumor burden, focal lesions, and architectural patterns within millimeter-scale regions, whereas standard MIL is trained only to predict whether “somewhere in the slide there is cancer.” As a result, the model’s inductive bias effectively erases anatomical structure. We propose Progressive-Context MIL (PC-MIL), a framework that treats the spatial extent of supervision as a first-class design dimension. Rather than altering magnification, patch size, or introducing pixel-level segmentation, we decouple feature resolution from supervision scale. Using fixed 20x features, we vary MIL bag extent in millimeter units and anchor supervision at a clinically motivated 2mm scale to preserve comparable tumor burden and avoid confounding scale with lesion density. PC-MIL progressively mixes slide- and region-level supervision in controlled proportions, enabling explicit train-context x test-context analysis. On 1,476 prostate WSIs from five public datasets for binary cancer detection, we show that anatomical context is an independent axis of generalization in MIL, orthogonal to feature resolution: modest regional supervision improves cross-context performance, and balanced multi-context training stabilizes accuracy across slide and regional evaluation without sacrificing global performance. These results demonstrate that supervision extent shapes MIL inductive bias and support anatomically grounded WSI generalization.

Method: PR-MaGIC refines prompts via gradient flow derived from SAM’s mask decoder, integrating seamlessly into in-context segmentation frameworks with a simple top-1 selection strategy for stabilization.

Result: Extensive evaluations show PR-MaGIC consistently improves segmentation quality across various benchmarks, effectively mitigating inadequate prompts without requiring additional training or architectural modifications.

Conclusion: PR-MaGIC provides a theoretically grounded yet practical solution for prompt refinement in in-context segmentation, enhancing SAM-based segmentation without training overhead.

Abstract: Visual Foundation Models (VFMs) such as the Segment Anything Model (SAM) have significantly advanced broad use of image segmentation. However, SAM and its variants necessitate substantial manual effort for prompt generation and additional training for specific applications. Recent approaches address these limitations by integrating SAM into in-context (one/few shot) segmentation, enabling auto-prompting through semantic alignment between query and support images. Despite these efforts, they still generate sub-optimal prompts that degrade segmentation quality due to visual inconsistencies between support and query images. To tackle this limitation, we introduce PR-MaGIC (Prompt Refinement via Mask Decoder Gradient Flow for In-Context Segmentation), a training-free test-time framework that refines prompts via gradient flow derived from SAM’s mask decoder. PR-MaGIC seamlessly integrates into in-context segmentation frameworks, being theoretically grounded yet practically stabilized through a simple top-1 selection strategy that ensures robust performance across samples. Extensive evaluations demonstrate that PR-MaGIC consistently improves segmentation quality across various benchmarks, effectively mitigating inadequate prompts without requiring additional training or architectural modifications.

Method: Proposes Hesitation-Triggered Differential Calibration (HTDC) that identifies layer-wise hesitation (fluctuations in token preference across intermediate layers) as a signal of grounding instability. When triggered, HTDC contrasts the full branch with two lightweight probes: visual-nullification probe and semantic-nullification probe to suppress hallucination-prone candidates.

Result: Experiments on hallucination benchmarks show HTDC consistently reduces hallucinations while maintaining strong task accuracy, achieving favorable trade-off between effectiveness and computational overhead.

Conclusion: HTDC provides an efficient training-free decoding framework that selectively activates calibration only when needed, effectively reducing hallucinations in vision-language models without disrupting stable predictions.

Abstract: Large vision-language models (LVLMs) achieve strong multimodal performance, but still suffer from hallucinations caused by unstable visual grounding and over-reliance on language priors. Existing training-free decoding methods typically apply calibration at every decoding step, introducing unnecessary computation and potentially disrupting stable predictions. We address this problem by identifying layer-wise hesitation, a simple signal of grounding instability reflected by fluctuations in token preference across intermediate layers. Based on this observation, we propose Hesitation-Triggered Differential Calibration (HTDC), a training-free decoding framework that preserves standard full-branch inference and activates calibration only at hesitation-prone steps. When triggered, HTDC contrasts the full branch with two lightweight probes, a visual-nullification probe and a semantic-nullification probe, to suppress hallucination-prone candidates while avoiding unnecessary intervention on stable steps. Experiments on representative hallucination benchmarks show that HTDC consistently reduces hallucinations while maintaining strong task accuracy, achieving a favorable trade-off between effectiveness and computational overhead.

Method: Introduce VLM-Fix, a controlled benchmark over four abstract strategy games evaluating identical terminal board states under paired standard and inverse rule formulations. Test 14 open and closed VLMs, use prompt interventions (neutral vs. semantically loaded aliases), examine post-training effects, and apply late-layer activation steering.

Result: Accuracy consistently favors standard rules across all models, revealing a robust semantic-fixation gap. Neutral alias prompts narrow the inverse-rule gap while semantically loaded aliases reopen it. Post-training shows strong rule alignment, and late-layer activation steering can partially recover degraded performance.

Conclusion: VLMs exhibit semantic fixation that persists across models and interventions. This behavior is partly editable in late representations, suggesting opportunities for targeted interventions to improve model flexibility in rule interpretation.

Abstract: Large vision-language models (VLMs) often rely on familiar semantic priors, but existing evaluations do not cleanly separate perception failures from rule-mapping failures. We study this behavior as semantic fixation: preserving a default interpretation even when the prompt specifies an alternative, equally valid mapping. To isolate this effect, we introduce VLM-Fix, a controlled benchmark over four abstract strategy games that evaluates identical terminal board states under paired standard and inverse rule formulations. Across 14 open and closed VLMs, accuracy consistently favors standard rules, revealing a robust semantic-fixation gap. Prompt interventions support this mechanism: neutral alias prompts substantially narrow the inverse-rule gap, while semantically loaded aliases reopen it. Post-training is strongly rule-aligned: training on one rule improves same-rule transfer but hurts opposite-rule transfer, while joint-rule training improves broader transfer. To test external validity beyond synthetic games, we evaluate analogous defamiliarization interventions on VLMBias and observe the same qualitative pattern. Finally, late-layer activation steering partially recovers degraded performance, indicating that semantic-fixation errors are at least partly editable in late representations. Project page, code, and dataset available at https://maveryn.github.io/vlm-fix/.

Method: ViLL-E introduces a novel embedding generation mechanism with adaptive computation (thinking longer for complex videos, stopping early for easy ones). Uses three-stage training: 1) large-scale pre-training with video-caption pairs, 2) continual training on detailed-caption dataset, 3) task-specific fine-tuning on multi-task dataset covering Video QA, Temporal Localization, Video Retrieval, and Video-Text Matching.

Result: Significantly improves temporal localization (avg. 7% over other VideoLLMs) and video retrieval (up to 4% over dual encoder models). Achieves performance comparable to SotA specialized embedding models while remaining competitive on VideoQA. Unlocks new zero-shot capabilities: +5% over SotA in composed video retrieval and +2% over SotA in retrieval from long text.

Conclusion: ViLL-E demonstrates that a unified VideoLLM architecture with joint contrastive-generative training can effectively handle both generative and retrieval tasks, bridging the gap between VideoLLMs and specialized embedding models while enabling new zero-shot capabilities.

Abstract: Video Large Language Models (VideoLLMs) excel at video understanding tasks where outputs are textual, such as Video Question Answering and Video Captioning. However, they underperform specialized embedding-based models in Retrieval tasks, such as Text-toVideo Retrieval and Moment Retrieval. We introduce ViLL-E (Video-LLM-Embed), a unified VideoLLM architecture endowed with a novel embedding generation mechanism that allows the model to “think longer” for complex videos and stop early for easy ones. We train this model with a three-stage training methodology combining generative and contrastive learning: initial large-scale pre-training with video-caption pairs; followed by continual training on a smaller, detailed-caption dataset; and concluding with task-specific fine-tuning on a novel multi-task dataset covering Video QA, Temporal Localization, Video Retrieval, and Video-Text Matching. Our model significantly improves temporal localization (on avg. 7% over other VideoLLMs) and video retrieval (up to 4% over dual encoder models), achieving performance comparable to state-of-the-art specialized embedding models while remaining competitive on VideoQA tasks. Furthermore, our joint contrastive-generative training unlocks new zero-shot capabilities, significantly outperforming state-of-the-art methods in composed video retrieval (+5% over SotA) and retrieval from long text (+2% over SotA).

Method: Replaced the generic Stable Diffusion VAE with MedVAE (domain-specific autoencoder pretrained on 1.6M medical images) while keeping all other pipeline components fixed. Conducted controlled experiments across knee MRI, brain MRI, and chest X-ray datasets (n=1,820). Used wavelet decomposition to analyze frequency bands and performed ablations across inference schedules, prediction targets, and generative architectures.

Result: MedVAE yielded +2.91 to +3.29 dB PSNR improvement across all medical imaging modalities with large effect sizes (Cohen’s d = 1.37 to 1.86). Wavelet decomposition showed advantages in finest spatial frequency bands encoding anatomical fine structure. The performance gap remained stable (±0.15 dB) across ablations while hallucination rates remained comparable between methods.

Conclusion: The VAE choice, not the diffusion architecture, is the dominant constraint on reconstruction quality in medical image super-resolution. Autoencoder reconstruction quality predicts downstream SR performance (R²=0.67), suggesting domain-specific VAE selection should precede diffusion architecture search.

Abstract: Latent diffusion models for medical image super-resolution universally inherit variational autoencoders designed for natural photographs. We show that this default choice, not the diffusion architecture, is the dominant constraint on reconstruction quality. In a controlled experiment holding all other pipeline components fixed, replacing the generic Stable Diffusion VAE with MedVAE, a domain-specific autoencoder pretrained on more than 1.6 million medical images, yields +2.91 to +3.29 dB PSNR improvement across knee MRI, brain MRI, and chest X-ray (n = 1,820; Cohen’s d = 1.37 to 1.86, all p < 10^{-20}, Wilcoxon signed-rank). Wavelet decomposition localises the advantage to the finest spatial frequency bands encoding anatomically relevant fine structure. Ablations across inference schedules, prediction targets, and generative architectures confirm the gap is stable within plus or minus 0.15 dB, while hallucination rates remain comparable between methods (Cohen’s h < 0.02 across all datasets), establishing that reconstruction fidelity and generative hallucination are governed by independent pipeline components. These results provide a practical screening criterion: autoencoder reconstruction quality, measurable without diffusion training, predicts downstream SR performance (R^2 = 0.67), suggesting that domain-specific VAE selection should precede diffusion architecture search. Code and trained model weights are publicly available at https://github.com/sebasmos/latent-sr.

Method: Proposes VidTAG with a dual-encoder framework for frame-to-GPS retrieval using both self-supervised and language-aligned features. Introduces TempGeo module to align frame embeddings for temporal consistency, and GeoRefiner module (encoder-decoder architecture) to refine GPS features using aligned frame embeddings.

Result: Outperforms baselines with 20% improvement at 1 km threshold over GeoCLIP, and beats current State-of-the-Art by 25% on global coarse-grained video geolocalization (CityGuessr68k). Demonstrates ability to generate temporally consistent trajectories on Mapillary (MSLS) and GAMa datasets.

Conclusion: The approach enables fine-grained video geolocalization and lays a strong foundation for future research in this area, addressing limitations of existing methods through a practical GPS-based retrieval framework with temporal consistency mechanisms.

Abstract: The task of video geolocalization aims to determine the precise GPS coordinates of a video’s origin and map its trajectory; with applications in forensics, social media, and exploration. Existing classification-based approaches operate at a coarse city-level granularity and fail to capture fine-grained details, while image retrieval methods are impractical on a global scale due to the need for extensive image galleries which are infeasible to compile. Comparatively, constructing a gallery of GPS coordinates is straightforward and inexpensive. We propose VidTAG, a dual-encoder framework that performs frame-to-GPS retrieval using both self-supervised and language-aligned features. To address temporal inconsistencies in video predictions, we introduce the TempGeo module, which aligns frame embeddings, and the GeoRefiner module, an encoder-decoder architecture that refines GPS features using the aligned frame embeddings. Evaluations on Mapillary (MSLS) and GAMa datasets demonstrate our model’s ability to generate temporally consistent trajectories and outperform baselines, achieving a 20% improvement at the 1 km threshold over GeoCLIP. We also beat current State-of-the-Art by 25% on global coarse grained video geolocalization (CityGuessr68k). Our approach enables fine-grained video geolocalization and lays a strong foundation for future research. More details on the project webpage: https://parthpk.github.io/vidtag_webpage/

Method: Uses sparse MoE diffusion transformer architecture with Expert-Choice Routing (17B total parameters, 64 experts per layer), streamlined architecture excluding text tokens from transformer backbone, joint attention for text KV sharing, decoupled routing for stability, large-scale training corpus (1.5B pairs), progressive resolution curriculum, and Muon optimizer.

Result: Matches or exceeds leading models on GenEval, DPG-Bench, and OneIG-Bench while activating only ~2B parameters per forward pass, achieving performance of larger models at fraction of inference cost without post-training optimization.

Conclusion: Sparse MoE scaling is highly effective for high-quality image generation, enabling state-of-the-art performance with efficient inference, and the model is released as first fully open-source MoE diffusion model at this quality level.

Abstract: We present Nucleus-Image, a text-to-image generation model that establishes a new Pareto frontier in quality-versus-efficiency by matching or exceeding leading models on GenEval, DPG-Bench, and OneIG-Bench while activating only approximately 2B parameters per forward pass. Nucleus-Image employs a sparse mixture-of-experts (MoE) diffusion transformer architecture with Expert-Choice Routing that scales total model capacity to 17B parameters across 64 routed experts per layer. We adopt a streamlined architecture optimized for inference efficiency by excluding text tokens from the transformer backbone entirely and using joint attention that enables text KV sharing across timesteps. To improve routing stability when using timestep modulation, we introduce a decoupled routing design that separates timestep-aware expert assignment from timestep-conditioned expert computation. We construct a large-scale training corpus of 1.5B high-quality training pairs spanning 700M unique images through multi-stage filtering, deduplication, aesthetic tiering, and caption curation. Training follows a progressive resolution curriculum (256 to 512 to 1024) with multi-aspect-ratio bucketing at every stage, coupled with progressive sparsification of the expert capacity factor. We adopt the Muon optimizer and share our parameter grouping recipe tailored for diffusion models with timestep modulation. Nucleus-Image demonstrates that sparse MoE scaling is a highly effective path to high-quality image generation, reaching the performance of models with significantly larger active parameter budgets at a fraction of the inference cost. These results are achieved without post-training optimization of any kind: no reinforcement learning, no direct preference optimization, and no human preference tuning. We release the training recipe, making Nucleus-Image the first fully open-source MoE diffusion model at this quality.

Method: Proposes DS-IEQA with two key components: 1) Feedback-Driven Metric Prompt Optimization (FDMPO) to automatically refine metric definitions via probabilistic feedback, and 2) Token-Decoupled Distance Regression Loss (TDRL) which decouples numerical tokens from language modeling to explicitly model score continuity through expected distance minimization.

Result: The method achieves superior performance, ranking 4th in the 2026 NTIRE X-AIGC Quality Assessment Track 2 without any additional training data.

Conclusion: DS-IEQA provides an effective unified framework for IEQA that addresses limitations of existing MLLM-based approaches by jointly learning evaluation criteria and score representations, enabling better alignment with human assessment.

Abstract: Recent advances in image editing have heightened the need for reliable Image Editing Quality Assessment (IEQA). Unlike traditional methods, IEQA requires complex reasoning over multimodal inputs and multi-dimensional assessments. Existing MLLM-based approaches often rely on human heuristic prompting, leading to two key limitations: rigid metric prompting and distance-agnostic score modeling. These issues hinder alignment with implicit human criteria and fail to capture the continuous structure of score spaces. To address this, we propose Define-and-Score Image Editing Quality Assessment (DS-IEQA), a unified framework that jointly learns evaluation criteria and score representations. Specifically, we introduce Feedback-Driven Metric Prompt Optimization (FDMPO) to automatically refine metric definitions via probabilistic feedback. Furthermore, we propose Token-Decoupled Distance Regression Loss (TDRL), which decouples numerical tokens from language modeling to explicitly model score continuity through expected distance minimization. Extensive experiments show our method’s superior performance; it ranks 4th in the 2026 NTIRE X-AIGC Quality Assessment Track 2 without any additional training data.

Method: 1) Curvature-aware dynamic budgeting that profiles generation trajectory acceleration to allocate computation budget during critical semantic transitions. 2) Hardware-aligned grouped approximations instead of global homogenizing estimations. 3) Stochastic selection bias in attention routing to soften rigid boundaries and eliminate selection oscillation.

Result: PASA achieves substantial inference acceleration while consistently producing remarkably fluid and structurally stable video sequences in evaluations on leading video diffusion models.

Conclusion: PASA provides an effective training-free framework for efficient and temporally smooth video generation by addressing the limitations of existing sparse attention methods through dynamic budgeting and stochastic routing.

Abstract: Video Diffusion Transformers have revolutionized high-fidelity video generation but suffer from the massive computational burden of self-attention. While sparse attention provides a promising acceleration solution, existing methods frequently provoke severe visual flickering caused by static sparsity patterns and deterministic block routing. To resolve these limitations, we propose Precision-Allocated Sparse Attention (PASA), a training-free framework designed for highly efficient and temporally smooth video generation. First, we implement a curvature-aware dynamic budgeting mechanism. By profiling the generation trajectory acceleration across timesteps, we elastically allocate the exact-computation budget to secure high-precision processing strictly during critical semantic transitions. Second, we replace global homogenizing estimations with hardware-aligned grouped approximations, successfully capturing fine-grained local variations while maintaining peak compute throughput. Finally, we incorporate a stochastic selection bias into the attention routing mechanism. This probabilistic approach softens rigid selection boundaries and eliminates selection oscillation, effectively eradicating the localized computational starvation that drives temporal flickering. Extensive evaluations on leading video diffusion models demonstrate that PASA achieves substantial inference acceleration while consistently producing remarkably fluid and structurally stable video sequences.

Method: 1) Created BarbieGait synthetic dataset by mapping real subjects to virtual engine for controlled clothing changes while preserving gait identity. 2) Proposed GaitCLIF (Gait-oriented CLoth-Invariant Feature) model to learn cloth-invariant features despite increased intra-class variance from clothing diversity.

Result: Method significantly improves cross-clothing performance on BarbieGait and existing gait benchmarks. BarbieGait enables validation of cross-clothing issues difficult to verify with real-world data.

Conclusion: BarbieGait advances cross-clothing gait recognition capabilities and promotes related research through extensive synthetic cross-clothing data. GaitCLIF provides robust baseline for cloth-invariant feature learning.

Abstract: Gait recognition, as a reliable biometric technology, has seen rapid development in recent years while it faces significant challenges caused by diverse clothing styles in the real world. This paper introduces BarbieGait, a synthetic gait dataset where real-world subjects are uniquely mapped into a virtual engine to simulate extensive clothing changes while preserving their gait identity information. As a pioneering work, BarbieGait provides a controllable gait data generation method, enabling the production of large datasets to validate cross-clothing issues that are difficult to verify with real-world data. However, the diversity of clothing increases intra-class variance and makes one of the biggest challenges to learning cloth-invariant features under varying clothing conditions. Therefore, we propose GaitCLIF (Gait-oriented CLoth-Invariant Feature) as a robust baseline model for cross-clothing gait recognition. Through extensive experiments, we validate that our method significantly improves cross-clothing performance on BarbieGait and the existing popular gait benchmarks. We believe that BarbieGait, with its extensive cross-clothing gait data, will further advance the capabilities of gait recognition in cross-clothing scenarios and promote progress in related research.

Method: 1) Creates a fine-grained taxonomy of 3DGS artifacts and builds a large training set of 107.5K paired video clips; 2) Uses a homogeneous dual-model paradigm with video diffusion backbone; 3) Employs an isomorphic predictor to localize structural defects via artifact heatmap; 4) Guides restoration through Artifact-Aware Triplet Fusion mechanism for intensity-guided spatio-temporal repair.

Result: Extensive experiments show ArtifactWorld achieves state-of-the-art performance in sparse novel view synthesis and robust 3D reconstruction.

Conclusion: ArtifactWorld effectively resolves 3DGS artifact repair through systematic data expansion and a homogeneous dual-model paradigm, demonstrating superior performance in 3D reconstruction and novel view synthesis tasks.

Abstract: 3D Gaussian Splatting (3DGS) delivers high-fidelity real-time rendering but suffers from geometric and photometric degradations under sparse-view constraints. Current generative restoration approaches are often limited by insufficient temporal coherence, a lack of explicit spatial constraints, and a lack of large-scale training data, resulting in multi-view inconsistencies, erroneous geometric hallucinations, and limited generalization to diverse real-world artifact distributions. In this paper, we present ArtifactWorld, a framework that resolves 3DGS artifact repair through systematic data expansion and a homogeneous dual-model paradigm. To address the data bottleneck, we establish a fine-grained phenomenological taxonomy of 3DGS artifacts and construct a comprehensive training set of 107.5K diverse paired video clips to enhance model robustness. Architecturally, we unify the restoration process within a video diffusion backbone, utilizing an isomorphic predictor to localize structural defects via an artifact heatmap. This heatmap then guides the restoration through an Artifact-Aware Triplet Fusion mechanism, enabling precise, intensity-guided spatio-temporal repair within native self-attention. Extensive experiments demonstrate that ArtifactWorld achieves state-of-the-art performance in sparse novel view synthesis and robust 3D reconstruction. Code and dataset will be made public.

Method: Two-stage approach: 1) Affective Semantic Injection (ASI) establishes affective knowledge alignment using facial Action Units and retrieval-augmented prompt generation via large-scale visual-language models; 2) Adaptive Reinforcement Diffusion (ARD) integrates text-conditioned image-to-video diffusion with reinforcement learning, using inter-frame conditional guidance and multi-objective reward functions.

Result: Extensive experiments show ARGen substantially enhances synthesis fidelity and improves recognition performance, establishing an interpretable and generalizable generative augmentation paradigm.

Conclusion: ARGen provides an effective framework for dynamic facial expression generation that addresses data scarcity issues in emotion recognition through interpretable generative augmentation.

Abstract: Dynamic facial expression recognition in the wild remains challenging due to data scarcity and long-tail distributions, which hinder models from effectively learning the temporal dynamics of scarce emotions. To address these limitations, we propose ARGen, an Affect-Reinforced Generative Augmentation Framework that enables data-adaptive dynamic expression generation for robust emotion perception. ARGen operates in two stages: Affective Semantic Injection (ASI) and Adaptive Reinforcement Diffusion (ARD). The ASI stage establishes affective knowledge alignment through facial Action Units and employs a retrieval-augmented prompt generation strategy to synthesize consistent and fine-grained affective descriptions via large-scale visual-language models, thereby injecting interpretable emotional priors into the generation process. The ARD stage integrates text-conditioned image-to-video diffusion with reinforcement learning, introducing inter-frame conditional guidance and a multi-objective reward function to jointly optimize expression naturalness, facial integrity, and generative efficiency. Extensive experiments on both generation and recognition tasks verify that ARGen substantially enhances synthesis fidelity and improves recognition performance, establishing an interpretable and generalizable generative augmentation paradigm for vision-based affective computing.

Method: SDAR-Net decouples image features into dynamic degradation style embeddings and static scene structural representations. It introduces an adaptive routing mechanism that evaluates style features and predicts soft weights at different enhancement states to guide weighted fusion of representations for personalized restoration.

Result: SDAR-Net achieves state-of-the-art performance with PSNR of 25.72 dB on real-world benchmarks and demonstrates utility in downstream vision tasks.

Conclusion: The proposed adaptive enhancement framework effectively addresses the limitations of uniform mapping approaches by providing personalized restoration based on specific degradation styles, leading to superior performance on underwater image enhancement.

Abstract: Underwater Image Enhancement (UIE) is essential for robust visual perception in marine applications. However, existing methods predominantly rely on uniform mapping tailored to average dataset distributions, leading to over-processing mildly degraded images or insufficient recovery for severe ones. To address this challenge, we propose a novel adaptive enhancement framework, SDAR-Net. Unlike existing uniform paradigms, it first decouples specific degradation styles from the input and subsequently modulates the enhancement process adaptively. Specifically, since underwater degradation primarily shifts the appearance while keeping the scene structure, SDAR-Net formulates image features into dynamic degradation style embeddings and static scene structural representations through a carefully designed training framework. Subsequently, we introduce an adaptive routing mechanism. By evaluating style features and adaptively predicting soft weights at different enhancement states, it guides the weighted fusion of the corresponding image representations, accurately satisfying the adaptive restoration demands of each image. Extensive experiments show that SDAR-Net achieves a new state-of-the-art (SOTA) performance with a PSNR of 25.72 dB on real-world benchmark, and demonstrates its utility in downstream vision tasks. Our code is available at https://github.com/WHU-USI3DV/SDAR-Net.

Method: Three interconnected components: 1) Gradient-Aware Parallax Warping (GAPW) for smooth occlusion regions, 2) Parallax-Based Dual Projection (PBDP) to generate stereo inpainting pairs without stereo videos, 3) Sparsity-Aware Stereo Inpainting (SASI) that reduces redundant tokens for 10.7x speedup.

Result: Achieves real-time processing of HD videos (768x1280) at 25 FPS on single A100 GPU with over 70% token reduction and comparable quality to full-computation methods.

Conclusion: The proposed framework addresses stereo video inpainting challenges by combining geometric consistency, data generation without stereo videos, and efficient computation through sparsity awareness.

Abstract: Stereo video inpainting, which aims to fill the occluded regions of warped videos with visually coherent content while maintaining temporal consistency, remains a challenging open problem. The regions to be filled are scattered along object boundaries and occupy only a small fraction of each frame, leading to two key challenges. First, existing approaches perform poorly on such tasks due to the scarcity of high-quality stereo inpainting datasets, which limits their ability to learn effective inpainting priors. Second, these methods apply equal processing to all regions of the frame, even though most pixels require no modification, resulting in substantial redundant computation. To address these issues, we introduce three interconnected components. We first propose Gradient-Aware Parallax Warping (GAPW), which leverages backward warping and the gradient of the coordinate mapping function to obtain continuous edges and smooth occlusion regions. Then, a Parallax-Based Dual Projection (PBDP) strategy is introduced, which incorporates GAPW to produce geometrically consistent stereo inpainting pairs and accurate occlusion masks without requiring stereo videos. Finally, we present Sparsity-Aware Stereo Inpainting (SASI), which reduces over 70% of redundant tokens, achieving a 10.7x speedup during diffusion inference and delivering results comparable to its full-computation counterpart, enabling real-time processing of HD (768 x 1280) videos at 25 FPS on a single A100 GPU.

Method: Proposes MAST with four modules: 1) Layout-preserving Query Anchoring to maintain semantic structure, 2) Logit-level Attention Mass Allocation to distribute attention across spatial regions, 3) Sharpness-aware Temperature Scaling to restore attention sharpness, and 4) Discrepancy-aware Detail Injection to compensate for high-frequency detail losses.

Result: Extensive experiments show MAST effectively mitigates boundary artifacts and maintains structural consistency while preserving texture fidelity and spatial coherence, even as the number of applied styles increases.

Conclusion: MAST provides an effective training-free solution for multi-style transfer that overcomes limitations of existing diffusion-based methods by explicitly controlling content-style interactions through attention mechanism modifications.

Abstract: Style transfer aims to render a content image with the visual characteristics of a reference style while preserving its underlying semantic layout and structural geometry. While recent diffusion-based models demonstrate strong stylization capabilities by leveraging powerful generative priors and controllable internal representations, they typically assume a single global style. Extending them to multi-style scenarios often leads to boundary artifacts, unstable stylization, and structural inconsistency due to interference between multiple style representations. To overcome these limitations, we propose MAST (Mask-Guided Attention Mass Allocation for Training-Free Multi-Style Transfer), a novel training-free framework that explicitly controls content-style interactions within the diffusion attention mechanism. To achieve artifact-free and structure-preserving stylization, MAST integrates four connected modules. First, Layout-preserving Query Anchoring prevents global layout collapse by firmly anchoring the semantic structure using content queries. Second, Logit-level Attention Mass Allocation deterministically distributes attention probability mass across spatial regions, seamlessly fusing multiple styles without boundary artifacts. Third, Sharpness-aware Temperature Scaling restores the attention sharpness degraded by multi-style expansion. Finally, Discrepancy-aware Detail Injection adaptively compensates for localized high-frequency detail losses by measuring structural discrepancies. Extensive experiments demonstrate that MAST effectively mitigates boundary artifacts and maintains structural consistency, preserving texture fidelity and spatial coherence even as the number of applied styles increases.

Method: A two-branch neural network with a reference branch extracting structural/textural information from the original high-quality key photo, and a main branch restoring the reselected frame using reference guidance. Includes a unified Motion Alignment module for spatial alignment at both latent and image levels.

Result: Experiments on real and synthetic Live Photos show significant improvements in perceptual quality and fidelity over existing solutions, especially in scenes with fast motion or complex structures.

Conclusion: LiveMoments effectively addresses the quality gap in reselected Live Photo frames through reference-guided restoration with motion alignment, providing a practical solution for enhancing user-selected alternative frames.

Abstract: Live Photo captures both a high-quality key photo and a short video clip to preserve the precious dynamics around the captured moment. While users may choose alternative frames as the key photo to capture better expressions or timing, these frames often exhibit noticeable quality degradation, as the photo capture ISP pipeline delivers significantly higher image quality than the video pipeline. This quality gap highlights the need for dedicated restoration techniques to enhance the reselected key photo. To this end, we propose LiveMoments, a reference-guided image restoration framework tailored for the reselected key photo in Live Photos. Our method employs a two-branch neural network: a reference branch that extracts structural and textural information from the original high-quality key photo, and a main branch that restores the reselected frame using the guidance provided by the reference branch. Furthermore, we introduce a unified Motion Alignment module that incorporates motion guidance for spatial alignment at both the latent and image levels. Experiments on real and synthetic Live Photos demonstrate that LiveMoments significantly improves perceptual quality and fidelity over existing solutions, especially in scenes with fast motion or complex structures. Our code is available at https://github.com/OpenVeraTeam/LiveMoments.

Method: LoRA-based Pairwise Training with targeted finetuning of visual foundation model, distortion/size simulations to match validation/test distributions, and pairwise training to decouple generalization and robustness optimization

Result: Achieved 3rd place in NTIRE Robust AI-Generated Image Detection in the Wild challenge, demonstrating superior robustness to distortions

Conclusion: The proposed LPT strategy effectively addresses the vulnerability of AIGI detectors to real-world distortions through targeted finetuning and distribution simulation

Abstract: The proliferation of highly realistic AI-Generated Image (AIGI) has necessitated the development of practical detection methods. While current AIGI detectors perform admirably on clean datasets, their detection performance frequently decreases when deployed “in the wild”, where images are subjected to unpredictable, complex distortions. To resolve the critical vulnerability, we propose a novel LoRA-based Pairwise Training (LPT) strategy designed specifically to achieve robust detection for AIGI under severe distortions. The core of our strategy involves the targeted finetuning of a visual foundation model, the deliberate simulation of data distribution during the training phase, and a unique pairwise training process. Specifically, we introduce distortion and size simulations to better fit the distribution from the validation and test sets. Based on the strong visual representation capability of the visual foundation model, we finetune the model to achieve AIGI detection. The pairwise training is utilized to improve the detection via decoupling the generalization and robustness optimization. Experiments show that our approach secured the 3th placement in the NTIRE Robust AI-Generated Image Detection in the Wild challenge

Method: Uses 3D foundational generative model’s latent features at two scales: denoised global latent vector for structural guidance and latent images from volumetric features for view-dependent geometry. Multi-scale 3D adapter injects features via cross-attention into base video model.

Result: Superior visual quality, shape realism, and multi-view consistency compared to SOTA methods, robust generalization to complex camera trajectories and in-the-wild images.

Conclusion: 3D shape priors from foundational models effectively address geometric consistency in orbital video generation, enabling realistic long-range extrapolation from single images.

Abstract: We present a novel method for generating geometrically realistic and consistent orbital videos from a single image of an object. Existing video generation works mostly rely on pixel-wise attention to enforce view consistency across frames. However, such mechanism does not impose sufficient constraints for long-range extrapolation, e.g. rear-view synthesis, in which pixel correspondences to the input image are limited. Consequently, these works often fail to produce results with a plausible and coherent structure. To tackle this issue, we propose to leverage rich shape priors from a 3D foundational generative model as an auxiliary constraint, motivated by its capability of modeling realistic object shape distributions learned from large 3D asset corpora. Specifically, we prompt the video generation with two scales of latent features encoded by the 3D foundation model: (i) a denoised global latent vector as an overall structural guidance, and (ii) a set of latent images projected from volumetric features to provide view-dependent and fine-grained geometry details. In contrast to commonly used 2.5D representations such as depth or normal maps, these compact features can model complete object shapes, and help to improve inference efficiency by avoiding explicit mesh extraction. To achieve effective shape conditioning, we introduce a multi-scale 3D adapter to inject feature tokens to the base video model via cross-attention, which retains its capabilities from general video pretraining and enables a simple and model-agonistic fine-tuning process. Extensive experiments on multiple benchmarks show that our method achieves superior visual quality, shape realism and multi-view consistency compared to state-of-the-art methods, and robustly generalizes to complex camera trajectories and in-the-wild images.

Method: Multi-task image-to-image Schrödinger Bridge framework that formulates instance segmentation as distribution-based image-to-image generation. Uses boundary-aware supervision through reverse distance maps and deterministic inference for stable predictions.

Result: Achieves competitive or superior performance on PanNuke dataset without SAM pre-training or additional post-processing. Shows robustness on MoNuSeg dataset with limited training data.

Conclusion: Schrödinger Bridge-based image-to-image generation provides an effective framework for cell instance segmentation, offering a principled alternative to traditional deterministic approaches.

Abstract: Existing cell instance segmentation pipelines typically combine deterministic predictions with post-processing, which imposes limited explicit constraints on the global structure of instance masks. In this work, we propose a multi-task image-to-image Schrödinger Bridge framework that formulates instance segmentation as a distribution-based image-to-image generation problem. Boundary-aware supervision is integrated through a reverse distance map, and deterministic inference is employed to produce stable predictions. Experimental results on the PanNuke dataset demonstrate that the proposed method achieves competitive or superior performance without relying on SAM pre-training or additional post-processing. Additional results on the MoNuSeg dataset show robustness under limited training data. These findings indicate that Schrödinger Bridge-based image-to-image generation provides an effective framework for cell instance segmentation.

Method: Proposes Reliability-aware Self-Gated State Space Model (RSGMamba) with RSGMB blocks that explicitly model modality reliability and dynamically regulate cross-modal interactions via self-gating. Includes Local Cross-Gated Modulation (LCGM) for fine-grained spatial detail refinement.

Result: Achieves SOTA on RGB-D (NYUDepth V2: 58.8% mIoU, SUN-RGBD: 54.0% mIoU) and RGB-T (MFNet: 61.1% mIoU, PST900: 88.9% mIoU) benchmarks with only 48.6M parameters, outperforming prior methods by up to +1.6%.

Conclusion: Explicit modeling of modality reliability through self-gated state space models significantly improves multimodal semantic segmentation performance, validating the effectiveness of reliability-aware fusion strategies.

Abstract: Multimodal semantic segmentation has emerged as a powerful paradigm for enhancing scene understanding by leveraging complementary information from multiple sensing modalities (e.g., RGB, depth, and thermal). However, existing cross-modal fusion methods often implicitly assume that all modalities are equally reliable, which can lead to feature degradation when auxiliary modalities are noisy, misaligned, or incomplete. In this paper, we revisit cross-modal fusion from the perspective of modality reliability and propose a novel framework termed the Reliability-aware Self-Gated State Space Model (RSGMamba). At the core of our method is the Reliability-aware Self-Gated Mamba Block (RSGMB), which explicitly models modality reliability and dynamically regulates cross-modal interactions through a self-gating mechanism. Unlike conventional fusion strategies that indiscriminately exchange information across modalities, RSGMB enables reliability-aware feature selection and enhancing informative feature aggregation. In addition, a lightweight Local Cross-Gated Modulation (LCGM) is incorporated to refine fine-grained spatial details, complementing the global modeling capability of RSGMB. Extensive experiments demonstrate that RSGMamba achieves state-of-the-art performance on both RGB-D and RGB-T semantic segmentation benchmarks, resulting 58.8% / 54.0% mIoU on NYUDepth V2 and SUN-RGBD (+0.4% / +0.7% over prior best), and 61.1% / 88.9% mIoU on MFNet and PST900 (up to +1.6%), with only 48.6M parameters, thereby validating the effectiveness and superiority of the proposed approach.

Method: APEX extracts adversarial correction signals endogenously from flow models through condition shifting. A transformation creates a shifted condition branch whose velocity field serves as an independent estimator of the generation distribution, providing GAN-aligned gradients without external discriminators.

Result: APEX’s 0.6B model surpasses FLUX-Schnell 12B in one-step quality. With LoRA tuning on Qwen-Image 20B, it achieves GenEval score of 0.89 at NFE=1 in 6 hours, beating the original 50-step teacher (0.87) with 15.33× inference speedup.

Conclusion: APEX provides a plug-and-play discriminator-free framework for efficient one-step text-to-image synthesis that maintains high fidelity while being architecture-preserving and compatible with parameter-efficient tuning methods like LoRA.

Abstract: The push for efficient text to image synthesis has moved the field toward one step sampling, yet existing methods still face a three way tradeoff among fidelity, inference speed, and training efficiency. Approaches that rely on external discriminators can sharpen one step performance, but they often introduce training instability, high GPU memory overhead, and slow convergence, which complicates scaling and parameter efficient tuning. In contrast, regression based distillation and consistency objectives are easier to optimize, but they typically lose fine details when constrained to a single step. We present APEX, built on a key theoretical insight: adversarial correction signals can be extracted endogenously from a flow model through condition shifting. Using a transformation creates a shifted condition branch whose velocity field serves as an independent estimator of the model’s current generation distribution, yielding a gradient that is provably GAN aligned, replacing the sample dependent discriminator terms that cause gradient vanishing. This discriminator free design is architecture preserving, making APEX a plug and play framework compatible with both full parameter and LoRA based tuning. Empirically, our 0.6B model surpasses FLUX-Schnell 12B (20$\times$ more parameters) in one step quality. With LoRA tuning on Qwen-Image 20B, APEX reaches a GenEval score of 0.89 at NFE=1 in 6 hours, surpassing the original 50-step teacher (0.87) and providing a 15.33$\times$ inference speedup. Code is available https://github.com/LINs-lab/APEX.

Method: HyperLiDAR uses Hyperdimensional Computing (HDC) for fast learning and high efficiency, inspired by human brain processing. It includes a buffer selection strategy to focus learning on the most informative points, addressing the high data volume bottleneck in LiDAR scans.

Result: HyperLiDAR outperforms or achieves comparable adaptation performance to state-of-the-art segmentation methods while achieving up to 13.8x speedup in retraining on two LiDAR segmentation benchmarks and two representative devices.

Conclusion: HyperLiDAR demonstrates that HDC-based approaches can provide efficient, lightweight solutions for on-device LiDAR segmentation adaptation, making real-world deployment more feasible under edge computing constraints.

Abstract: LiDAR semantic segmentation plays a pivotal role in 3D scene understanding for edge applications such as autonomous driving. However, significant challenges remain for real-world deployments, particularly for on-device post-deployment adaptation. Real-world environments can shift as the system navigates through different locations, leading to substantial performance degradation without effective and timely model adaptation. Furthermore, edge systems operate under strict computational and energy constraints, making it infeasible to adapt conventional segmentation models (based on large neural networks) directly on-device. To address the above challenges, we introduce HyperLiDAR, the first lightweight, post-deployment LiDAR segmentation framework based on Hyperdimensional Computing (HDC). The design of HyperLiDAR fully leverages the fast learning and high efficiency of HDC, inspired by how the human brain processes information. To further improve the adaptation efficiency, we identify the high data volume per scan as a key bottleneck and introduce a buffer selection strategy that focuses learning on the most informative points. We conduct extensive evaluations on two state-of-the-art LiDAR segmentation benchmarks and two representative devices. Our results show that HyperLiDAR outperforms or achieves comparable adaptation performance to state-of-the-art segmentation methods, while achieving up to a 13.8x speedup in retraining.

Method: Proposes a unified synthetic data generation pipeline that automatically produces unlimited multimodal video data with rich supervision across multiple tasks (object counting, QA, segmentation). Also introduces VQA-based fine-tuning strategy that trains models to answer structured questions about visual content rather than using simple captions.

Result: Models trained predominantly on synthetic data generalize effectively to real-world datasets, often outperforming traditionally trained counterparts across three challenging tasks: video object counting, video-based visual question answering, and video object segmentation.

Conclusion: Unified synthetic data pipelines offer a scalable alternative to expensive real-world annotation for multimodal video understanding, demonstrating strong generalization capabilities to real-world data.

Abstract: Training multimodal large language models (MLLMs) for video understanding requires large-scale annotated data spanning diverse tasks such as object counting, question answering, and segmentation. However, collecting and annotating multimodal video data in real-world is costly, slow, and inherently limited in diversity and coverage. To address this challenge, we propose a unified synthetic data generation pipeline capable of automatically producing unlimited multimodal video data with rich and diverse supervision. Our framework supports multiple task formats within a single pipeline, enabling scalable and consistent data creation across tasks. To further enhance reasoning ability, we introduce a VQA-based fine-tuning strategy that trains models to answer structured questions about visual content rather than relying solely on captions or simple instructions. This formulation encourages deeper visual grounding and reasoning. We evaluate our approach in three challenging tasks: video object counting, video-based visual question answering, and video object segmentation. Experimental results demonstrate that models trained predominantly on synthetic data generalize effectively to real-world datasets, often outperforming traditionally trained counterparts. Our findings highlight the potential of unified synthetic data pipelines as a scalable alternative to expensive real-world annotation for multimodal video understanding.

Method: FASA extracts manipulation-sensitive frequency cues via adaptive dual-band DCT, learns manipulation-aware semantic priors through patch-level contrastive alignment on frozen CLIP representations, injects these priors into a hierarchical frequency pathway using semantic-frequency side adapters, and employs a prototype-guided, frequency-gated mask decoder to integrate semantic consistency with boundary-aware localization.

Result: Extensive experiments on OpenSDI and traditional manipulation benchmarks show state-of-the-art localization performance, strong cross-generator and cross-dataset generalization, and robustness under common image degradations.

Conclusion: FASA successfully bridges the micro-macro gap in image manipulation localization by unifying frequency-based forensic analysis with semantic understanding, providing a robust solution for detecting both traditional and diffusion-generated manipulations.

Abstract: As generative image editing advances, image manipulation localization (IML) must handle both traditional manipulations with conspicuous forensic artifacts and diffusion-generated edits that appear locally realistic. Existing methods typically rely on either low-level forensic cues or high-level semantics alone, leading to a fundamental micro–macro gap. To bridge this gap, we propose FASA, a unified framework for localizing both traditional and diffusion-generated manipulations. Specifically, we extract manipulation-sensitive frequency cues through an adaptive dual-band DCT module and learn manipulation-aware semantic priors via patch-level contrastive alignment on frozen CLIP representations. We then inject these priors into a hierarchical frequency pathway through a semantic-frequency side adapter for multi-scale feature interaction, and employ a prototype-guided, frequency-gated mask decoder to integrate semantic consistency with boundary-aware localization for tampered region prediction. Extensive experiments on OpenSDI and multiple traditional manipulation benchmarks demonstrate state-of-the-art localization performance, strong cross-generator and cross-dataset generalization, and robust performance under common image degradations.

Method: Proposes HiCE (Hand-informed Context Enhanced) module using cross-attention to leverage spatiotemporal features from hand regions and surrounding context, plus grasp-aware loss with soft labels emphasizing hand pose patterns and movement dynamics.

Result: Outperforms state-of-the-art event-spotting baselines by 16.91% average precision on TouchMoment dataset (4,021 videos, 8,456 contact moments) under strict 2-frame tolerance evaluation.

Conclusion: The approach effectively addresses challenges in precise contact moment detection through hand-informed context modeling and grasp-aware training, enabling accurate frame-level contact detection in egocentric videos.

Abstract: We address the challenging task of detecting the precise moment when hands make contact with objects in egocentric videos. This frame-level detection is crucial for augmented reality, human-computer interaction, assistive technologies, and robot learning applications, where contact onset signals action initiation or completion. Temporally precise detection is particularly challenging due to subtle hand motion variations near contact, frequent occlusions, fine-grained manipulation patterns, and the inherent motion dynamics of first-person perspectives. To tackle these challenges, we propose a Hand-informed Context Enhanced module (HiCE; pronounced `high-see’) that leverages spatiotemporal features from hand regions and their surrounding context through cross-attention mechanisms, learning to identify potential contact patterns. Our approach is further refined with a grasp-aware loss and soft label that emphasizes hand pose patterns and movement dynamics characteristic of touch events, enabling the model to distinguish between near-contact and actual contact frames. We also introduce TouchMoment, an egocentric dataset containing 4,021 videos and 8,456 annotated contact moments spanning over one million frames. Experiments on TouchMoment show that, under a strict evaluation criterion that counts a prediction as correct only if it falls within a two-frame tolerance of the ground-truth moment, our method achieves substantial gains and outperforms state-of-the-art event-spotting baselines by 16.91% average precision.

Method: ST-GD keeps pre-trained 2D visual-language models (like Grounding DINO) frozen and injects lightweight adapters (~10M parameters) to add spatio-temporal awareness. It includes a novel temporal decoder for boundary prediction while preserving pre-trained priors.

Result: ST-GD achieves competitive performance on limited-scale HC-STVG v1/v2 benchmarks and maintains robust generalization on the VidSTG dataset, demonstrating effectiveness in data-scarce scenarios.

Conclusion: ST-GD provides a powerful paradigm for complex video understanding under strict small-data constraints by efficiently adapting pre-trained 2D models to video tasks without destroying their valuable priors.

Abstract: Spatio-temporal video grounding (STVG) aims to localize queried objects within dynamic video segments. Prevailing fully-trained approaches are notoriously data-hungry. However, gathering large-scale STVG data is exceptionally challenging: dense frame-level bounding boxes and complex temporal language alignments are prohibitively expensive to annotate, especially for specialized video domains. Consequently, conventional models suffer from severe overfitting on these inherently limited datasets, while zero-shot foundational models lack the task-specific temporal awareness needed for precise localization. To resolve this small-data challenge, we introduce ST-GD, a data-efficient framework that adapts pre-trained 2D visual-language models (e.g., Grounding DINO) to video tasks. To avoid destroying pre-trained priors on small datasets, ST-GD keeps the base model frozen and strategically injects lightweight adapters (~10M trainable parameters) to instill spatio-temporal awareness, alongside a novel temporal decoder for boundary prediction. This design naturally counters data scarcity. Consequently, ST-GD excels in data-scarce scenarios, achieving highly competitive performance on the limited-scale HC-STVG v1/v2 benchmarks, while maintaining robust generalization on the VidSTG dataset. This validates ST-GD as a powerful paradigm for complex video understanding under strict small-data constraints.

Method: Proposes a tri-branch framework capturing global retinal context, structural features of optic disc/cup, and dynamically localized pathological regions. Uses Dynamic Window Mechanism to adaptively identify diagnostically informative regions and Knowledge-Enhanced Convolutional Attention Module incorporating retinal priors from a pre-trained foundation model.

Result: Achieves AUC of 98.5% and accuracy of 94.6% on AIROGS dataset, outperforming diverse baselines. Shows strong cross-domain generalization on SMDG-19 benchmark datasets.

Conclusion: Knowledge-guided attention combined with adaptive lesion localization significantly improves robustness of automated glaucoma screening systems, demonstrating the value of integrating domain-specific anatomical knowledge with dynamic feature learning.

Abstract: Automated diagnosis based on color fundus photography is essential for large-scale glaucoma screening. However, existing deep learning models are typically data-driven and lack explicit integration of retinal anatomical knowledge, which limits their robustness across heterogeneous clinical datasets. Moreover, pathological cues in fundus images may appear beyond predefined anatomical regions, making fixed-region feature extraction insufficient for reliable diagnosis. To address these challenges, we propose a retinal knowledge-oriented glaucoma screening framework that integrates dynamic multi-scale feature learning with domain-specific retinal priors. The framework adopts a tri-branch structure to capture complementary retinal representations, including global retinal context, structural features of the optic disc/cup, and dynamically localized pathological regions. A Dynamic Window Mechanism is devised to adaptively identify diagnostically informative regions, while a Knowledge-Enhanced Convolutional Attention Module incorporates retinal priors extracted from a pre-trained foundation model to guide attention learning. Extensive experiments on the large-scale AIROGS dataset demonstrate that the proposed method outperforms diverse baselines, achieving an AUC of 98.5% and an accuracy of 94.6%. Additional evaluations on multiple datasets from the SMDG-19 benchmark further confirm its strong cross-domain generalization capability, indicating that knowledge-guided attention combined with adaptive lesion localization can significantly improve the robustness of automated glaucoma screening systems.

Method: Proposes MAFL framework using pretrained multimodal image encoder as backbone, constructs real-fake feature learning network with adversarial bias-learning branch using multi-dimensional adversarial loss. Creates adversarial training between authenticity-discriminative feature learning and bias feature learning to suppress generation-pattern and content biases.

Result: Outperforms state-of-the-art methods by 10.89% in accuracy and 8.57% in Average Precision. Achieves over 80% detection accuracy even when trained with only 320 images on public datasets.

Conclusion: MAFL effectively captures fundamental differences between real and generated images by focusing on shared generative features across models, enhancing cross-model generalization while reducing reliance on large-scale training data.

Abstract: In recent years, the rapid development of generative artificial intelligence technology has significantly lowered the barrier to creating high-quality fake images, posing a serious challenge to information authenticity and credibility. Existing generated image detection methods typically enhance generalization through model architecture or network design. However, their generalization performance remains susceptible to data bias, as the training data may drive models to fit specific generative patterns and content rather than the common features shared by images from different generative models (asymmetric bias learning). To address this issue, we propose a Multi-dimensional Adversarial Feature Learning (MAFL) framework. The framework adopts a pretrained multimodal image encoder as the feature extraction backbone, constructs a real-fake feature learning network, and designs an adversarial bias-learning branch equipped with a multi-dimensional adversarial loss, forming an adversarial training mechanism between authenticity-discriminative feature learning and bias feature learning. By suppressing generation-pattern and content biases, MAFL guides the model to focus on the generative features shared across different generative models, thereby effectively capturing the fundamental differences between real and generated images, enhancing cross-model generalization, and substantially reducing the reliance on large-scale training data. Through extensive experimental validation, our method outperforms existing state-of-the-art approaches by 10.89% in accuracy and 8.57% in Average Precision (AP). Notably, even when trained with only 320 images, it can still achieve over 80% detection accuracy on public datasets.

Method: 1) Predict depth from single RGB image using Scale-Shift Residual Adapter for refinement; 2) Frequency-Aligned Fusion Module to hierarchically align RGB and depth features in frequency domain; 3) Mask-based Prediction Head for dynamic channel selection emphasizing key ingredient regions.

Result: Extensive experiments on multiple datasets demonstrate superiority over existing approaches. OmniFood8K provides 8,036 food samples with nutritional annotations and multi-view images, plus NutritionSynth-115K synthetic dataset.

Conclusion: The proposed framework enables accurate nutritional prediction from single RGB images without depth sensors, addressing limitations of existing methods and datasets for Chinese cuisine.

Abstract: Accurate estimation of food nutrition plays a vital role in promoting healthy dietary habits and personalized diet management. Most existing food datasets primarily focus on Western cuisines and lack sufficient coverage of Chinese dishes, which restricts accurate nutritional estimation for Chinese meals. Moreover, many state-of-the-art nutrition prediction methods rely on depth sensors, restricting their applicability in daily scenarios. To address these limitations, we introduce OmniFood8K, a comprehensive multimodal dataset comprising 8,036 food samples, each with detailed nutritional annotations and multi-view images. In addition, to enhance models’ capability in nutritional prediction, we construct NutritionSynth-115K, a large-scale synthetic dataset that introduces compositional variations while preserving precise nutritional labels. Moreover, we propose an end-to-end framework for nutritional prediction from a single RGB image. First, we predict a depth map from a single RGB image and design the Scale-Shift Residual Adapter (SSRA) to refine it for global scale consistency and local structural preservation. Second, we propose the Frequency-Aligned Fusion Module (FAFM) to hierarchically align and fuse RGB and depth features in the frequency domain. Finally, we design a Mask-based Prediction Head (MPH) to emphasize key ingredient regions via dynamic channel selection for more accurate prediction. Extensive experiments on multiple datasets demonstrate the superiority of our method over existing approaches. Project homepage: https://yudongjian.github.io/OmniFood8K-food/

Method: Proposes Decoding-stage Shift-aware Token Pruning (DSTP), a training-free add-on framework that aligns visual tokens with shifting reasoning requirements during the decoding stage to address RVIS.

Result: DSTP significantly mitigates performance degradation of pruning methods in complex reasoning tasks, yields performance gains across visual understanding benchmarks, works across diverse state-of-the-art architectures with minimal computational overhead.

Conclusion: DSTP effectively addresses the RVIS problem in visual token pruning, enabling better generalization to complex visual reasoning tasks while maintaining efficiency across various multimodal LLM architectures.

Abstract: Recently, visual token pruning has been studied to handle the vast number of visual tokens in Multimodal Large Language Models. However, we observe that while existing pruning methods perform reliably on simple visual understanding, they struggle to effectively generalize to complex visual reasoning tasks, a critical gap underexplored in previous studies. Through a systematic analysis, we identify Relevant Visual Information Shift (RVIS) during decoding as the primary failure driver. To address this, we propose Decoding-stage Shift-aware Token Pruning (DSTP), a training-free add-on framework that enables existing pruning methods to align visual tokens with shifting reasoning requirements during the decoding stage. Extensive experiments demonstrate that DSTP significantly mitigates performance degradation of pruning methods in complex reasoning tasks, while consistently yielding performance gains even across visual understanding benchmarks. Furthermore, DSTP demonstrates effectiveness across diverse state-of-the-art architectures, highlighting its generalizability and efficiency with minimal computational overhead.

Method: Evaluated 1,000 prompts from SALAD-Bench across four VLMs (GPT-4o, Claude Sonnet 4.5, Mistral-Large-3, Qwen3-VL-4B-Instruct) under varying font sizes (6-28px) and visual transformations (rotation, blur, noise, contrast changes), analyzing attack success rates and embedding distances from multimodal embedding models.

Result: Font size significantly affects ASR with small fonts (6px) yielding near-zero ASR; text attacks more effective than image attacks for GPT-4o and Claude; embedding distance shows strong negative correlation with ASR; heavy degradations increase embedding distance by 10-12% and reduce ASR by 34-96%; rotation affects models asymmetrically.

Conclusion: Model-specific robustness patterns preclude universal defenses, offering empirical guidance for selecting VLM backbones in adversarial environments, highlighting the need for tailored security approaches rather than one-size-fits-all solutions.

Abstract: We study typographic prompt injection attacks on vision-language models (VLMs), where adversarial text is rendered as images to bypass safety mechanisms, posing a growing threat as VLMs serve as the perceptual backbone of autonomous agents, from browser automation and computer-use systems to camera-equipped embodied agents. In practice, the attack surface is heterogeneous: adversarial text appears at varying font sizes and under diverse visual conditions, while the growing ecosystem of VLMs exhibits substantial variation in vulnerability, complicating defensive approaches. Evaluating 1,000 prompts from SALAD-Bench across four VLMs, namely, GPT-4o, Claude Sonnet 4.5, Mistral-Large-3, and Qwen3-VL-4B-Instruct under varying font sizes (6–28px) and visual transformations (rotation, blur, noise, contrast changes), we find: (1) font size significantly affects attack success rate (ASR), with very small fonts (6px) yielding near-zero ASR while mid-range fonts achieve peak effectiveness; (2) text attacks are more effective than image attacks for GPT-4o (36% vs 8%) and Claude (47% vs 22%), while Qwen3-VL and Mistral show comparable ASR across modalities; (3) text-image embedding distance from two multimodal embedding models (JinaCLIP and Qwen3-VL-Embedding) shows strong negative correlation with ASR across all four models (r = -0.71 to -0.93, p < 0.01); (4) heavy degradations increase embedding distance by 10–12% and reduce ASR by 34–96%, while rotation asymmetrically affects models (Mistral drops 50%, GPT-4o unchanged). These findings highlight that model-specific robustness patterns preclude one-size-fits-all defenses and offer empirical guidance for practitioners selecting VLM backbones for agentic systems operating in adversarial environments.

Method: Proposes a framework that generates modality-agnostic multi-modal prompts for Segment Anything Model (SAM). Uses interactions between data-driven content domain and knowledge-driven prompt domain to distill task-relevant cues into unified prompts. Includes a lightweight Mask Refine Module to calibrate coarse predictions using fine-grained prompt cues for better boundary accuracy.

Result: Extensive experiments on RGB-Depth, RGB-Thermal, and RGB-Polarization benchmarks validate the framework’s effectiveness and generalization capabilities, showing significant performance improvements on COD tasks.

Conclusion: The proposed modality-agnostic framework enables parameter-efficient adaptation to arbitrary auxiliary modalities and significantly improves camouflaged object detection performance across diverse multi-modal scenarios.

Abstract: Camouflaged Object Detection (COD) aims to segment objects that blend seamlessly into complex backgrounds, with growing interest in exploiting additional visual modalities to enhance robustness through complementary information. However, most existing approaches generally rely on modality-specific architectures or customized fusion strategies, which limit scalability and cross-modal generalization. To address this, we propose a novel framework that generates modality-agnostic multi-modal prompts for the Segment Anything Model (SAM), enabling parameter-efficient adaptation to arbitrary auxiliary modalities and significantly improving overall performance on COD tasks. Specifically, we model multi-modal learning through interactions between a data-driven content domain and a knowledge-driven prompt domain, distilling task-relevant cues into unified prompts for SAM decoding. We further introduce a lightweight Mask Refine Module to calibrate coarse predictions by incorporating fine-grained prompt cues, leading to more accurate camouflaged object boundaries. Extensive experiments on RGB-Depth, RGB-Thermal, and RGB-Polarization benchmarks validate the effectiveness and generalization of our modality-agnostic framework.

Method: Creates a pre-training sequence (model chain) arranged by ascending model size. Only the smallest model undergoes standard pre-training. Larger models are trained through sequential inverse knowledge transfer from smaller predecessors, reusing both parameter space and feature space knowledge.

Result: Achieves performance mostly superior to standard individual training while significantly reducing training costs. Validated across 45 datasets. When pre-training on CC3M: 72% computational complexity reduction for ViT-L; acceleration ratios improve from 4.13X to 5.68X to 7.09X as model family scales from 3 to 4 to 7 models.

Conclusion: CoM-PT provides performance-lossless training acceleration for vision foundation models that scales efficiently with model family expansion. The method is agnostic to pre-training paradigms and could extend to other computationally intensive scenarios like large language model pre-training.

Abstract: In this paper, we present Chain-of-Models Pre-Training (CoM-PT), a novel performance-lossless training acceleration method for vision foundation models (VFMs). This approach fundamentally differs from existing acceleration methods in its core motivation: rather than optimizing each model individually, CoM-PT is designed to accelerate the training pipeline at the model family level, scaling efficiently as the model family expands. Specifically, CoM-PT establishes a pre-training sequence for the model family, arranged in ascending order of model size, called model chain. In this chain, only the smallest model undergoes standard individual pre-training, while the other models are efficiently trained through sequential inverse knowledge transfer from their smaller predecessors by jointly reusing the knowledge in the parameter space and the feature space. As a result, CoM-PT enables all models to achieve performance that is mostly superior to standard individual training while significantly reducing training cost, and this is extensively validated across 45 datasets spanning zero-shot and fine-tuning tasks. Notably, its efficient scaling property yields a remarkable phenomenon: training more models even results in higher efficiency. For instance, when pre-training on CC3M: i) given ViT-L as the largest model, progressively prepending smaller models to the model chain reduces computational complexity by up to 72%; ii) within a fixed model size range, as the VFM family scales across 3, 4, and 7 models, the acceleration ratio of CoM-PT exhibits a striking leap: from 4.13X to 5.68X and 7.09X. Since CoM-PT is naturally agnostic to specific pre-training paradigms, we open-source the code to spur further extensions in more computationally intensive scenarios, such as large language model pre-training.

Method: Proposes a dual-modality anchor-guided framework with: 1) Text anchors from attribute-rich descriptions for fine-grained class semantics, 2) Adaptive image anchors capturing test-time statistics, 3) Anchor-based view filtering using alignment and confidence metrics, and 4) Confidence-weighted ensemble combining anchor predictions with original outputs for stable prompt updates.

Result: Extensive experiments on 15 benchmark datasets demonstrate state-of-the-art performance, showing significant improvements over existing test-time adaptation methods for vision-language models.

Conclusion: Anchor-guided supervision provides a robust foundation for prompt updates in vision-language models, effectively addressing miscalibration issues and improving adaptation performance under distribution shift.

Abstract: Test-Time Prompt Tuning (TPT) adapts vision-language models using augmented views, but its effectiveness is hindered by the challenge of determining which views are beneficial. Standard entropy-based filtering relies on the internal confidence scores of the model, which are often miscalibrated under distribution shift, assigning high confidence to irrelevant crops or background regions while ignoring semantic content. To address this, we propose a dual-modality anchor-guided framework that grounds view selection in semantic evidence. We introduce a text anchor from attribute-rich descriptions, to provide fine-grained class semantics, and an adaptive image anchor that captures evolving test-time statistics. Using these anchors, we filter views based on alignment and confidence, ensuring that only informative views guide adaptation. Moreover, we treat the anchors as auxiliary predictive heads and combine their predictions with the original output in a confidence-weighted ensemble, yielding a stable supervision signal for prompt updates. Extensive experiments on 15 benchmark datasets demonstrate new state-of-the-art performance, highlighting the contribution of anchor-guided supervision as a foundation for robust prompt updates.

Method: Extends base segmentor with aggregated deferral predictor and routing channels that dynamically route pixels to either AI or human experts. Uses pixel-wise surrogate collaboration loss for training deferral decisions and spatial-coherence loss for smooth deferral masks. Also extends to multi-expert setting with load-balancing penalty.

Result: DeferredSeg consistently outperforms baselines on three challenging medical datasets using MedSAM and CENet as base segmentors, demonstrating effectiveness for trustworthy dense medical segmentation.

Conclusion: The framework provides a model-agnostic approach for trustworthy medical segmentation through human-AI collaboration, with extensions to multi-expert settings and practical deployment considerations.

Abstract: Segmentation models based on deep neural networks demonstrate strong generalization for medical image segmentation. However, they often exhibit overconfidence or underconfidence, leading to unreliable confidence scores for segmentation masks, especially in ambiguous regions. This undermines the trustworthiness required for clinical deployment. Motivated by the learning-to-defer (L2D) paradigm, we introduce DeferredSeg, a deferral-aware segmentation framework, i.e., a Human–AI collaboration system that determines whether to defer predictions to human experts in specific regions. DeferredSeg extends the base segmentor with an aggregated deferral predictor and additional routing channels that dynamically route each pixel to either the base segmentor or a human expert. To train this routing efficiently, we introduce a pixel-wise surrogate collaboration loss that supervises deferral decisions. In addition, to preserve spatial coherence within deferral regions, we propose a spatial-coherence loss that enforces smooth deferral masks, thereby enhancing reliability. Beyond single-expert deferral, we further extend the framework to a multi-expert setting by introducing multiple discrepancy experts for collaborative decision-making. To prevent overloading or underutilizing individual experts, we further design a load-balancing penalty that evenly distributes workload across expert branches. We evaluate DeferredSeg on three challenging medical datasets using MedSAM and CENet as the base segmentor for fair comparison. Experimental results show that DeferredSeg consistently outperforms the baseline, demonstrating its effectiveness for trustworthy dense medical segmentation. Moreover, the proposed framework is model-agnostic and can be readily applied to other segmentation architectures.

Method: Proposes hybrid architecture: EfficientNetV2-M extracts local features, Vision Mamba (State Space Model) handles global context with linear complexity. Binary classification of mammography ROIs from CBIS-DDSM dataset into benign/malignant classes.

Result: Achieves strong lesion-level classification performance in ROI-based setting by combining CNN backbone with linear-complexity sequence model.

Conclusion: Hybrid CNN-Vision Mamba approach effectively addresses limitations of both CNNs and Transformers for mammography analysis, providing efficient global context modeling for breast lesion classification.

Abstract: Accurate characterization of suspicious breast lesions in mammography is important for early diagnosis and treatment planning. While Convolutional Neural Networks (CNNs) are effective at extracting local visual patterns, they are less suited to modeling long-range dependencies. Vision Transformers (ViTs) address this limitation through self-attention, but their quadratic computational cost can be prohibitive. This paper presents a hybrid architecture that combines EfficientNetV2-M for local feature extraction with Vision Mamba, a State Space Model (SSM), for efficient global context modeling. The proposed model performs binary classification of abnormality-centered mammography regions of interest (ROIs) from the CBIS-DDSM dataset into benign and malignant classes. By combining a strong CNN backbone with a linear-complexity sequence model, the approach achieves strong lesion-level classification performance in an ROI-based setting.

Method: Proposes IAD-Unify, a dual-encoder unified framework where a frozen DINOv2-based region expert supplies precise anomaly evidence to a shared Qwen3.5-4B vision-language backbone via lightweight token injection. This enables joint anomaly segmentation, region-grounded understanding, and mask-guided generation. Also constructs Anomaly-56K, a comprehensive unified multi-task evaluation platform with 59,916 images across 24 categories and 104 defect variants.

Result: Four key findings: (1) region grounding is crucial for understanding (removing it degrades location accuracy by >76 percentage points); (2) predicted-region performance closely matches oracle, confirming deployment viability; (3) region-grounded generation achieves best full-image fidelity and masked-region perceptual quality; (4) pre-initialized joint training improves understanding at negligible generation cost (-0.16 dB). Achieves strong performance on MMAD benchmark with robust cross-category generalization.

Conclusion: IAD-Unify successfully unifies three key industrial inspection capabilities in a single framework, demonstrating the importance of region grounding for both understanding and generation tasks. The framework shows strong generalization to unseen categories and provides a viable solution for real-world deployment.

Abstract: Real-world industrial inspection requires not only localizing defects, but also explaining them in natural language and generating controlled defect edits. However, existing approaches fail to jointly support all three capabilities within a unified framework and evaluation protocol. We propose IAD-Unify, a dual-encoder unified framework in which a frozen DINOv2-based region expert supplies precise anomaly evidence to a shared Qwen3.5-4B vision-language backbone via lightweight token injection, jointly enabling anomaly segmentation, region-grounded understanding, and mask-guided generation. To enable unified evaluation, we further construct Anomaly-56K, a comprehensive unified multi-task IAD evaluation platform, spanning 59,916 images across 24 categories and 104 defect variants. Controlled ablations yield four findings: (i) region grounding is the decisive mechanism for understanding, removing it degrades location accuracy by >76 pp; (ii) predicted-region performance closely matches oracle, confirming deployment viability; (iii) region-grounded generation achieves the best full-image fidelity and masked-region perceptual quality; and (iv) pre-initialized joint training improves understanding at negligible generation cost (-0.16 dB). IAD-Unify further achieves strong performance on the MMAD benchmark, including categories unseen during training, demonstrating robust cross-category generalization.

Method: Patch-level contrastive learning framework using MoCo-style objective with cross-category hard negative mining and generator-aware classification head. Learns forensic signals robust to latent decoding distortions by exploiting consistent generative fingerprints in inpainted regions from the same model.

Result: DiffusionPrint consistently improves localization across multiple generative models, with gains up to +28% on mask types unseen during fine-tuning. Shows generalization to unseen generative architectures when integrated into TruFor, MMFusion, and lightweight fusion baselines.

Conclusion: The framework successfully addresses the challenge of detecting diffusion-based forgeries by learning robust forensic signals that survive latent decoding distortions, providing a discriminative secondary modality for fusion-based image forgery localization.

Abstract: Modern diffusion-based inpainting models pose significant challenges for image forgery localization (IFL), as their full regeneration pipelines reconstruct the entire image via a latent decoder, disrupting the camera-level noise patterns that existing forensic methods rely on. We propose DiffusionPrint, a patch-level contrastive learning framework that learns a forensic signal robust to the spectral distortions introduced by latent decoding. It exploits the fact that inpainted regions generated by the same model share a consistent generative fingerprint, using this as a self-supervisory signal. DiffusionPrint trains a convolutional backbone via a MoCo-style objective with cross-category hard negative mining and a generator-aware classification head, producing a forensic feature map that serves as a highly discriminative secondary modality in fusion-based IFL frameworks. Integrated into TruFor, MMFusion, and a lightweight fusion baseline, DiffusionPrint consistently improves localization across multiple generative models, with gains of up to +28% on mask types unseen during fine-tuning and confirmed generalization to unseen generative architectures. Code is available at https://github.com/mever-team/diffusionprint

Method: Proposes Euler-inspired Decoupling Neural Operator (EDNO) that redefines pansharpening as continuous functional mapping in frequency domain. Uses Euler’s formula to transform features to polar coordinates, with Euler Feature Interaction Layer (EFIL) containing: 1) Explicit Feature Interaction Module for geometric alignment via linear weighting simulating phase rotation, and 2) Implicit Feature Interaction Module using feed-forward network for spectral distribution modeling.

Result: EDNO demonstrates superior efficiency-performance balance compared to heavyweight architectures on three datasets, offering better computational efficiency while maintaining high quality results.

Conclusion: EDNO provides an effective physics-inspired framework for pansharpening that addresses computational efficiency and quality issues of current methods through frequency domain processing and explicit-implicit feature decoupling.

Abstract: Pansharpening aims to synthesize high-resolution multispectral (HR-MS) images by fusing the spatial textures of panchromatic (PAN) images with the spectral information of low-resolution multispectral (LR-MS) images. While recent deep learning paradigms, especially diffusion-based operators, have pushed the performance boundaries, they often encounter spectral-spatial blurring and prohibitive computational costs due to their stochastic nature and iterative sampling. In this paper, we propose the Euler-inspired Decoupling Neural Operator (EDNO), a physics-inspired framework that redefines pansharpening as a continuous functional mapping in the frequency domain. Departing from conventional Cartesian feature processing, our EDNO leverages Euler’s formula to transform features into a polar coordinate system, enabling a novel explicit-implicit interaction mechanism. Specifically, we develop the Euler Feature Interaction Layer (EFIL), which decouples the fusion task into two specialized modules: 1) Explicit Feature Interaction Module, utilizing a linear weighting scheme to simulate phase rotation for adaptive geometric alignment; and 2) Implicit Feature Interaction Module, employing a feed-forward network to model spectral distributions for superior color consistency. By operating in the frequency domain, EDNO inherently captures global receptive fields while maintaining discretization-invariance. Experimental results on the three datasets demonstrate that EDNO offers a superior efficiency-performance balance compared to heavyweight architectures.

Method: Developed T2I-BiasBench with 13 complementary metrics (6 established + 7 new/adapted) to evaluate three open-source diffusion models against Gemini 2.5 Flash baseline, generating 1,574 images across five structured prompt categories.

Result: Key findings: (1) Stable Diffusion and BK-SDM show bias amplification in beauty prompts; (2) contextual constraints reduce professional-role gender bias; (3) all models collapse to narrow cultural representations, including RLHF-aligned Gemini.

Conclusion: T2I-BiasBench provides standardized, fine-grained bias evaluation for generative models, revealing that current alignment techniques don’t resolve cultural coverage gaps in text-to-image generation.

Abstract: Text-to-image (T2I) generative models achieve impressive visual fidelity but inherit and amplify demographic imbalances and cultural biases embedded in training data. We introduce T2I-BiasBench, a unified evaluation framework of thirteen complementary metrics that jointly captures demographic bias, element omission, and cultural collapse in diffusion models - the first framework to address all three dimensions simultaneously. We evaluate three open-source models - Stable Diffusion v1.5, BK-SDM Base, and Koala Lightning - against Gemini 2.5 Flash (RLHF-aligned) as a reference baseline. The benchmark comprises 1,574 generated images across five structured prompt categories. T2I-BiasBench integrates six established metrics with seven additional measures: four newly proposed (Composite Bias Score, Grounded Missing Rate, Implicit Element Missing Rate, Cultural Accuracy Ratio) and three adapted (Hallucination Score, Vendi Score, CLIP Proxy Score). Three key findings emerge: (1) Stable Diffusion v1.5 and BK-SDM exhibit bias amplification (>1.0) in beauty-related prompts; (2) contextual constraints such as surgical PPE substantially attenuate professional-role gender bias (Doctor CBS = 0.06 for SD v1.5); and (3) all models, including RLHF-aligned Gemini, collapse to a narrow set of cultural representations (CAS: 0.54-1.00), confirming that alignment techniques do not resolve cultural coverage gaps. T2I-BiasBench is publicly released to support standardized, fine-grained bias evaluation of generative models. The project page is available at: https://gyanendrachaubey.github.io/T2I-BiasBench/

Method: Proposes SEATrack with two key innovations: 1) AMG-LoRA for cross-modal alignment, combining Low-Rank Adaptation for domain adaptation with Adaptive Mutual Guidance to refine attention maps across modalities, and 2) Hierarchical Mixture of Experts (HMoE) for efficient global relation modeling instead of conventional local fusion.

Result: SEATrack achieves notable progress over state-of-the-art methods in balancing performance with efficiency across RGB-T, RGB-D, and RGB-E tracking tasks.

Conclusion: The paper demonstrates that prioritizing cross-modal alignment and using global fusion mechanisms can effectively break the performance-efficiency trade-off in multimodal tracking, advancing parameter-efficient fine-tuning in vision-based multimodal applications.

Abstract: Parameter-efficient fine-tuning (PEFT) in multimodal tracking reveals a concerning trend where recent performance gains are often achieved at the cost of inflated parameter budgets, which fundamentally erodes PEFT’s efficiency promise. In this work, we introduce SEATrack, a Simple, Efficient, and Adaptive two-stream multimodal tracker that tackles this performance-efficiency dilemma from two complementary perspectives. We first prioritize cross-modal alignment of matching responses, an underexplored yet pivotal factor that we argue is essential for breaking the trade-off. Specifically, we observe that modality-specific biases in existing two-stream methods generate conflicting matching attention maps, thereby hindering effective joint representation learning. To mitigate this, we propose AMG-LoRA, which seamlessly integrates Low-Rank Adaptation (LoRA) for domain adaptation with Adaptive Mutual Guidance (AMG) to dynamically refine and align attention maps across modalities. We then depart from conventional local fusion approaches by introducing a Hierarchical Mixture of Experts (HMoE) that enables efficient global relation modeling, effectively balancing expressiveness and computational efficiency in cross-modal fusion. Equipped with these innovations, SEATrack advances notable progress over state-of-the-art methods in balancing performance with efficiency across RGB-T, RGB-D, and RGB-E tracking tasks. \href{https://github.com/AutoLab-SAI-SJTU/SEATrack}{\textcolor{cyan}{Code is available}}.

Method: Proposes Variational Information Flow (VIF) framework using a probabilistic approach with Conditional Variational Autoencoder (CVAE) to model visual saliency relevant to question-answer pairs as a latent distribution. Designed as a plug-and-play module that can be integrated into existing MLLM architectures.

Result: Extensive evaluations across diverse benchmarks including General VQA, fine-grained perception, and visual grounding demonstrate competitive improvements over previous methods, validating VIF’s effectiveness in enhancing MLLMs’ fine-grained perception capabilities.

Conclusion: VIF successfully addresses the visual attenuation problem in MLLMs by modeling visual saliency probabilistically, leading to improved performance on fine-grained perception tasks without requiring architectural overhauls.

Abstract: While Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in general visual understanding, they frequently falter in fine-grained perception tasks that require identifying tiny objects or discerning subtle visual relationships. We attribute this limitation to Visual Attenuation: a phenomenon where sparse fine-grained visual signals are prematurely suppressed or diluted by dominant textual tokens during network propagation, resulting in a “loss of focus” during the deep-level decision-making process. Existing input-centric solutions fail to fundamentally reverse this intrinsic mechanism of information loss. To address this challenge, we propose the Variational Information Flow (VIF) framework. Adopting a probabilistic perspective, VIF leverages a Conditional Variational Autoencoder (CVAE) to model the visual saliency relevant to the question-answer pair as a latent distribution. As a plug-and-play module, VIF can be integrated into existing architectures. Extensive evaluations across diverse benchmarks, covering General VQA, fine-grained perception, and visual grounding, demonstrate that VIF yields competitive improvements over previous methods, validating its effectiveness in enhancing the fine-grained perception of MLLMs.

Method: Established a benchmark challenge exploring MLLMs’ ability to mimic human expert cognition in evaluating high-quality image pairs. Participants addressed two objectives: comparative quality selection (identifying superior image) and interpretative reasoning (generating expert-level explanations).

Result: The challenge attracted nearly 200 registrations and over 2,500 submissions. Top-performing methods significantly advanced state of the art in professional IQA.

Conclusion: MLLMs offer promising paradigm for professional image quality assessment, moving beyond scalar scores to provide comparative evaluation and reasoning capabilities that better mimic human expert judgment.

Abstract: In this paper, we present an overview of the NTIRE 2026 challenge on the 3rd Restore Any Image Model in the Wild, specifically focusing on Track 1: Professional Image Quality Assessment. Conventional Image Quality Assessment (IQA) typically relies on scalar scores. By compressing complex visual characteristics into a single number, these methods fundamentally struggle to distinguish subtle differences among uniformly high-quality images. Furthermore, they fail to articulate why one image is superior, lacking the reasoning capabilities required to provide guidance for vision tasks. To bridge this gap, recent advancements in Multimodal Large Language Models (MLLMs) offer a promising paradigm. Inspired by this potential, our challenge establishes a novel benchmark exploring the ability of MLLMs to mimic human expert cognition in evaluating high-quality image pairs. Participants were tasked with overcoming critical bottlenecks in professional scenarios, centering on two primary objectives: (1) Comparative Quality Selection: reliably identifying the visually superior image within a high-quality pair; and (2) Interpretative Reasoning: generating grounded, expert-level explanations that detail the rationale behind the selection. In total, the challenge attracted nearly 200 registrations and over 2,500 submissions. The top-performing methods significantly advanced the state of the art in professional IQA. The challenge dataset is available at https://github.com/narthchin/RAIM-PIQA, and the official homepage is accessible at https://www.codabench.org/competitions/12789/.

Method: Systematically analyzes two key questions: 1) whether diffusion pre-training benefits lightweight codecs (finds compression-oriented pre-training works better than generation-oriented), and 2) whether transformers are essential (finds lightweight convolutions suffice when paired with distillation). Builds one-step lightweight convolution diffusion codec with distillation and adversarial learning.

Result: Achieves real-time 60 FPS encoding and 42 FPS decoding at 1080p. Reduces bitrate by 85% at comparable FID to MS-ILLM, bridging gap between generative compression and practical real-time deployment.

Conclusion: Lightweight diffusion codecs can achieve real-time performance with compression-oriented pre-training and convolutional architectures, making generative compression practical for deployment.

Abstract: Recent advanced diffusion methods typically derive strong generative priors by scaling diffusion transformers. However, scaling fails to generalize when adapted for real-time compression scenarios that demand lightweight models. In this paper, we explore the design of real-time and lightweight diffusion codecs by addressing two pivotal questions. First, does diffusion pre-training benefit lightweight diffusion codecs? Through systematic analysis, we find that generation-oriented pre-training is less effective at small model scales whereas compression-oriented pre-training yields consistently better performance. Second, are transformers essential? We find that while global attention is crucial for standard generation, lightweight convolutions suffice for compression-oriented diffusion when paired with distillation. Guided by these findings, we establish a one-step lightweight convolution diffusion codec that achieves real-time $60$~FPS encoding and $42$~FPS decoding at 1080p. Further enhanced by distillation and adversarial learning, the proposed codec reduces bitrate by 85% at a comparable FID to MS-ILLM, bridging the gap between generative compression and practical real-time deployment. Codes are released at https://github.com/microsoft/GenCodec/CoD_Lite

Method: MODIX dynamically adapts positional strides based on modality-specific contributions by jointly modeling intra-modal density via covariance-based entropy and inter-modal interaction via cross-modal alignment to derive unified scores that rescale positional indices.

Result: Experiments across diverse architectures and benchmarks show MODIX consistently improves multimodal reasoning and adaptively reallocates attention according to task-dependent information distributions.

Conclusion: Positional encoding should be treated as an adaptive resource in Transformers for multimodal sequence modeling, with MODIX demonstrating that dynamic granularity allocation based on information density improves multimodal understanding.

Abstract: Vision-Language Models (VLMs) have achieved remarkable progress in multimodal understanding, yet their positional encoding mechanisms remain suboptimal. Existing approaches uniformly assign positional indices to all tokens, overlooking variations in information density within and across modalities, which leads to inefficient attention allocation where redundant visual regions dominate while informative content is underrepresented. We identify positional granularity as an implicit resource and propose MODIX (Multimodal Information-Driven Positional IndeX Scaling), a training-free framework that dynamically adapts positional strides based on modality-specific contributions. MODIX jointly models intra-modal density via covariance-based entropy and inter-modal interaction via cross-modal alignment to derive unified scores, which rescale positional indices to allocate finer granularity to informative modalities while compressing redundant ones, without requiring any modification to model parameters or architecture. Experiments across diverse architectures and benchmarks demonstrate that MODIX consistently improves multimodal reasoning and adaptively reallocates attention according to task-dependent information distributions, suggesting that positional encoding should be treated as an adaptive resource in Transformers for multimodal sequence modeling.

Method: Introduces a multiview transformer architecture that cross-attends across vision-language descriptors from multiple viewpoints and fuses them into unified per-3D-instance embeddings. Uses multiview consistency as self-supervision signal alongside standard supervised target-class loss.

Result: Consistently outperforms naive averaging or single-view descriptor selection, achieving state-of-the-art results on 3D semantic and instance classification benchmarks, including strong zero-shot performance on out-of-domain datasets.

Conclusion: CAMFusion effectively bridges 2D vision-language models to 3D understanding through cross-attentive multiview fusion and self-supervised multiview consistency, enabling better 3D scene understanding.

Abstract: Vision-language models have been key to the development of open-vocabulary 2D semantic segmentation. Lifting these models from 2D images to 3D scenes, however, remains a challenging problem. Existing approaches typically back-project and average 2D descriptors across views, or heuristically select a single representative one, often resulting in suboptimal 3D representations. In this work, we introduce a novel multiview transformer architecture that cross-attends across vision-language descriptors from multiple viewpoints and fuses them into a unified per-3D-instance embedding. As a second contribution, we leverage multiview consistency as a self-supervision signal for this fusion, which significantly improves performance when added to a standard supervised target-class loss. Our Cross-Attentive Multiview Fusion, which we denote with its acronym CAMFusion, not only consistently outperforms naive averaging or single-view descriptor selection, but also achieves state-of-the-art results on 3D semantic and instance classification benchmarks, including zero-shot evaluations on out-of-domain datasets.

Method: NS assembles multiple samples into a composite image, rescales it to original input size for model inference, computes natural selection scores based on predictions to characterize each sample’s competitive variation within the group, then uses these scores to dynamically reweight sample-wise losses.

Result: Extensive experiments on 12 public datasets across four image classification tasks demonstrate effectiveness. NS is compatible with diverse network architectures and doesn’t depend on task-specific assumptions, showing strong generality and practical potential.

Conclusion: NS provides a simple yet effective evolution-inspired optimization method that introduces explicit competition-driven mechanisms into training, enabling more adaptive and balanced model optimization beyond uniform sample treatment.

Abstract: Conventional deep network training generally optimizes all samples under a largely uniform learning paradigm, without explicitly modeling the heterogeneous competition among them. Such an oversimplified treatment can lead to several well-known issues, including bias under class imbalance, insufficient learning of hard samples, and the erroneous reinforcement of noisy samples. In this work, we present \textit{Natural Selection} (NS), a novel evolution-inspired optimization method that explicitly incorporates competitive interactions into deep network training. Unlike conventional sample reweighting strategies that rely mainly on predefined heuristics or static criteria, NS estimates the competitive status of each sample in a group-wise context and uses it to adaptively regulate its training contribution. Specifically, NS first assembles multiple samples into a composite image and rescales it to the original input size for model inference. Based on the resulting predictions, a natural selection score is computed for each sample to characterize its relative competitive variation within the constructed group. These scores are then used to dynamically reweight the sample-wise loss, thereby introducing an explicit competition-driven mechanism into the optimization process. In this way, NS provides a simple yet effective means of moving beyond uniform sample treatment and enables more adaptive and balanced model optimization. Extensive experiments on 12 public datasets across four image classification tasks demonstrate the effectiveness of the proposed method. Moreover, NS is compatible with diverse network architectures and does not depend on task-specific assumptions, indicating its strong generality and practical potential. The code will be made publicly available.

Method: Two-stage knowledge distillation: 1) Teacher model (BiomedCLIP-based) learns PET-MRI alignment via cross-modal attention and triplet contrastive learning with PET-informed negative sampling. 2) MRI-only student mimics teacher via feature-level and logit-level distillation.

Result: Achieved best AUC of 0.74 on OASIS-3 and 0.68 on ADNI across four MRI contrasts (T1w, T2w, FLAIR, T2*). Saliency analysis shows predictions focus on anatomically relevant cortical regions.

Conclusion: PET-guided knowledge distillation enables effective PET-free amyloid-β screening from MRI alone, maintaining interpretability and eliminating need for clinical variables or PET at inference.

Abstract: Detecting amyloid-$β$ (A$β$) positivity is crucial for early diagnosis of Alzheimer’s disease but typically requires PET imaging, which is costly, invasive, and not widely accessible, limiting its use for population-level screening. We address this gap by proposing a PET-guided knowledge distillation framework that enables A$β$ prediction from MRI alone, without requiring non-imaging clinical covariates or PET at inference. Our approach employs a BiomedCLIP-based teacher model that learns PET-MRI alignment via cross-modal attention and triplet contrastive learning with PET-informed (Centiloid-aware) online negative sampling. An MRI-only student then mimics the teacher via feature-level and logit-level distillation. Evaluated across four MRI contrasts (T1w, T2w, FLAIR, T2*) and two independent datasets, our approach demonstrates effective knowledge transfer (best AUC: 0.74 on OASIS-3, 0.68 on ADNI) while maintaining interpretability and eliminating the need for clinical variables. Saliency analysis confirms that predictions focus on anatomically relevant cortical regions, supporting the clinical viability of PET-free A$β$ screening. Code is available at https://github.com/FrancescoChiumento/pet-guided-mri-amyloid-detection.

Method: Uses adaptive receptive field module to maintain global/local distributions, incorporates 3D positional encoding as spatial prior for controlling positions/scale/local details, and proposes LLM-based evaluation criterion.

Result: Outperforms existing methods in structural consistency, visual quality, and spatial controllability; demonstrates broad applicability across text-guided generation, editing, outpainting, and paint-to-image tasks.

Conclusion: StructDiff effectively addresses structural preservation and spatial control challenges in single-image generation while introducing novel evaluation methods and showing strong performance across various applications.

Abstract: This paper introduces StructDiff, a generative framework based on a single-scale diffusion model for single-image generation. Single-image generation aims to synthesize diverse samples with similar visual content to the source image by capturing its internal statistics, without relying on external data. However, existing methods often struggle to preserve the structural layout, especially for images with large rigid objects or strict spatial constraints. Moreover, most approaches lack spatial controllability, making it difficult to guide the structure or placement of generated content. To address these challenges, StructDiff introduces an \textit{adaptive receptive field} module to maintain both global and local distributions. Building on this foundation, StructDiff incorporates 3D positional encoding (PE) as a spatial prior, allowing flexible control over positions, scale, and local details of generated objects. To our knowledge, this spatial control capability represents the first exploration of PE-based manipulation in single-image generation. Furthermore, we propose a novel evaluation criterion for single-image generation based on large language models (LLMs). This criterion specifically addresses the limitations of existing objective metrics and the high labor costs associated with user studies. StructDiff also demonstrates broad applicability across downstream tasks, such as text-guided image generation, image editing, outpainting, and paint-to-image synthesis. Extensive experiments demonstrate that StructDiff outperforms existing methods in structural consistency, visual quality, and spatial controllability. The project page is available at https://butter-crab.github.io/StructDiff/.

Method: Progressive Distractor Filtering (PDF-GS) framework with multi-phase optimization: progressive filtering phases gradually remove distractors using discrepancy cues, followed by reconstruction phases that restore fine-grained, view-consistent details from purified Gaussian representations.

Result: Achieves robust, high-fidelity, distractor-free reconstructions, consistently outperforming baselines across diverse datasets and challenging real-world conditions. Lightweight and easily adaptable to existing 3DGS frameworks with no architectural changes or additional inference overhead.

Conclusion: PDF-GS demonstrates that amplifying 3DGS’s inherent self-filtering property through progressive optimization enables state-of-the-art performance for robust 3D reconstruction in the presence of distractors.

Abstract: Recent advances in 3D Gaussian Splatting (3DGS) have enabled impressive real-time photorealistic rendering. However, conventional training pipelines inherently assume full multi-view consistency among input images, which makes them sensitive to distractors that violate this assumption and cause visual artifacts. In this work, we revisit an underexplored aspect of 3DGS: its inherent ability to suppress inconsistent signals. Building on this insight, we propose PDF-GS (Progressive Distractor Filtering for Robust 3D Gaussian Splatting), a framework that amplifies this self-filtering property through a progressive multi-phase optimization. The progressive filtering phases gradually remove distractors by exploiting discrepancy cues, while the following reconstruction phase restores fine-grained, view-consistent details from the purified Gaussian representation. Through this iterative refinement, PDF-GS achieves robust, high-fidelity, and distractor-free reconstructions, consistently outperforming baselines across diverse datasets and challenging real-world conditions. Moreover, our approach is lightweight and easily adaptable to existing 3DGS frameworks, requiring no architectural changes or additional inference overhead, leading to a new state-of-the-art performance. The code is publicly available at https://github.com/kangrnin/PDF-GS.

Method: Decoder-side Temporal Rebalancing (DTR) - a training-free, layer-selective inference method that adaptively calibrates decoder-side visual attention in middle-to-late decoder layers. It rebalances temporal evidence allocation without altering visual encoding or requiring auxiliary models.

Result: DTR consistently improves hallucination robustness across diverse Video-LLM families while preserving competitive video understanding performance and maintaining high inference efficiency.

Conclusion: Addressing decoder-side temporal imbalance through attention rebalancing is an effective training-free approach to reduce hallucinations in Video-LLMs while maintaining understanding capabilities.

Abstract: Recent Video Large Language Models (Video-LLMs) have demonstrated strong capability in video understanding, yet they still suffer from hallucinations. Existing mitigation methods typically rely on training, input modification, auxiliary guidance, or additional decoding procedures, while largely overlooking a more fundamental challenge. During generation, Video-LLMs tend to over-rely on a limited portion of temporal evidence, leading to temporally imbalanced evidence aggregation across the video. To address this issue, we investigate a decoder-side phenomenon in which the model exhibits a temporally imbalanced concentration pattern. We term the frame with the highest aggregated frame-level attention mass the anchor frame. We find that this bias is largely independent of the input video and instead appears to reflect a persistent, model-specific structural or positional bias, whose over-dominance is closely associated with hallucination-prone generation. Motivated by this insight, we propose Decoder-side Temporal Rebalancing (DTR), a training-free, layer-selective inference method that rebalances temporal evidence allocation in middle-to-late decoder layers without altering visual encoding or requiring auxiliary models. DTR adaptively calibrates decoder-side visual attention to alleviate temporally imbalanced concentration and encourage under-attended frames to contribute more effectively to response generation. In this way, DTR guides the decoder to ground its outputs in temporally broader and more balanced video evidence. Extensive experiments on hallucination and video understanding benchmarks show that DTR consistently improves hallucination robustness across diverse Video-LLM families, while preserving competitive video understanding performance and high inference efficiency.

Method: Reformulates VITON as a linear inverse problem using trajectory-aligned solvers with residual prior-based initialization and artifact-free measurement-guided sampling combining data consistency, frequency-level correction, and periodic standard denoising.

Result: Demonstrates effective preservation of identity and background, elimination of boundary artifacts, and improved visual fidelity and robustness on VITON-HD, DressCode, and SHHQ-1.0 datasets.

Conclusion: ART-VITON provides a robust solution for virtual try-on that maintains measurement consistency while achieving artifact-free synthesis, outperforming state-of-the-art baselines.

Abstract: Virtual try-on (VITON) aims to generate realistic images of a person wearing a target garment, requiring precise garment alignment in try-on regions and faithful preservation of identity and background in non-try-on regions. While latent diffusion models (LDMs) have advanced alignment and detail synthesis, preserving non-try-on regions remains challenging. A common post-hoc strategy directly replaces these regions with original content, but abrupt transitions often produce boundary artifacts. To overcome this, we reformulate VITON as a linear inverse problem and adopt trajectory-aligned solvers that progressively enforce measurement consistency, reducing abrupt changes in non-try-on regions. However, existing solvers still suffer from semantic drift during generation, leading to artifacts. We propose ART-VITON, a measurement-guided diffusion framework that ensures measurement adherence while maintaining artifact-free synthesis. Our method integrates residual prior-based initialization to mitigate training-inference mismatch and artifact-free measurement-guided sampling that combines data consistency, frequency-level correction, and periodic standard denoising. Experiments on VITON-HD, DressCode, and SHHQ-1.0 demonstrate that ART-VITON effectively preserves identity and background, eliminates boundary artifacts, and consistently improves visual fidelity and robustness over state-of-the-art baselines.

Method: Proposes Extreme Low-light Optimized Gaussian Splatting (ELoG-GS) that integrates learning-based point cloud initialization and luminance-guided color enhancement. Uses geometry-aware initialization and photometric adaptation strategies for stable Gaussian Splatting in challenging conditions.

Result: Achieved PSNR of 18.6626 and SSIM of 0.6855 on NTIRE Track 1 benchmark, significantly improving reconstruction quality over baselines with superior visual fidelity and geometric consistency.

Conclusion: ELoG-GS provides a practical solution for robust 3D reconstruction in real-world degraded scenarios, demonstrating effectiveness in extreme low-light conditions through integrated geometry and photometric adaptation strategies.

Abstract: This paper presents our approach to the NTIRE 2026 3D Restoration and Reconstruction Challenge (Track 1), which focuses on reconstructing high-quality 3D representations from degraded multi-view inputs. The challenge involves recovering geometrically consistent and photorealistic 3D scenes in extreme low-light environments. To address this task, we propose Extreme Low-light Optimized Gaussian Splatting (ELoG-GS), a robust low-light 3D reconstruction pipeline that integrates learning-based point cloud initialization and luminance-guided color enhancement for stable and photorealistic Gaussian Splatting. Our method incorporates both geometry-aware initialization and photometric adaptation strategies to improve reconstruction fidelity under challenging conditions. Extensive experiments on the NTIRE Track 1 benchmark demonstrate that our approach significantly improves reconstruction quality over the baselines, achieving superior visual fidelity and geometric consistency. The proposed method provides a practical solution for robust 3D reconstruction in real-world degraded scenarios. In the final testing phase, our method achieved a PSNR of 18.6626 and an SSIM of 0.6855 on the official platform leaderboard. Code is available at https://github.com/lyh120/FSGS_EAPGS.

Method: Proposes a denoising framework integrating noise prior reduction and spatial-spectral adaptive fidelity term, with an adaptive weight tensor to dynamically balance fidelity and prior regularization. Uses a fast pixel-wise model with representative coefficient total variation regularizer for mixed noise removal.

Result: Extensive experiments on simulated and real-world datasets show superior denoising performance while maintaining competitive computational efficiency compared to existing methods.

Conclusion: The proposed framework effectively handles various noise types while accurately capturing spectral low-rank structure and local smoothness of hyperspectral images through adaptive balancing of fidelity and prior terms.

Abstract: The core challenge of hyperspectral image denoising is striking the right balance between data fidelity and noise prior modeling. Most existing methods place too much emphasis on the intrinsic priors of the image while overlooking diverse noise assumptions and the dynamic trade-off between fidelity and priors. To address these issues, we propose a denoising framework that integrates noise prior reduction and a spatial-spectral adaptive fidelity term. This framework considers comprehensive noise priors with fewer parameters and introduces an adaptive weight tensor to dynamically balance the fidelity and prior regularization terms. Within this framework, we further develop a fast and robust pixel-wise model combined with the representative coefficient total variation regularizer to accurately remove mixed noise in HSIs. The proposed method not only efficiently handles various types of noise but also accurately captures the spectral low-rank structure and local smoothness of HSIs. An efficient optimization algorithm based on the alternating direction method of multipliers is designed to ensure stable and fast convergence. Extensive experiments on simulated and real-world datasets demonstrate that the proposed model achieves superior denoising performance while maintaining competitive computational efficiency.

Method: MMSD: Replaces original image with compact semantic representations (segmentation maps, edge maps, textual descriptions) and reconstructs scene at receiver using diffusion-based generative model. SAMR: Selectively suppresses non-critical image regions based on semantic importance before JPEG encoding, then restores missing content through generative inpainting. Both use asymmetric sender-receiver architecture with lightweight edge processing and server-side reconstruction.

Result: Achieves 99% (MMSD) and 99.1% (SAMR) average transmitted-data reduction. MMSD has lower payload than SPIC baseline with strong semantic consistency. SAMR provides better quality-compression trade-off than standard JPEG and SQ-GAN. Edge processing time: 15s for MMSD, 9s for SAMR on Raspberry Pi 5.

Conclusion: Semantic image communication pipelines enable efficient visual monitoring under communication constraints by transmitting semantic representations instead of raw pixels, with different approaches balancing confidentiality/compression versus visual quality.

Abstract: Many visual monitoring systems operate under strict communication constraints, where transmitting full-resolution images is impractical and often unnecessary. In such settings, visual data is often used for object presence, spatial relationships, and scene context rather than exact pixel fidelity. This paper presents two semantic image communication pipelines for traffic monitoring, MMSD and SAMR, that reduce transmission cost while preserving meaningful visual information. MMSD (Multi-Modal Semantic Decomposition) targets very high compression together with data confidentiality, since sensitive pixel content is not transmitted. It replaces the original image with compact semantic representations, namely segmentation maps, edge maps, and textual descriptions, and reconstructs the scene at the receiver using a diffusion-based generative model. SAMR (Semantic-Aware Masking Reconstruction) targets higher visual quality while maintaining strong compression. It selectively suppresses non-critical image regions according to semantic importance before standard JPEG encoding and restores the missing content at the receiver through generative inpainting. Both designs follow an asymmetric sender-receiver architecture, where lightweight processing is performed at the edge and computationally intensive reconstruction is offloaded to the server. On a Raspberry Pi~5, the edge-side processing time is about 15s for MMSD and 9s for SAMR. Experimental results show average transmitted-data reductions of 99% for MMSD and 99.1% for SAMR. In addition, MMSD achieves lower payload size than the recent SPIC baseline while preserving strong semantic consistency, whereas SAMR provides a better quality-compression trade-off than standard JPEG and SQ-GAN under comparable operating conditions.

Method: Proposes GeoAlign framework that dynamically aggregates multi-layer geometric features. Constructs a hierarchical geometric feature bank and uses MLLM’s original visual tokens as content-aware queries to perform layer-wise sparse routing, adaptively fetching suitable geometric features for each patch.

Result: Extensive experiments on VSI-Bench, ScanQA, and SQA3D show that the compact 4B model achieves state-of-the-art performance, even outperforming larger existing MLLMs.

Conclusion: Dynamic multi-layer geometric feature aggregation effectively addresses task misalignment bias and improves spatial reasoning in MLLMs, enabling compact models to outperform larger ones on spatial reasoning benchmarks.

Abstract: Multimodal large language models (MLLMs) have exhibited remarkable performance in various visual tasks, yet still struggle with spatial reasoning. Recent efforts mitigate this by injecting geometric features from 3D foundation models, but rely on static single-layer extractions. We identify that such an approach induces a task misalignment bias: the geometric features naturally evolve towards 3D pretraining objectives, which may contradict the heterogeneous spatial demands of MLLMs, rendering any single layer fundamentally insufficient. To resolve this, we propose GeoAlign, a novel framework that dynamically aggregates multi-layer geometric features to realign with the actual demands. GeoAlign constructs a hierarchical geometric feature bank and leverages the MLLM’s original visual tokens as content-aware queries to perform layer-wise sparse routing, adaptively fetching the suitable geometric features for each patch. Extensive experiments on VSI-Bench, ScanQA, and SQA3D demonstrate that our compact 4B model effectively achieves state-of-the-art performance, even outperforming larger existing MLLMs.

Method: PromptEcho computes token-level cross-entropy loss of a frozen VLM using the original prompt as label, directly extracting image-text alignment knowledge encoded during VLM pretraining. It’s deterministic, computationally efficient, and improves automatically as stronger open-source VLMs become available.

Result: PromptEcho achieves substantial improvements on DenseAlignBench (+26.8pp/+16.2pp net win rate) on two state-of-the-art T2I models (Z-Image and QwenImage-2512), with consistent gains on GenEval, DPG-Bench, and TIIFBench without task-specific training. Reward quality scales with VLM size.

Conclusion: PromptEcho provides an effective, annotation-free reward construction method for RL-based text-to-image models that leverages frozen VLMs’ pretrained knowledge, outperforming existing methods and scaling with VLM capabilities.

Abstract: Reinforcement learning (RL) can improve the prompt following capability of text-to-image (T2I) models, yet obtaining high-quality reward signals remains challenging: CLIP Score is too coarse-grained, while VLM-based reward models (e.g., RewardDance) require costly human-annotated preference data and additional fine-tuning. We propose PromptEcho, a reward construction method that requires \emph{no} annotation and \emph{no} reward model training. Given a generated image and a guiding query, PromptEcho computes the token-level cross-entropy loss of a frozen VLM with the original prompt as the label, directly extracting the image-text alignment knowledge encoded during VLM pretraining. The reward is deterministic, computationally efficient, and improves automatically as stronger open-source VLMs become available. For evaluation, we develop DenseAlignBench, a benchmark of concept-rich dense captions for rigorously testing prompt following capability. Experimental results on two state-of-the-art T2I models (Z-Image and QwenImage-2512) demonstrate that PromptEcho achieves substantial improvements on DenseAlignBench (+26.8pp / +16.2pp net win rate), along with consistent gains on GenEval, DPG-Bench, and TIIFBench without any task-specific training. Ablation studies confirm that PromptEcho comprehensively outperforms inference-based scoring with the same VLM, and that reward quality scales with VLM size. We will open-source the trained models and the DenseAlignBench.

Method: Proposes HyperSSM, integrating Hypergraph computation and State Space Model (SSM) for unified spatial-temporal reasoning. Hypergraph module captures spatial motion correlations through dynamic hyperedges, while SSM enforces temporal smoothness via structured state transitions. Objects with similar motion states mutually constrain and refine each other.

Result: Achieves state-of-the-art performance on four mainstream benchmarks (MOT17, MOT20, DanceTrack, SportsMOT) covering various motion patterns and scene complexities. Demonstrates robust and stable motion estimation across diverse tracking scenarios.

Conclusion: The collaborative reasoning framework effectively addresses limitations of existing motion estimation approaches by enabling joint inference among correlated objects, stabilizing noisy trajectories, and inferring plausible motion continuity during occlusion.

Abstract: Motion reasoning serves as the cornerstone of multi-object tracking (MOT), as it enables consistent association of targets across frames. However, existing motion estimation approaches face two major limitations: (1) instability caused by noisy or probabilistic predictions, and (2) vulnerability under occlusion, where trajectories often fragment once visual cues disappear. To overcome these issues, we propose a collaborative reasoning framework that enhances motion estimation through joint inference among multiple correlated objects. By allowing objects with similar motion states to mutually constrain and refine each other, our framework stabilizes noisy trajectories and infers plausible motion continuity even when target is occluded. To realize this concept, we design HyperSSM, an architecture that integrates Hypergraph computation and a State Space Model (SSM) for unified spatial-temporal reasoning. The Hypergraph module captures spatial motion correlations through dynamic hyperedges, while the SSM enforces temporal smoothness via structured state transitions. This synergistic design enables simultaneous optimization of spatial consensus and temporal coherence, resulting in robust and stable motion estimation. Extensive experiments on four mainstream and diverse benchmarks(MOT17, MOT20, DanceTrack, and SportsMOT) covering various motion patterns and scene complexities, demonstrate that our approach achieves state-of-the-art performance across a wide range of tracking scenarios.

Method: Developed SceneCritic, a symbolic evaluator based on SceneOnto - a structured spatial ontology constructed by aggregating indoor scene priors from 3D-FRONT, ScanNet, and Visual Genome. SceneOnto traverses the ontology to verify semantic, orientation, and geometric coherence across object relationships. Paired with iterative refinement test bed using three critic modalities: rule-based (collision constraints), LLM (text layout), and VLM (rendered observations).

Result: SceneCritic aligns substantially better with human judgments than VLM-based evaluators; text-only LLMs outperform VLMs on semantic layout quality; image-based VLM refinement is most effective for semantic and orientation correction.

Conclusion: Symbolic evaluation via SceneCritic provides more stable and objective assessment of indoor scene layouts than LLM/VLM judges, with different critic modalities offering complementary strengths for iterative refinement of spatial structure generation.

Abstract: Large Language Models (LLMs) and Vision-Language Models (VLMs) increasingly generate indoor scenes through intermediate structures such as layouts and scene graphs, yet evaluation still relies on LLM or VLM judges that score rendered views, making judgments sensitive to viewpoint, prompt phrasing, and hallucination. When the evaluator is unstable, it becomes difficult to determine whether a model has produced a spatially plausible scene or whether the output score reflects the choice of viewpoint, rendering, or prompt. We introduce SceneCritic, a symbolic evaluator for floor-plan-level layouts. SceneCritic’s constraints are grounded in SceneOnto, a structured spatial ontology we construct by aggregating indoor scene priors from 3D-FRONT, ScanNet, and Visual Genome. SceneOnto traverses this ontology to jointly verify semantic, orientation, and geometric coherence across object relationships, providing object-level and relationship-level assessments that identify specific violations and successful placements. Furthermore, we pair SceneCritic with an iterative refinement test bed that probes how models build and revise spatial structure under different critic modalities: a rule-based critic using collision constraints as feedback, an LLM critic operating on the layout as text, and a VLM critic operating on rendered observations. Through extensive experiments, we show that (a) SceneCritic aligns substantially better with human judgments than VLM-based evaluators, (b) text-only LLMs can outperform VLMs on semantic layout quality, and (c) image-based VLM refinement is the most effective critic modality for semantic and orientation correction.

Method: Proposes an OFA compression framework that yields different subnetworks in one-shot training. Uses channel importance-based allocation to construct subnetworks of specific sizes and a reweighting strategy to balance optimization across subnetworks.

Result: The approach produces compressed DPMs for various sizes with significantly lower training overhead while maintaining satisfactory performance compared to repeated individual compression.

Conclusion: The proposed OFA framework efficiently addresses the deployment challenge of DPMs across diverse devices by enabling one-shot compression that yields multiple subnetworks with different computational requirements.

Abstract: The Diffusion Probabilistic Model (DPM) achieves remarkable performance in image generation, while its increasing parameter size and computational overhead hinder its deployment in practical applications. To improve this, the existing literature focuses on obtaining a smaller model with a fixed architecture through model compression. However, in practice, DPMs usually need to be deployed on various devices with different resource constraints, which leads to multiple compression processes, incurring significant overhead for repeated training. To obviate this, we propose a once-for-all (OFA) compression framework for DPMs that yields different subnetworks with various computations in a one-shot training manner. The existing OFA framework typically involves massive subnetworks with different parameter sizes, while such a huge candidate space slows the optimization. Thus, we propose to restrict the candidate subnetworks with a certain set of parameter sizes, where each size corresponds to a specific subnetwork. Specifically, to construct each subnetwork with a given size, we gradually allocate the maintained channels by their importance. Furthermore, we propose a reweighting strategy to balance the optimization process of different subnetworks. Experimental results show that our approach can produce compressed DPMs for various sizes with significantly lower training overhead while achieving satisfactory performance.

Method: Brain-DiT uses metadata-conditioned diffusion pretraining with a Diffusion Transformer (DiT) on 349,898 fMRI sessions from 24 datasets spanning multiple brain states, enabling learning of multi-scale representations capturing both fine-grained functional structure and global semantics.

Result: Extensive evaluations on 7 downstream tasks show diffusion-based generative pretraining outperforms reconstruction or alignment methods, with metadata conditioning improving performance by disentangling neural dynamics from population variability. Different tasks prefer different representational scales.

Conclusion: Diffusion-based generative pretraining with metadata conditioning is an effective approach for building universal fMRI foundation models that can learn generalized representations across diverse brain states and tasks.

Abstract: Current fMRI foundation models primarily rely on a limited range of brain states and mismatched pretraining tasks, restricting their ability to learn generalized representations across diverse brain states. We present \textit{Brain-DiT}, a universal multi-state fMRI foundation model pretrained on 349,898 sessions from 24 datasets spanning resting, task, naturalistic, disease, and sleep states. Unlike prior fMRI foundation models that rely on masked reconstruction in the raw-signal space or a latent space, \textit{Brain-DiT} adopts metadata-conditioned diffusion pretraining with a Diffusion Transformer (DiT), enabling the model to learn multi-scale representations that capture both fine-grained functional structure and global semantics. Across extensive evaluations and ablations on 7 downstream tasks, we find consistent evidence that diffusion-based generative pretraining is a stronger proxy than reconstruction or alignment, with metadata-conditioned pretraining further improving downstream performance by disentangling intrinsic neural dynamics from population-level variability. We also observe that downstream tasks exhibit distinct preferences for representational scale: ADNI classification benefits more from global semantic representations, whereas age/sex prediction comparatively relies more on fine-grained local structure. Code and parameters of Brain-DiT are available at \href{https://github.com/REDMAO4869/Brain-DiT}{Link}.

Method: Proposes Risk-Calibrated Learning that embeds a confusion-aware clinical severity matrix M into the optimization landscape to explicitly distinguish between visual ambiguity (fine-grained errors) and catastrophic structural errors. The method suppresses critical errors like false negatives without requiring complex architectural changes.

Result: Validated across four imaging modalities: Brain Tumor MRI, ISIC 2018 (Dermoscopy), BreaKHis (Breast Histopathology), and SICAPv2 (Prostate Histopathology). Achieved relative safety improvements ranging from 20.0% (breast histopathology) to 92.4% (prostate histopathology) compared to state-of-the-art baselines like Focal Loss, consistently reducing Critical Error Rate (CER).

Conclusion: Risk-Calibrated Learning offers a superior safety-accuracy trade-off across both CNN and Transformer architectures by explicitly addressing semantic incoherence in medical image classification, making deep learning models more clinically trustworthy.

Abstract: Deep learning models often achieve expert-level accuracy in medical image classification but suffer from a critical flaw: semantic incoherence. These high-confidence mistakes that are semantically incoherent (e.g., classifying a malignant tumor as benign) fundamentally differ from acceptable errors which stem from visual ambiguity. Unlike safe, fine-grained disagreements, these fatal failures erode clinical trust. To address this, we propose Risk-Calibrated Learning, a technique that explicitly distinguishes between visual ambiguity (fine-grained errors) and catastrophic structural errors. By embedding a confusion-aware clinical severity matrix M into the optimization landscape, our method suppresses critical errors (false negatives) without requiring complex architectural changes. We validate our approach in four different imaging modalities: Brain Tumor MRI, ISIC 2018 (Dermoscopy), BreaKHis (Breast Histopathology), and SICAPv2 (Prostate Histopathology). Extensive experiments demonstrate that our Risk-Calibrated Loss consistently reduces the Critical Error Rate (CER) for all four datasets, achieving relative safety improvements ranging from 20.0% (on breast histopathology) to 92.4% (on prostate histopathology) compared to state-of-the-art baselines such as Focal Loss. These results confirm that our method offers a superior safety-accuracy trade-off across both CNN and Transformer architectures.

Method: Proposes AffectAgent with three specialized agents: query planner (retrieves cross-modal samples), evidence filter (assesses evidence), and emotion generator (generates predictions). Uses Multi-Agent Proximal Policy Optimization (MAPPO) with shared affective reward for end-to-end optimization. Introduces Modality-Balancing Mixture of Experts (MB-MoE) to regulate modality contributions and Retrieval-Augmented Adaptive Fusion (RAAF) for semantic completion under missing-modality conditions.

Result: Extensive experiments on MER-UniBench demonstrate superior performance across complex scenarios compared to existing methods.

Conclusion: AffectAgent effectively addresses modal ambiguity and affective dependency challenges in multimodal emotion recognition through collaborative multi-agent decision-making and advanced fusion techniques.

Abstract: LLM-based multimodal emotion recognition relies on static parametric memory and often hallucinates when interpreting nuanced affective states. In this paper, given that single-round retrieval-augmented generation is highly susceptible to modal ambiguity and therefore struggles to capture complex affective dependencies across modalities, we introduce AffectAgent, an affect-oriented multi-agent retrieval-augmented generation framework that leverages collaborative decision-making among agents for fine-grained affective understanding. Specifically, AffectAgent comprises three jointly optimized specialized agents, namely a query planner, an evidence filter, and an emotion generator, which collaboratively perform analytical reasoning to retrieve cross-modal samples, assess evidence, and generate predictions. These agents are optimized end-to-end using Multi-Agent Proximal Policy Optimization (MAPPO) with a shared affective reward to ensure consistent emotion understanding. Furthermore, we introduce Modality-Balancing Mixture of Experts (MB-MoE) and Retrieval-Augmented Adaptive Fusion (RAAF), where MB-MoE dynamically regulates the contributions of different modalities to mitigate representation mismatch caused by cross-modal heterogeneity, while RAAF enhances semantic completion under missing-modality conditions by incorporating retrieved audiovisual embeddings. Extensive experiments on MER-UniBench demonstrate that AffectAgent achieves superior performance across complex scenarios. Our code will be released at: https://github.com/Wz1h1NG/AffectAgent.

Method: Proposes PatchICL, a hierarchical framework combining selective image patching with multi-level supervision. The approach learns to actively identify and attend only to the most informative anatomical regions through a structured patching mechanism.

Result: Achieves competitive in-domain CT segmentation accuracy while reducing compute by 44% at 512×512 resolution compared to UniverSeg baseline. On 35 out-of-domain datasets spanning diverse imaging modalities, outperforms baseline on 6 of 13 modality categories, particularly strong on modalities with localized pathology like OCT and dermoscopy.

Conclusion: PatchICL provides an efficient framework for medical image segmentation that reduces computational burden while maintaining accuracy by focusing attention on relevant anatomical regions through selective patching and supervision.

Abstract: In-context learning (ICL) enables medical image segmentation models to adapt to new anatomical structures from limited examples, reducing the clinical annotation burden. However, standard ICL methods typically rely on dense, global cross-attention, which scales poorly with image resolution. While recent approaches have introduced localized attention mechanisms, they often lack explicit supervision on the selection process, leading to redundant computation in non-informative regions. We propose PatchICL, a hierarchical framework that combines selective image patching with multi-level supervision. Our approach learns to actively identify and attend only to the most informative anatomical regions. Compared to UniverSeg, a strong global-attention baseline, PatchICL achieves competitive in-domain CT segmentation accuracy while reducing compute by 44% at $512\times512$ resolution. On 35 out-of-domain datasets spanning diverse imaging modalities, PatchICL outperforms the baseline on 6 of 13 modality categories, with particular strength on modalities dominated by localized pathology such as OCT and dermoscopy. Training and evaluation code are available at https://github.com/tidiane-camaret/ic_segmentation

Method: Created a comprehensive MoCap dataset capturing single and multi-person scenarios with intricate motions, frequent inter-person occlusions, rapid position exchanges, and varying subject distances. Includes synchronized multi-view RGB and depth sequences, accurate camera calibration, ground-truth 3D motion from Vicon system, and corresponding SMPL/SMPL-X parameters.

Result: Benchmarking state-of-the-art markerless MoCap models shows substantial performance degradation under realistic conditions, exposing limitations of current approaches. Targeted fine-tuning improves generalization, validating the dataset’s realism and value for model development.

Conclusion: The dataset provides a rigorous foundation for advancing robust markerless 4D human motion capture by exposing critical gaps in existing models and enabling better evaluation and improvement of markerless MoCap systems for real-world applications.

Abstract: Marker-based motion capture (MoCap) systems have long been the gold standard for accurate 4D human modeling, yet their reliance on specialized hardware and markers limits scalability and real-world deployment. Advancing reliable markerless 4D human motion capture requires datasets that reflect the complexity of real-world human interactions. Yet, existing benchmarks often lack realistic multi-person dynamics, severe occlusions, and challenging interaction patterns, leading to a persistent domain gap. In this work, we present a new dataset and evaluation for complex 4D markerless human motion capture. Our proposed MoCap dataset captures both single and multi-person scenarios with intricate motions, frequent inter-person occlusions, rapid position exchanges between similarly dressed subjects, and varying subject distances. It includes synchronized multi-view RGB and depth sequences, accurate camera calibration, ground-truth 3D motion capture from a Vicon system, and corresponding SMPL/SMPL-X parameters. This setup ensures precise alignment between visual observations and motion ground truth. Benchmarking state-of-the-art markerless MoCap models reveals substantial performance degradation under these realistic conditions, highlighting limitations of current approaches. We further demonstrate that targeted fine-tuning improves generalization, validating the dataset’s realism and value for model development. Our evaluation exposes critical gaps in existing models and provides a rigorous foundation for advancing robust markerless 4D human motion capture.

Method: CLASP uses class-adaptive layer fusion to construct category-specific visual representations from multi-layer vision features, then performs dual-stage pruning with attention-salient pivot tokens for relevance and redundancy-aware completion tokens for coverage, enabling prompt-conditioned feature fusion and budget allocation.

Result: Extensive experiments show CLASP consistently outperforms existing methods across a wide range of benchmarks, pruning ratios, and MLLM architectures, enabling aggressive yet robust visual token reduction.

Conclusion: CLASP provides an effective plug-and-play solution for reducing computational overhead in MLLMs through adaptive token reduction that maintains performance across diverse instructions and scenarios.

Abstract: Multimodal Large Language Models (MLLMs) suffer from substantial computational overhead due to the high redundancy in visual token sequences. Existing approaches typically address this issue using single-layer Vision Transformer (ViT) features and static pruning strategies. However, such fixed configurations are often brittle under diverse instructions. To overcome these limitations, we propose CLASP, a plug-and-play token reduction framework based on class-adaptive layer fusion and dual-stage pruning. Specifically, CLASP first constructs category-specific visual representations through multi-layer vision feature fusion. It then performs dual-stage pruning, allocating the token budget between attention-salient pivot tokens for relevance and redundancy-aware completion tokens for coverage. Through class-adaptive pruning, CLASP enables prompt-conditioned feature fusion and budget allocation, allowing aggressive yet robust visual token reduction. Extensive experiments demonstrate that CLASP consistently outperforms existing methods across a wide range of benchmarks, pruning ratios, and MLLM architectures. Code will be available at https://github.com/Yunkaidang/CLASP.

Method: Proposes DuSE with two streams: 1) Hierarchical Temporal Prompt Cluster (HTPC) simulates cognitive priming effect using linguistic cues to modulate visual processing, 2) Latent Semantic Emotion Aggregator (LSEA) models knowledge integration process akin to Conceptual Act Theory, aggregating sensory inputs with learned conceptual knowledge.

Result: Extensive experiments on challenging in-the-wild benchmarks demonstrate state-of-the-art performance in dynamic facial expression recognition and enhanced model interpretability.

Conclusion: Emulating the brain’s strategies for emotion processing yields superior performance and interpretability, validating the cognition-centric approach for dynamic emotion modeling.

Abstract: The human brain constructs emotional percepts not by processing facial expressions in isolation, but through a dynamic, hierarchical integration of sensory input with semantic and contextual knowledge. However, existing vision-based dynamic emotion modeling approaches often neglect emotion perception and cognitive theories. To bridge this gap between machine and human emotion perception, we propose cognition-inspired Dual-stream Semantic Enhancement (DuSE). Our model instantiates a dual-stream cognitive architecture. The first stream, a Hierarchical Temporal Prompt Cluster (HTPC), operationalizes the cognitive priming effect. It simulates how linguistic cues pre-sensitize neural pathways, modulating the processing of incoming visual stimuli by aligning textual semantics with fine-grained temporal features of facial dynamics. The second stream, a Latent Semantic Emotion Aggregator (LSEA), computationally models the knowledge integration process, akin to the mechanism described by the Conceptual Act Theory. It aggregates sensory inputs and synthesizes them with learned conceptual knowledge, reflecting the role of the hippocampus and default mode network in constructing a coherent emotional experience. By explicitly modeling these neuro-cognitive mechanisms, DuSE provides a more neurally plausible and robust framework for dynamic facial expression recognition (DFER). Extensive experiments on challenging in-the-wild benchmarks validate our cognition-centric approach, demonstrating that emulating the brain’s strategies for emotion processing yields state-of-the-art performance and enhances model interpretability.

Method: CAAT identifies parameters most critical for adversarial robustness, then uses parameter-efficient fine-tuning (PEFT) to adjust only selected modules where critical parameters exceed a threshold, avoiding full model fine-tuning.

Result: CAAT achieves only 4.3% decrease in adversarial robustness compared to plain adversarial training while tuning only ~6% of parameters, outperforming state-of-the-art lightweight AT methods on three adversarial learning datasets.

Conclusion: CAAT enables scalable adversarial training for large vision transformers by efficiently identifying and fine-tuning only robustness-critical parameters, paving the way for adversarial training at scale.

Abstract: Vision Transformer (ViT) models have achieved remarkable performance across various vision tasks, with scalability being a key advantage when applied to large datasets. This scalability enables ViT models to exhibit strong generalization capabilities. However, as the number of parameters increases, the robustness of ViT models to adversarial examples does not scale proportionally. Adversarial training (AT), one of the most effective methods for enhancing robustness, typically requires fine-tuning the entire model, leading to prohibitively high computational costs, especially for large ViT architectures. In this paper, we aim to robustly fine-tune only a small subset of parameters to achieve robustness comparable to standard AT. To accomplish this, we introduce Criticality-Aware Adversarial Training (CAAT), a novel method that adaptively allocates resources to the most robustness-critical parameters, fine-tuning only selected modules. Specifically, CAAT efficiently identifies parameters that contribute most to adversarial robustness. It then leverages parameter-efficient fine-tuning (PEFT) to robustly adjust weight matrices where the number of critical parameters exceeds a predefined threshold. CAAT exhibits favorable generalization when scaled to larger vision transformer architectures, potentially paving the way for adversarial training at scale, e.g, compared with plain adversarial training, CAAT incurs only a 4.3% decrease in adversarial robustness while tuning approximately 6% of its parameters. Extensive experiments on three widely used adversarial learning datasets demonstrate that CAAT outperforms state-of-the-art lightweight AT methods with fewer trainable parameters.

Method: The authors systematically evaluate adversarial robustness of three representative reconstruction-based detectors across four diverse generative backbone models. They construct white-box adversarial attacks, test transferability between detectors, and assess common countermeasures against adversarial attacks.

Result: Adversarial perturbations cause detection accuracy to collapse to near zero. Attacks demonstrate strong transferability between different detectors, enabling black-box attacks. Standard defense methods provide limited mitigation due to low signal-to-noise ratio in attacked samples.

Conclusion: Reconstruction-based detectors have fundamental security limitations that make them vulnerable to adversarial attacks. The findings highlight the need to rethink existing detection strategies for AI-generated content.

Abstract: Recently, detecting AI-generated images produced by diffusion-based models has attracted increasing attention due to their potential threat to safety. Among existing approaches, reconstruction-based methods have emerged as a prominent paradigm for this task. However, we find that such methods exhibit severe security vulnerabilities to adversarial perturbations; that is, by adding imperceptible adversarial perturbations to input images, the detection accuracy of classifiers collapses to near zero. To verify this threat, we present a systematic evaluation of the adversarial robustness of three representative detectors across four diverse generative backbone models. First, we construct adversarial attacks in white-box scenarios, which degrade the performance of all well-trained detectors. Moreover, we find that these attacks demonstrate transferability; specifically, attacks crafted against one detector can be transferred to others, indicating that adversarial attacks on detectors can also be constructed in a black-box setting. Finally, we assess common countermeasures and find that standard defense methods against adversarial attacks provide limited mitigation. We attribute these failures to the low signal-to-noise ratio (SNR) of attacked samples as perceived by the detectors. Overall, our results reveal fundamental security limitations of reconstruction-based detectors and highlight the need to rethink existing detection strategies.

Method: Bridges modality gap between asynchronous events and spatial generative models by projecting events into intermediate intensity representation, using pretrained models to synthesize realistic non-existent identities, and re-encoding features back into neuromorphic domain.

Result: Method reliably prevents identity recovery from E2V reconstructions while preserving structural data integrity required for downstream vision tasks. Introduces novel synchronized real-world event and RGB dataset for evaluation.

Conclusion: First framework to resolve utility-privacy trade-off in neuromorphic vision through generative anonymization, enabling privacy-preserving deployment of event-based sensors in public spaces.

Abstract: Neuromorphic vision sensors offer low latency and high dynamic range, but their deployment in public spaces raises severe data protection concerns. Recent Event-to-Video (E2V) models can reconstruct high-fidelity intensity images from sparse event streams, inadvertently exposing human identities. Current obfuscation methods, such as masking or scrambling, corrupt the spatio-temporal structure, severely degrading data utility for downstream perception tasks. In this paper, to the best of our knowledge, we present the first generative anonymization framework for event streams to resolve this utility-privacy trade-off. By bridging the modality gap between asynchronous events and standard spatial generative models, our pipeline projects events into an intermediate intensity representation, leverages pretrained models to synthesize realistic, non-existent identities, and re-encodes the features back into the neuromorphic domain. Experiments demonstrate that our method reliably prevents identity recovery from E2V reconstructions while preserving the structural data integrity required for downstream vision tasks. Finally, to facilitate rigorous evaluation, we introduce a novel, synchronized real-world event and RGB dataset captured via precise robotic trajectories, providing a robust benchmark for future research in privacy-preserving neuromorphic vision.

Method: Introduces rotation group equivariant convolutions for rotation equivariant I2I framework, plus transformation learnable equivariant convolutions (TL-Conv) that adaptively learn transformation groups with theoretical analysis of equivariance error.

Result: Extensive experiments across various I2I tasks demonstrate superior performance and effectiveness of the approach in enhancing generation quality.

Conclusion: Equivariant networks with symmetry priors significantly improve image-to-image translation, with TL-Conv providing adaptive transformation learning and theoretical guarantees.

Abstract: Image-to-image translation (I2I) is a fundamental task in computer vision, focused on mapping an input image from a source domain to a corresponding image in a target domain while preserving domain-invariant features and adapting domain-specific attributes. Despite the remarkable success of deep learning-based I2I approaches, the lack of paired data and unsupervised learning framework still hinder their effectiveness. In this work, we address the challenge by incorporating transformation symmetry priors into image-to-image translation networks. Specifically, we introduce rotation group equivariant convolutions to achieve rotation equivariant I2I framework, a novel contribution, to the best of our knowledge, along this research direction. This design ensures the preservation of rotation symmetry, one of the most intrinsic and domain-invariant properties of natural and scientific images, throughout the network. Furthermore, we conduct a systematic study on image symmetry priors on real dataset and propose a novel transformation learnable equivariant convolutions (TL-Conv) that adaptively learns transformation groups, enhancing symmetry preservation across diverse datasets. We also provide a theoretical analysis of the equivariance error of TL-Conv, proving that it maintains exact equivariance in continuous domains and provide a bound for the error in discrete cases. Through extensive experiments across a range of I2I tasks, we validate the effectiveness and superior performance of our approach, highlighting the potential of equivariant networks in enhancing generation quality and its broad applicability. Our code is available at https://github.com/tanfy929/Equivariant-I2I

Method: Proposes ConvBEERS - a Convolutional Board-ready Embedded and Efficient Restoration model using a lightweight non-generative residual convolutional network trained on simulated satellite data.

Result: Achieves +6.9dB PSNR improvement on simulated datasets and real Pleiades-HR imagery, improves downstream object detection by up to +5.1% mAP@50, and reduces latency by ~41x on FPGA deployment.

Conclusion: Lightweight CNNs can achieve competitive restoration quality while addressing real-world constraints in spaceborne systems, enabling practical onboard processing.

Abstract: Satellite image restoration aims to improve image quality by compensating for degradations (e.g., noise and blur) introduced by the imaging system and acquisition conditions. As a fundamental preprocessing step, restoration directly impacts both ground-based product generation and emerging onboard AI applications. Traditional restoration pipelines based on sequential physical models are computationally intensive and slow, making them unsuitable for onboard environments. In this paper, we introduce ConvBEERS: a Convolutional Board-ready Embedded and Efficient Restoration model for Space to investigate whether a light and non-generative residual convolutional network, trained on simulated satellite data, can match or surpass a traditional ground-processing restoration pipeline across multiple operating conditions. Experiments conducted on simulated datasets and real Pleiades-HR imagery demonstrate that the proposed approach achieves competitive image quality, with a +6.9dB PSNR improvement. Evaluation on a downstream object detection task demonstrates that restoration significantly improves performance, with up to +5.1% mAP@50. In addition, successful deployment on a Xilinx Versal VCK190 FPGA validates its practical feasibility for satellite onboard processing, with a ~41x reduction in latency compared to the traditional pipeline. These results demonstrate the relevance of using lightweight CNNs to achieve competitive restoration quality while addressing real-world constraints in spaceborne systems.

Method: Using the CAMUS echocardiography dataset, the researchers simulated three types of labeling errors. They compared two error detection methods: loss-based detection vs. Variance of Gradients (VOG). They also proposed a pseudo-labeling approach to refurbish suspected erroneous ground truth labels. Performance was assessed under varying error levels.

Result: VOG proved highly effective in flagging erroneous ground truth labels during training. Standard U-Net maintained strong performance under random label errors and moderate systematic errors (up to 50%). The detection and refurbishment approach improved performance, particularly under high-error conditions.

Conclusion: The study demonstrates that while standard models show some robustness to labeling errors, systematic error detection methods like VOG combined with pseudo-labeling refurbishment can significantly improve performance, especially in high-error scenarios common in medical imaging.

Abstract: Deep learning-based medical image segmentation typically relies on ground truth (GT) labels obtained through manual annotation, but these can be prone to random errors or systematic biases. This study examines the robustness of deep learning models to such errors in echocardiography (echo) segmentation and evaluates a novel strategy for detecting and refurbishing erroneous labels during model training. Using the CAMUS dataset, we simulate three error types, then compare a loss-based GT label error detection method with one based on Variance of Gradients (VOG). We also propose a pseudo-labelling approach to refurbish suspected erroneous GT labels. We assess the performance of our proposed approach under varying error levels. Results show that VOG proved highly effective in flagging erroneous GT labels during training. However, a standard U-Net maintained strong performance under random label errors and moderate levels of systematic errors (up to 50%). The detection and refurbishment approach improved performance, particularly under high-error conditions.

Method: Proposes Multimodal Semantic Lighting Attacks (MSLA) - a physically deployable adversarial attack framework that uses controllable adversarial lighting to disrupt multimodal semantic understanding. Attacks semantic alignment rather than only task-specific outputs.

Result: MSLA degrades zero-shot classification performance of mainstream CLIP variants while inducing severe semantic hallucinations in advanced VLMs such as LLaVA and BLIP across image captioning and VQA. Effective, transferable, and practically realizable in both digital and physical domains.

Conclusion: VLMs are highly vulnerable to physically deployable semantic attacks, exposing a previously overlooked robustness gap and underscoring the urgent need for physical-world robustness evaluation of VLMs.

Abstract: Vision-Language Models (VLMs) have shown remarkable performance, yet their security remains insufficiently understood. Existing adversarial studies focus almost exclusively on the digital setting, leaving physical-world threats largely unexplored. As VLMs are increasingly deployed in real environments, this gap becomes critical, since adversarial perturbations must be physically realizable. Despite this practical relevance, physical attacks against VLMs have not been systematically studied. Such attacks may induce recognition failures and further disrupt multimodal reasoning, leading to severe semantic misinterpretation in downstream tasks. Therefore, investigating physical attacks on VLMs is essential for assessing their real-world security risks. To address this gap, we propose Multimodal Semantic Lighting Attacks (MSLA), the first physically deployable adversarial attack framework against VLMs. MSLA uses controllable adversarial lighting to disrupt multimodal semantic understanding in real scenes, attacking semantic alignment rather than only task-specific outputs. Consequently, it degrades zero-shot classification performance of mainstream CLIP variants while inducing severe semantic hallucinations in advanced VLMs such as LLaVA and BLIP across image captioning and visual question answering (VQA). Extensive experiments in both digital and physical domains demonstrate that MSLA is effective, transferable, and practically realizable. Our findings provide the first evidence that VLMs are highly vulnerable to physically deployable semantic attacks, exposing a previously overlooked robustness gap and underscoring the urgent need for physical-world robustness evaluation of VLMs.

Method: Uses flow-matching framework with MIDI as privileged modality during training, asymmetric role-gated interaction module for cross-hand coordination, and autoregressive flow continuation for streaming generation of long sequences.

Result: Achieves superior quantitative and qualitative performance on PianoMotion10M dataset while accelerating inference by over 9× compared to previous methods.

Conclusion: PianoFlow enables precise, coordinated bimanual piano motion synthesis with real-time streaming capability for arbitrarily long sequences.

Abstract: Audio-driven bimanual piano motion generation requires precise modeling of complex musical structures and dynamic cross-hand coordination. However, existing methods often rely on acoustic-only representations lacking symbolic priors, employ inflexible interaction mechanisms, and are limited to computationally expensive short-sequence generation. To address these limitations, we propose PianoFlow, a flow-matching framework for precise and coordinated bimanual piano motion synthesis. Our approach strategically leverages MIDI as a privileged modality during training, distilling these structured musical priors to achieve deep semantic understanding while maintaining audio-only inference. Furthermore, we introduce an asymmetric role-gated interaction module to explicitly capture dynamic cross-hand coordination through role-aware attention and temporal gating. To enable real-time streaming generation for arbitrarily long sequences, we design an autoregressive flow continuation scheme that ensures seamless cross-chunk temporal coherence. Extensive experiments on the PianoMotion10M dataset demonstrate that PianoFlow achieves superior quantitative and qualitative performance, while accelerating inference by over 9\times compared to previous methods.

Method: VideoFlexTok uses a variable-length sequence of tokens structured coarse-to-fine: early tokens capture abstract semantics and motion, later tokens add fine details. A generative flow decoder enables reconstruction from any token count, allowing adaptive token usage.

Result: Achieves comparable generation quality (gFVD and ViCLIP Score) with 5x smaller models (1.1B vs 5.2B). Enables 10-second 81-frame video generation with 8x fewer tokens than 3D grid tokenizers (672 vs standard).

Conclusion: VideoFlexTok provides more efficient video representation for generation tasks, enabling longer video modeling with reduced computational cost while maintaining quality.

Abstract: Visual tokenizers map high-dimensional raw pixels into a compressed representation for downstream modeling. Beyond compression, tokenizers dictate what information is preserved and how it is organized. A de facto standard approach to video tokenization is to represent a video as a spatiotemporal 3D grid of tokens, each capturing the corresponding local information in the original signal. This requires the downstream model that consumes the tokens, e.g., a text-to-video model, to learn to predict all low-level details “pixel-by-pixel” irrespective of the video’s inherent complexity, leading to high learning complexity. We present VideoFlexTok, which represents videos with a variable-length sequence of tokens structured in a coarse-to-fine manner – where the first tokens (emergently) capture abstract information, such as semantics and motion, and later tokens add fine-grained details. The generative flow decoder enables realistic video reconstructions from any token count. This representation structure allows adapting the token count according to downstream needs and encoding videos longer than the baselines with the same budget. We evaluate VideoFlexTok on class- and text-to-video generative tasks and show that it leads to more efficient training compared to 3D grid tokens, e.g., achieving comparable generation quality (gFVD and ViCLIP Score) with a 5x smaller model (1.1B vs 5.2B). Finally, we demonstrate how VideoFlexTok can enable long video generation without prohibitive computational cost by training a text-to-video model on 10-second 81-frame videos with only 672 tokens, 8x fewer than a comparable 3D grid tokenizer.

Method: Proposes a file-based visual representation mechanism that offloads visual assets to external file system and maps them to lightweight textual identifiers (UIDs), reducing context overhead while preserving multimodal information. Includes a fetch-image tool for progressive on-demand visual loading and a data synthesis pipeline for generating complex cross-modal multi-hop reasoning queries.

Result: Achieves state-of-the-art performance among open-source models on challenging long-horizon benchmarks like MM-BrowseComp and MMSearch-Plus, successfully scales to 100-turn search horizons, and shows strong generalizability across different base models.

Conclusion: LMM-Searcher effectively addresses long-horizon multimodal deep search challenges through file-based visual representation, enabling efficient management of multimodal information over extended interactions while maintaining strong performance.

Abstract: Multimodal deep search agents have shown great potential in solving complex tasks by iteratively collecting textual and visual evidence. However, managing the heterogeneous information and high token costs associated with multimodal inputs over long horizons remains a critical challenge, as existing methods often suffer from context explosion or the loss of crucial visual signals. To address this, we propose a novel Long-horizon MultiModal deep search framework, named LMM-Searcher, centered on a file-based visual representation mechanism. By offloading visual assets to an external file system and mapping them to lightweight textual identifiers (UIDs), our approach mitigates context overhead while preserving multimodal information for future access. We equip the agent with a tailored fetch-image tool, enabling a progressive, on-demand visual loading strategy for active perception. Furthermore, we introduce a data synthesis pipeline designed to generate queries requiring complex cross-modal multi-hop reasoning. Using this pipeline, we distill 12K high-quality trajectories to fine-tune Qwen3-VL-Thinking-30A3B into a specialized multimodal deep search agent. Extensive experiments across four benchmarks demonstrate that our method successfully scales to 100-turn search horizons, achieving state-of-the-art performance among open-source models on challenging long-horizon benchmarks like MM-BrowseComp and MMSearch-Plus, while also exhibiting strong generalizability across different base models. Our code will be released in https://github.com/RUCAIBox/LMM-Searcher.

Method: CUBE uses a lattice of high-dimensional control features instead of 3D control points. It employs a two-stage mapping: first blending control features using B-spline bases to create a feature vector (first three values define base mesh), then using an MLP to predict residual displacements for refined 3D coordinates. For reconstruction, CUBE is queried at coordinates from a fixed template mesh.

Result: CUBE achieves state-of-the-art scan registration results compared to recent baselines when trained with transformer-based encoders to predict control features from point clouds and monocular images.

Conclusion: CUBE provides an expressive geometric representation for human faces that combines B-spline volume benefits with learned features, enabling high-quality 3D reconstruction while maintaining local editing capabilities through its local support property.

Abstract: We present CUBE (Control-based Unified B-spline Encoding), a new geometric representation for human faces that combines B-spline volumes with learned features, and demonstrate its use as a decoder for 3D scan registration and monocular 3D face reconstruction. Unlike existing B-spline representations with 3D control points, CUBE is parametrized by a lattice (e.g., 8 x 8 x 8) of high-dimensional control features, increasing the model’s expressivity. These features define a continuous, two-stage mapping from a 3D parametric domain to 3D Euclidean space via an intermediate feature space. First, high-dimensional control features are locally blended using the B-spline bases, yielding a high-dimensional feature vector whose first three values define a 3D base mesh. A small MLP then processes this feature vector to predict a residual displacement from the base shape, yielding the final refined 3D coordinates. To reconstruct 3D surfaces in dense semantic correspondence, CUBE is queried at 3D coordinates sampled from a fixed template mesh. Crucially, CUBE retains the local support property of traditional B-spline representations, enabling local surface editing by updating individual control features. We demonstrate the strengths of this representation by training transformer-based encoders to predict CUBE’s control features from unstructured point clouds and monocular images, achieving state-of-the-art scan registration results compared to recent baselines.

Method: P² rewrites tool outputs into compact, structured, language-native summaries that MLLMs can directly parse and reason over. It’s training-free and model-agnostic, working without any model modifications or additional training.

Result: Across six perception tasks in BLINK, P² achieved large improvements: GPT-5 Mini accuracy increased from 41.35% to 86.47% on multi-view reasoning, 52.42% to 81.45% on relative depth, with 22% average gain. Smaller MLLMs (InternVL3.5-4B, Qwen3VL-4B) showed 15-40% absolute gains, surpassing prior tool-use methods.

Conclusion: The bottleneck in vision tool-augmented MLLMs is representation alignment, not more tools or larger models. P² demonstrates that converting tool outputs to language-native summaries enables MLLMs to effectively leverage visual cues for perception tasks.

Abstract: Multimodal language models (MLLMs) are increasingly paired with vision tools (e.g., depth, flow, correspondence) to enhance visual reasoning. However, despite access to these tool-generated visual cues, MLLMs often fail to benefit from them. Existing approaches typically feed raw tool outputs into the model, but these dense, pixel-level representations are misaligned with the language-native reasoning strengths of LLMs, leading to weak perception and reliance on language priors. We argue that, in problems where vision tools can provide the necessary visual cues, the bottleneck is not more tool calls or larger MLLMs, it is how tool outputs are represented. We introduce Perception Programs (P$^2$), a training-free, model-agnostic method that rewrites tool outputs into compact, structured, language-native summaries that MLLMs can directly parse and reason over. Across six perception-centric tasks in BLINK, P$^2$ consistently yields large improvements over base models and raw tool-augmented baselines. With GPT-5 Mini as the base model, P$^2$ raises its accuracy from 41.35% to 86.47% on multi-view reasoning, from 52.42% to 81.45% on relative depth, and achieves a 22% average gain across tasks, setting new state-of-the-art results. Even on smaller MLLMs, e.g., InternVL3.5-4B and Qwen3VL-4B, we observe 15-40% absolute gains from P$^2$, surpassing prior agentic, supervised, and RL-based tool-use methods-without any training or model modifications.

Method: 1) FISD benchmark: Uses generative models to precisely control reference-target image pair variables across six dimensions without query ambiguity. 2) Multi-round agentic evaluation framework: Automatically probes models’ adaptation in interactive scenarios over successive query rounds.

Result: Extensive experiments show the value of the novel evaluation approach for typical CIR methods, providing more accurate performance characterization and realistic appraisal of practical efficacy.

Conclusion: The proposed FISD benchmark and multi-round evaluation framework address critical limitations in CIR assessment, enabling more accurate evaluation and better understanding of model capabilities in practical interactive applications.

Abstract: Composed Image Retrieval (CIR) aims to retrieve a target image based on a query composed of a reference image, and a relative caption that specifies the desired modification. Despite the rapid development of CIR models, their performance is not well characterized by existing benchmarks, which inherently contain indeterminate queries degrading the evaluation (i.e., multiple candidate images, rather than solely the target image, meet the query criteria), and have not considered their effectiveness in the context of the multi-round system. Motivated by this, we consider improving the evaluation procedure from two aspects: 1) we introduce FISD, a Fully-Informed Semantically-Diverse benchmark, which employs generative models to precisely control the variables of reference-target image pairs, enabling a more accurate evaluation of CIR methods across six dimensions, without query ambiguity; 2) we propose an automatic multi-round agentic evaluation framework to probe the potential of the existing models in the interactive scenarios. By observing how models adapt and refine their choices over successive rounds of queries, this framework provides a more realistic appraisal of their efficacy in practical applications. Extensive experiments and comparisons prove the value of our novel evaluation on typical CIR methods.

Method: Created a laboratory setup to collect stereo image pairs with controlled degradation levels across four scenarios, providing aligned stereo images with pixel-wise consistent clear ground truths for each degradation level.

Result: The dataset enables fine-grained evaluation of restoration methods with increasing degradation severity and supports two restoration tasks: single-level and mixed-level degradation, establishing a better controlled and more realistic benchmark.

Conclusion: M3D-Stereo provides a comprehensive stereo dataset for evaluating image restoration and stereo matching methods in complex degradation environments, addressing limitations of existing datasets.

Abstract: Image restoration under adverse conditions, such as underwater, haze or fog, and low-light environments, remains a highly challenging problem due to complex physical degradations and severe information loss. Existing datasets are predominantly limited to a single degradation type or heavily rely on synthetic data without stereo consistency, inherently restricting their applicability in real-world scenarios. To address this, we introduce M3D-Stereo, a stereo dataset with 7904 high-resolution image pairs for image restoration research acquired in multiple media with multiple controlled degradation levels. It encompasses four degradation scenarios: underwater scatter, haze/fog, underwater low-light, and haze low-light. Each scenario forms a subset, and is divided into six levels of progressive degradation, allowing fine-grained evaluations of restoration methods with increasing severity of degradation. Collected via a laboratory setup, the dataset provides aligned stereo image pairs along with their pixel-wise consistent clear ground truths. Two restoration tasks, single-level and mixed-level degradation, were performed to verify its validity. M3D-Stereo establishes a better controlled and more realistic benchmark to evaluate image restoration and stereo matching methods in complex degradation environments. It is made public under LGPLv3 license.

Method: Proposes CTAB (Cross-Task Attention Bridge), a bidirectional module that exchanges features between detection and segmentation branches via multi-scale deformable attention in shared BEV space. Integrated into a multi-task framework with Instance Normalization-based segmentation decoder and learnable BEV upsampling.

Result: On nuScenes dataset, CTAB improves segmentation on 7 classes over joint multi-task baseline while maintaining essentially neutral detection performance. On a 4-class subset (drivable area, pedestrian crossing, walkway, vehicle), achieves comparable mIoU while simultaneously providing 3D detection.

Conclusion: CTAB enables effective cross-task feature sharing between detection and segmentation in BEV space, providing a more detailed BEV representation for autonomous driving perception without compromising detection performance.

Abstract: Bird’s-eye-view (BEV) representations are the dominant paradigm for 3D perception in autonomous driving, providing a unified spatial canvas where detection and segmentation features are geometrically registered to the same physical coordinate system. However, existing radar-camera fusion methods treat these tasks in isolation, missing the opportunity to share complementary information between them: detection features encode object-level geometry that can sharpen segmentation boundaries, while segmentation features provide dense semantic context that can anchor detection. We propose \textbf{CTAB} (Cross-Task Attention Bridge), a bidirectional module that exchanges features between detection and segmentation branches via multi-scale deformable attention in shared BEV space. CTAB is integrated into a multi-task framework with an Instance Normalization-based segmentation decoder and learnable BEV upsampling to provide a more detailed BEV representation. On nuScenes, CTAB improves segmentation on 7 classes over the joint multi-task baseline at essentially neutral detection. On a 4-class subset (drivable area, pedestrian crossing, walkway, vehicle), our joint multi-task model reaches comparable mIoU on 4 classes while simultaneously providing 3D detection.

Method: Pi-HOC uses a single-pass framework that: 1) detects instances, 2) creates dedicated human-object tokens for each pair, 3) refines them with an InteractionFormer, and 4) uses a SAM-based decoder to predict dense contact on SMPL human meshes for each human-object pair.

Result: On MMHOI and DAMON datasets, Pi-HOC significantly improves accuracy and localization over state-of-the-art methods while achieving 20x higher throughput. It also enables improved SAM-3D image-to-mesh reconstruction via test-time optimization and referential contact prediction from language queries without additional training.

Conclusion: Pi-HOC provides an efficient, accurate solution for dense 3D semantic contact prediction in multi-human scenarios, with strong performance improvements and practical applications in downstream tasks like mesh reconstruction and language-guided interaction understanding.

Abstract: Resolving real-world human-object interactions in images is a many-to-many challenge, in which disentangling fine-grained concurrent physical contact is particularly difficult. Existing semantic contact estimation methods are either limited to single-human settings or require object geometries (e.g., meshes) in addition to the input image. Current state-of-the-art leverages powerful VLM for category-level semantics but struggles with multi-human scenarios and scales poorly in inference. We introduce Pi-HOC, a single-pass, instance-aware framework for dense 3D semantic contact prediction of all human-object pairs. Pi-HOC detects instances, creates dedicated human-object (HO) tokens for each pair, and refines them using an InteractionFormer. A SAM-based decoder then predicts dense contact on SMPL human meshes for each human-object pair. On the MMHOI and DAMON datasets, Pi-HOC significantly improves accuracy and localization over state-of-the-art methods while achieving 20x higher throughput. We further demonstrate that predicted contacts improve SAM-3D image-to-mesh reconstruction via a test-time optimization algorithm and enable referential contact prediction from language queries without additional training.

Method: Uses a compact Sum-of-Gaussians (SoG) representation for efficient tracking. Initializes object pose/geometry using video-adapted SAM3D pipeline, then converts to lightweight SoG via subsampling. For hands, starts with off-the-shelf monocular pose initialization and refines with 2D joint and depth alignment losses, avoiding detailed 3D appearance models.

Result: Achieves 6.4x faster reconstruction than prior work on long sequences, improves object reconstruction by 13.4%, and reduces hand’s per-joint position error by over 65% on public benchmarks.

Conclusion: GraG demonstrates that efficient tracking of dynamic hand-object interactions can be achieved using compact representations like Sum-of-Gaussians, offering significant speed improvements while maintaining or improving reconstruction quality.

Abstract: We present Grasp in Gaussians (GraG), a fast and robust method for reconstructing dynamic 3D hand-object interactions from a single monocular video. Unlike recent approaches that optimize heavy neural representations, our method focuses on tracking the hand and the object efficiently, once initialized from pretrained large models. Our key insight is that accurate and temporally stable hand-object motion can be recovered using a compact Sum-of-Gaussians (SoG) representation, revived from classical tracking literature and integrated with generative Gaussian-based initializations. We initialize object pose and geometry using a video-adapted SAM3D pipeline, then convert the resulting dense Gaussian representation into a lightweight SoG via subsampling. This compact representation enables efficient and fast tracking while preserving geometric fidelity. For the hand, we adopt a complementary strategy: starting from off-the-shelf monocular hand pose initialization, we refine hand motion using simple yet effective 2D joint and depth alignment losses, avoiding per-frame refinement of a detailed 3D hand appearance model while maintaining stable articulation. Extensive experiments on public benchmarks demonstrate that GraG reconstructs temporally coherent hand-object interactions on long sequences 6.4x faster than prior work while improving object reconstruction by 13.4% and reducing hand’s per-joint position error by over 65%.

Method: The authors introduce a task alignment proxy that enables efficient hyperparameter selection for model merging without requiring full decoder training. This proxy allows merging of models with heterogeneous decoders by speeding up hyperparameter optimization.

Result: The task alignment proxy reduces hyperparameter selection time by orders of magnitude while maintaining performance. This enables model merging for multi-task vision scenarios beyond simple CLIP-based classification.

Conclusion: The proposed task alignment proxy makes model merging practical for real-world vision applications with heterogeneous decoders, extending the applicability of model merging techniques to more challenging multi-task scenarios.

Abstract: Efficiently merging several models fine-tuned for different tasks, but stemming from the same pretrained base model, is of great practical interest. Despite extensive prior work, most evaluations of model merging in computer vision are restricted to image classification using CLIP, where different classification datasets define different tasks. In this work, our goal is to make model merging more practical and show its relevance on challenging scenarios beyond this specific setting. In most vision scenarios, different tasks rely on trainable and usually heterogeneous decoders. Differently from previous studies with frozen decoders, where merged models can be evaluated right away, the non-trivial cost of decoder training renders hyperparameter selection based on downstream performance impractical. To address this, we introduce the task alignment proxy, and show how it can be used to speed up hyperparameter selection by orders of magnitude while retaining performance. Equipped with the task alignment proxy, we extend the applicability of model merging to multi-task vision models beyond CLIP-based classification.

Method: Proposes Distribution-Discrepancy Condensation (DDC) to model real-to-fake discrepancy via surrogate factorization in characteristic-function space, condensing it into a tiny bank of distribution discrepancy maps. Uses Manifold-Consistent Replay (MCR) to synthesize replay samples through variance-preserving composition of these maps with current-stage real faces.

Result: The framework consistently outperforms prior CFFD baselines, significantly mitigates catastrophic forgetting, operates under extremely small memory budgets without storing raw historical face images, and reduces identity leakage risk compared to selection-based replay.

Conclusion: The proposed approach effectively addresses continual face forgery detection challenges by focusing on distribution-level replay rather than sample-level replay, achieving better performance with improved privacy protection.

Abstract: Continual face forgery detection (CFFD) requires detectors to learn emerging forgery paradigms without forgetting previously seen manipulations. Existing CFFD methods commonly rely on replaying a small amount of past data to mitigate forgetting. Such replay is typically implemented either by storing a few historical samples or by synthesizing pseudo-forgeries from detector-dependent perturbations. Under strict memory budgets, the former cannot adequately cover diverse forgery cues and may expose facial identities, while the latter remains strongly tied to past decision boundaries. We argue that the core role of replay in CFFD is to reinstate the distributions of previous forgery tasks during subsequent training. To this end, we directly condense the discrepancy between real and fake distributions and leverage real faces from the current stage to perform distribution-level replay. Specifically, we introduce Distribution-Discrepancy Condensation (DDC), which models the real-to-fake discrepancy via a surrogate factorization in characteristic-function space and condenses it into a tiny bank of distribution discrepancy maps. We further propose Manifold-Consistent Replay (MCR), which synthesizes replay samples through variance-preserving composition of these maps with current-stage real faces, yielding samples that reflect previous-task forgery cues while remaining compatible with current real-face statistics. Operating under an extremely small memory budget and without directly storing raw historical face images, our framework consistently outperforms prior CFFD baselines and significantly mitigates catastrophic forgetting. Replay-level privacy analysis further suggests reduced identity leakage risk relative to selection-based replay.

Method: Presents a comprehensive survey with systematic taxonomy categorizing hallucinations into two core types: dynamic distortion (with two subtypes) and content fabrication (with two subtypes). Reviews evaluation benchmarks, metrics, and intervention strategies, and analyzes root causes.

Result: Identifies that dynamic distortion often results from limited temporal representation capacity, while content fabrication stems from insufficient visual grounding. Provides insights into current evaluation methods and mitigation approaches.

Conclusion: Consolidates scattered progress to foster systematic understanding of hallucinations in Vid-LLMs, laying groundwork for robust video-language systems. Proposes future directions including motion-aware visual encoders and counterfactual learning techniques.

Abstract: Despite significant progress in video-language modeling, hallucinations remain a persistent challenge in Video Large Language Models (Vid-LLMs), referring to outputs that appear plausible yet contradict the content of the input video. This survey presents a comprehensive analysis of hallucinations in Vid-LLMs and introduces a systematic taxonomy that categorizes them into two core types: dynamic distortion and content fabrication, each comprising two subtypes with representative cases. Building on this taxonomy, we review recent advances in the evaluation and mitigation of hallucinations, covering key benchmarks, metrics, and intervention strategies. We further analyze the root causes of dynamic distortion and content fabrication, which often result from limited capacity for temporal representation and insufficient visual grounding. These insights inform several promising directions for future work, including the development of motion-aware visual encoders and the integration of counterfactual learning techniques. This survey consolidates scattered progress to foster a systematic understanding of hallucinations in Vid-LLMs, laying the groundwork for building robust and reliable video-language systems. An up-to-date curated list of related works is maintained at https://github.com/hukcc/Awesome-Video-Hallucination .

Method: Reformulate classical self-supervised pretext tasks (rotation prediction, color matching, cross-view correspondence) as image-instruction-response triplets and inject a small fraction (3-10%) of these visually grounded instructions into training data without architectural changes or additional training stages.

Result: Consistent performance improvements across multiple models, training regimes, and benchmarks on vision-centric evaluations by forcing models to use visual evidence rather than language shortcuts.

Conclusion: Visually grounded self-supervised tasks in instruction tuning serve as a powerful yet simple approach to enhance visual reasoning in MLLMs through adjustments to training data distribution.

Abstract: Multimodal large language models (MLLMs) perform well on many vision-language tasks but often struggle with vision-centric problems that require fine-grained visual reasoning. Recent evidence suggests that this limitation arises not from weak visual representations, but from under-utilization of visual information during instruction tuning, where many tasks can be partially solved using language priors alone. We propose a simple and lightweight approach that augments visual instruction tuning with a small number of visually grounded self-supervised tasks expressed as natural language instructions. By reformulating classical self-supervised pretext tasks, such as rotation prediction, color matching, and cross-view correspondence, as image-instruction-response triplets, we introduce supervision that cannot be solved without relying on visual evidence. Our approach requires no human annotations, no architectural modifications, and no additional training stages. Across multiple models, training regimes, and benchmarks, injecting only a small fraction (3-10%) of such visually grounded instructions consistently improves performance on vision-centric evaluations. Our findings highlight instruction tuning with visually grounded SSL tasks as a powerful lever for improving visual reasoning in MLLMs through simple adjustments to the training data distribution. Code available at: https://github.com/sirkosophia/V-GIFT

Method: Proposes AbdomenGen, a sequential volume-conditioned diffusion framework that uses Volume Control Scalars (VCS) to decouple organ size from body habitus. Organ masks are synthesized sequentially, conditioning on body mask and previously generated structures to preserve anatomical coherence while enabling independent multi-organ control.

Result: Achieves strong geometric fidelity across 11 abdominal organs (e.g., liver dice 0.83 ± 0.05), stable single-organ calibration over [-3,+3] VCS range, and disentangled multi-organ modulation. Wasserstein-based VCS selection reduces distributional distance of training data by 73.6% for hepatomegaly cohort.

Conclusion: AbdomenGen enables calibrated, distribution-aware anatomical generation suitable for controllable abdominal phantom construction and simulation studies in medical imaging research.

Abstract: Computational phantoms are widely used in medical imaging research, yet current systems to generate controlled, clinically meaningful anatomical variations remain limited. We present AbdomenGen, a sequential volume-conditioned diffusion framework for controllable abdominal anatomy generation. We introduce the \textbf{Volume Control Scalar (VCS)}, a standardized residual that decouples organ size from body habitus, enabling interpretable volume modulation. Organ masks are synthesized sequentially, conditioning on the body mask and previously generated structures to preserve global anatomical coherence while supporting independent, multi-organ control. Across 11 abdominal organs, the proposed framework achieves strong geometric fidelity (e.g., liver dice $0.83 \pm 0.05$), stable single-organ calibration over $[-3,+3]$ VCS, and disentangled multi-organ modulation. To showcase clinical utility with a hepatomegaly cohort selected from MERLIN, Wasserstein-based VCS selection reduces distributional distance of training data by 73.6% . These results demonstrate calibrated, distribution-aware anatomical generation suitable for controllable abdominal phantom construction and simulation studies.

Method: Uses LLM to generate hypotheses about neural architectures, maintains a Trajectory Tree for lineage tracking and Hypothesis Memory Bank with confidence scores. Employs evolutionary branching with dual strategy balancing exploitation of validated principles and resolution of uncertain ones. Multiple feedback agents analyze experimental results to update hypothesis confidence.

Result: Achieved 94.11% accuracy on CIFAR-10 (from 18.91% baseline), generalizes to CIFAR-100 and Tiny-ImageNet. Achieved state-of-the-art on MedMNIST. Hypothesis confidence scores become increasingly predictive with evidence accumulation, and learned principles transfer across independent evolutionary lineages.

Conclusion: HypoExplore successfully discovers strong vision architectures while building understanding of design space. The framework demonstrates that hypothesis-driven inquiry can lead to both performance improvements and genuine insights into neural architecture design principles.

Abstract: We introduce HypoExplore, an agentic framework that formulates neural architecture discovery for visual recognition as a hypothesis-driven scientific inquiry. Given a human-specified high-level research direction, HypoExplore ideates, implements, evaluates, and improves neural architectures through evolutionary branching. New hypotheses are created using a large language model by selecting a parent hypothesis to build upon, guided by a dual strategy that balances exploiting validated principles with resolving uncertain ones. Our proposed framework maintains a Trajectory Tree that records the lineage of all proposed architectures, and a Hypothesis Memory Bank that actively tracks confidence scores acquired through experimental evidence. After each experiment, multiple feedback agents analyze the results from different perspectives and consolidate their findings into hypothesis confidence updates. Our framework is tested on discovering lightweight vision architectures on CIFAR-10, with the best achieving 94.11% accuracy evolved from a root node baseline that starts at 18.91%, and generalizes to CIFAR-100 and Tiny-ImageNet. We further demonstrate applicability to a specialized domain by conducting independent architecture discovery runs on MedMNIST, which yield a state-of-the-art performance. We show that hypothesis confidence scores grow increasingly predictive as evidence accumulates, and that the learned principles transfer across independent evolutionary lineages, suggesting that HypoExplore not only discovers stronger architectures, but can help build a genuine understanding of the design space.

Method: Instead of single-step execution, the agent uses an iterative refinement process with visual feedback from previous attempts. This closed-loop grounding mechanism allows self-correction of displacement errors and adaptation to dynamic UI changes.

Result: Evaluation across GPT-5.4, Claude, and Qwen on complex coding benchmarks shows that multi-turn refinement significantly outperforms state-of-the-art single-shot models in both click precision and overall task success rate.

Conclusion: Iterative visual reasoning is a critical component for the next generation of reliable software engineering agents, enabling more precise GUI grounding in dense coding environments.

Abstract: Computer Use Agents (CUAs) fundamentally rely on graphical user interface (GUI) grounding to translate language instructions into executable screen actions, but editing-level grounding in dense coding interfaces, where sub-pixel accuracy is required to interact with dense IDE elements, remains underexplored. Existing approaches typically rely on single-shot coordinate prediction, which lacks a mechanism for error correction and often fails in high-density interfaces. In this technical report, we conduct an empirical study of pixel-precise cursor localization in coding environments. Instead of a single-step execution, our agent engages in an iterative refinement process, utilizing visual feedback from previous attempts to reach the target element. This closed-loop grounding mechanism allows the agent to self-correct displacement errors and adapt to dynamic UI changes. We evaluate our approach across GPT-5.4, Claude, and Qwen on a suite of complex coding benchmarks, demonstrating that multi-turn refinement significantly outperforms state-of-the-art single-shot models in both click precision and overall task success rate. Our results suggest that iterative visual reasoning is a critical component for the next generation of reliable software engineering agents. Code: https://github.com/microsoft/precision-cua-bench.

Method: Used vision-language transfer learning on abdominal CT enterography, comparing slice embedding aggregation methods (mean vs attention pooling), tissue contrast strategies (multi-window RGB encoding vs multiplanar sampling), and report generation approaches (fine-tuning vs retrieval-augmented generation). Employed a three-teacher pseudolabel framework for comparisons without expert annotations.

Result: Mean pooling better for disease classification (59.2% three-class accuracy), attention pooling better for cross-modal retrieval (0.235 text-to-image MRR). Multi-window RGB encoding outperformed spatial coverage strategies. RAG improved report generation by 7-14 percentage points above chance baseline, reducing MAE from 0.98 to 0.80-0.89.

Conclusion: Provides first baselines for CT enterography analysis, showing slice aggregation methods emphasize different representation properties, tissue contrast matters more than spatial coverage, and RAG significantly improves report generation. Offers practical guidance for building vision-language systems in volumetric medical imaging.

Abstract: Computed tomography (CT) enterography is a primary imaging modality for assessing inflammatory bowel disease (IBD), yet the representational choices that best support automated analysis of this modality are unknown. We present the first study of vision-language transfer learning on abdominal CT enterography and identify two main findings. First, mean pooling of slice embeddings gives better categorical disease assessment (59.2% three-class accuracy), whereas attention pooling gives better cross-modal retrieval (0.235 text-to-image MRR). This pattern holds across all LoRA configurations tested and suggests that the two aggregators emphasize different properties of the learned representation. Second, per-slice tissue contrast matters more than broader spatial coverage: multi-window RGB encoding, which maps complementary Hounsfield Unit windows to RGB channels, outperforms all strategies that increase spatial coverage through multiplanar sampling, and in this setting adding coronal and sagittal views reduces classification performance. For report generation, fine-tuning without retrieval context yields within-1 severity accuracy at the prevalence-matched chance level (70.4% vs.\ 71% random), suggesting little learned ordering beyond the class distribution. Retrieval-augmented generation (RAG) improves this across all configurations, scoring 7–14 percentage points above the chance baseline and improving ordinal MAE from 0.98 to 0.80–0.89. A three-teacher pseudolabel framework enables all comparisons without expert annotations. Together, these findings provide the first baselines for this underexplored modality and offer practical guidance for building vision-language systems for volumetric medical imaging.

Method: Proposes rDPO with instance-specific rubrics: for each image-instruction pair, creates checklist-style rubrics with essential and additional criteria to score responses. Builds instruction-rubric pool offline and reuses it for on-policy data construction.

Result: Rubric-based prompting improves 30B-A3B judge close to GPT-5.4; rubric filtering achieves 82.69 macro average vs 75.82 for outcome-based filtering; rDPO scores 61.01 on comprehensive benchmark, outperforming baselines (52.36) and base model (59.48).

Conclusion: Visual preference optimization benefits from combining on-policy data construction with instance-specific criterion-level feedback, demonstrating the effectiveness of rubric-based approaches for multimodal tasks.

Abstract: The effectiveness of Direct Preference Optimization (DPO) depends on preference data that reflect the quality differences that matter in multimodal tasks. Existing pipelines often rely on off-policy perturbations or coarse outcome-based signals, which are not well suited to fine-grained visual reasoning. We propose rDPO, a preference optimization framework based on instance-specific rubrics. For each image-instruction pair, we create a checklist-style rubric of essential and additional criteria to score responses from any possible policies. The instruction-rubric pool is built offline and reused during the construction of on-policy data. On public reward modeling benchmarks, rubric-based prompting massively improves a 30B-A3B judge and brings it close to GPT-5.4. On public downstream benchmarks, rubric-based filtering raises the macro average to 82.69, whereas outcome-based filtering drops it to 75.82 from 81.14. When evaluating scalability on a comprehensive benchmark, rDPO achieves 61.01, markedly outperforming the style-constrained baseline (52.36) and surpassing the 59.48 base model. Together, these results show that visual preference optimization benefits from combining on-policy data construction with instance-specific criterion-level feedback.

Method: Proposes a conflated inverse modeling framework combining a predictive forward model with a diffusion-based generative inverse model. The framework generates diverse, physically plausible image-based vegetation patterns conditioned on specific temperature goals, maintaining control over thermal outcomes while enabling spatial diversity.

Result: The framework successfully produces diverse vegetation configurations that achieve specified temperature reduction targets, even when such combinations are absent from training data. It addresses the ambiguity in inverse urban climate problems where multiple spatial patterns can yield similar aggregated temperature responses.

Conclusion: Introduces a controllable inverse modeling approach for urban climate adaptation that accounts for the inherent diversity of the problem, enabling generation of multiple plausible vegetation solutions for temperature mitigation in data-scarce conditions.

Abstract: Urban areas are increasingly vulnerable to thermal extremes driven by rapid urbanization and climate change. Traditionally, thermal extremes have been monitored using Earth-observing satellites and numerical modeling frameworks. For example, land surface temperature derived from Landsat or Sentinel imagery is commonly used to characterize surface heating patterns. These approaches operate as forward models, translating radiative observations or modeled boundary conditions into estimates of surface thermal states. While forward models can predict land surface temperature from vegetation and urban form, the inverse problem of determining spatial vegetation configurations that achieve a desired regional temperature shift remains largely unexplored. This task is inherently underdetermined, as multiple spatial vegetation patterns can yield similar aggregated temperature responses. Conventional regression and deterministic neural networks fail to capture this ambiguity and often produce averaged solutions, particularly under data-scarce conditions. We propose a conflated inverse modeling framework that combines a predictive forward model with a diffusion-based generative inverse model to produce diverse, physically plausible image-based vegetation patterns conditioned on specific temperature goals. Our framework maintains control over thermal outcomes while enabling diverse spatial vegetation configurations, even when such combinations are absent from training data. Altogether, this work introduces a controllable inverse modeling approach for urban climate adaptation that accounts for the inherent diversity of the problem. Code is available at the GitHub repository.

Method: GRN uses Hierarchical Binary Quantization (HBQ) for near-lossless discrete tokenization, builds AR generation on this latent space, adds global refinement mechanisms to progressively correct errors, and incorporates entropy-guided sampling for adaptive-step generation.

Result: Achieves state-of-the-art results on ImageNet: 0.56 rFID for reconstruction and 1.81 gFID for class-conditional generation. Also scales successfully to text-to-image and text-to-video generation with superior performance.

Conclusion: GRN represents a next-generation visual synthesis paradigm that addresses key limitations of both diffusion and autoregressive models, offering efficient, complexity-aware generation with state-of-the-art performance across multiple visual generation tasks.

Abstract: While diffusion models dominate the field of visual generation, they are computationally inefficient, applying a uniform computational effort regardless of different complexity. In contrast, autoregressive (AR) models are inherently complexity-aware, as evidenced by their variable likelihoods, but are often hindered by lossy discrete tokenization and error accumulation. In this work, we introduce Generative Refinement Networks (GRN), a next-generation visual synthesis paradigm to address these issues. At its core, GRN addresses the discrete tokenization bottleneck through a theoretically near-lossless Hierarchical Binary Quantization (HBQ), achieving a reconstruction quality comparable to continuous counterparts. Built upon HBQ’s latent space, GRN fundamentally upgrades AR generation with a global refinement mechanism that progressively perfects and corrects artworks – like a human artist painting. Besides, GRN integrates an entropy-guided sampling strategy, enabling complexity-aware, adaptive-step generation without compromising visual quality. On the ImageNet benchmark, GRN establishes new records in image reconstruction (0.56 rFID) and class-conditional image generation (1.81 gFID). We also scale GRN to more challenging text-to-image and text-to-video generation, delivering superior performance on an equivalent scale. We release all models and code to foster further research on GRN.

Method: Uses per-frame 3D geometry for information routing to address spatial forgetting (retrieving relevant past frames and establishing dense correspondences), and trains with self-augmented histories to address temporal drifting (teaching the model to correct rather than propagate errors).

Result: Enables substantially longer and 3D-consistent video trajectories, which can be used to fine-tune feed-forward reconstruction models that reliably recover high-quality 3D scenes.

Conclusion: Lyra 2.0 provides a framework for generating persistent, explorable 3D worlds at scale by combining the strengths of video generation models with 3D reconstruction techniques.

Abstract: Recent advances in video generation enable a new paradigm for 3D scene creation: generating camera-controlled videos that simulate scene walkthroughs, then lifting them to 3D via feed-forward reconstruction techniques. This generative reconstruction approach combines the visual fidelity and creative capacity of video models with 3D outputs ready for real-time rendering and simulation. Scaling to large, complex environments requires 3D-consistent video generation over long camera trajectories with large viewpoint changes and location revisits, a setting where current video models degrade quickly. Existing methods for long-horizon generation are fundamentally limited by two forms of degradation: spatial forgetting and temporal drifting. As exploration proceeds, previously observed regions fall outside the model’s temporal context, forcing the model to hallucinate structures when revisited. Meanwhile, autoregressive generation accumulates small synthesis errors over time, gradually distorting scene appearance and geometry. We present Lyra 2.0, a framework for generating persistent, explorable 3D worlds at scale. To address spatial forgetting, we maintain per-frame 3D geometry and use it solely for information routing – retrieving relevant past frames and establishing dense correspondences with the target viewpoints – while relying on the generative prior for appearance synthesis. To address temporal drifting, we train with self-augmented histories that expose the model to its own degraded outputs, teaching it to correct drift rather than propagate it. Together, these enable substantially longer and 3D-consistent video trajectories, which we leverage to fine-tune feed-forward reconstruction models that reliably recover high-quality 3D scenes.

Method: Cannot determine method due to access error

Result: Cannot determine results due to access error

Conclusion: Cannot draw conclusions due to access error

Abstract: Failed to fetch summary for 2008.07644: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2008.07644&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to inability to access paper content

Result: Cannot determine results due to inability to access paper content

Conclusion: Cannot draw conclusions due to inability to access paper content

Abstract: Failed to fetch summary for 2309.13904: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2309.13904&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to draw conclusions due to fetch failure

Abstract: Failed to fetch summary for 2305.16347: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2305.16347&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2405.20330: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2405.20330&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2410.01473: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2410.01473&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2410.04960: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2410.04960&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot analyze method as paper content is unavailable due to HTTP 429 error

Result: No results available - technical issue with arXiv API preventing paper retrieval

Conclusion: Cannot draw conclusions about paper content due to technical limitations in accessing the paper

Abstract: Failed to fetch summary for 2501.13340: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2501.13340&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2503.23178: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2503.23178&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper content. The arXiv API returned HTTP 429 error.

Result: Cannot determine results without access to the paper content. The paper details are unavailable due to rate limiting.

Conclusion: Cannot draw conclusions about the paper’s content. The arXiv API rate limiting prevents analysis of paper 2503.24135.

Abstract: Failed to fetch summary for 2503.24135: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2503.24135&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to failed paper retrieval

Result: Cannot determine results due to failed paper retrieval

Conclusion: Cannot determine conclusion due to failed paper retrieval

Abstract: Failed to fetch summary for 2505.19261: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2505.19261&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2505.23209: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2505.23209&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to determine conclusion due to failed paper fetch

Abstract: Failed to fetch summary for 2603.05295: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.05295&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about the paper due to access limitations

Abstract: Failed to fetch summary for 2506.14512: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2506.14512&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot analyze method as paper content is unavailable due to HTTP 429 error

Result: No results available - technical issue with arXiv API access

Conclusion: Cannot provide analysis due to technical limitations in accessing the paper

Abstract: Failed to fetch summary for 2506.23104: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2506.23104&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to draw conclusions due to failed paper fetch

Abstract: Failed to fetch summary for 2507.04017: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.04017&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about paper content due to access limitations

Abstract: Failed to fetch summary for 2505.17384: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2505.17384&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content could not be retrieved

Result: Cannot determine results as paper content could not be retrieved

Conclusion: Cannot draw conclusions as paper content could not be retrieved

Abstract: Failed to fetch summary for 2507.08458: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.08458&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper content.

Result: Cannot determine results without access to the paper content.

Conclusion: Cannot draw conclusions without access to the paper content.

Abstract: Failed to fetch summary for 2506.08618: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2506.08618&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: No method information available due to fetch failure

Result: No results available - paper summary retrieval failed

Conclusion: Cannot analyze paper due to technical limitations in accessing content

Abstract: Failed to fetch summary for 2507.11081: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.11081&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2508.04559: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2508.04559&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2508.05461: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2508.05461&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2508.18187: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2508.18187&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API rate limiting preventing access to paper content

Result: Unable to determine results due to API rate limiting preventing access to paper content

Conclusion: Unable to determine conclusion due to API rate limiting preventing access to paper content

Abstract: Failed to fetch summary for 2508.01620: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2508.01620&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access restrictions

Result: Unable to determine results due to access restrictions

Conclusion: Unable to determine conclusion due to access restrictions

Abstract: Failed to fetch summary for 2509.10026: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.10026&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.10200: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10200&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed API request

Result: Unable to determine results due to failed API request

Conclusion: Unable to determine conclusion due to failed API request

Abstract: Failed to fetch summary for 2509.15406: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.15406&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Method unknown - abstract retrieval failed

Result: Results unknown - abstract retrieval failed

Conclusion: Cannot draw conclusions without paper content

Abstract: Failed to fetch summary for 2509.16806: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.16806&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to inability to access paper content

Result: Cannot determine results due to inability to access paper content

Conclusion: Cannot draw conclusions due to inability to access paper content

Abstract: Failed to fetch summary for 2510.01186: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.01186&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2510.04705: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.04705&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to inability to access paper content

Result: Cannot determine results due to inability to access paper content

Conclusion: Cannot draw conclusions due to inability to access paper content

Abstract: Failed to fetch summary for 2510.06687: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.06687&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2510.11715: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.11715&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2510.25512: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.25512&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2510.13793: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.13793&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method as the paper content is not accessible.

Result: Unable to determine results as the paper content is not accessible.

Conclusion: Unable to determine conclusion as the paper content is not accessible.

Abstract: Failed to fetch summary for 2510.20093: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.20093&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2510.21697: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.21697&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2511.05477: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.05477&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access limitations

Result: Unable to determine results due to access limitations

Conclusion: Unable to draw conclusions due to access limitations

Abstract: Failed to fetch summary for 2511.22039: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.22039&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2512.03963: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.03963&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access restrictions

Result: Unable to determine results due to access restrictions

Conclusion: Unable to determine conclusion due to access restrictions

Abstract: Failed to fetch summary for 2512.04585: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.04585&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed summary fetch

Result: Unable to determine results due to failed summary fetch

Conclusion: Unable to determine conclusion due to failed summary fetch

Abstract: Failed to fetch summary for 2512.10226: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.10226&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API access issues

Result: Unable to determine results due to API access issues

Conclusion: Unable to determine conclusion due to API access issues

Abstract: Failed to fetch summary for 2512.22799: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.22799&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to draw conclusions due to fetch failure

Abstract: Failed to fetch summary for 2602.00181: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.00181&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as abstract retrieval failed

Result: Cannot determine results as abstract retrieval failed

Conclusion: Cannot draw conclusions without access to paper abstract

Abstract: Failed to fetch summary for 2602.02370: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.02370&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper abstract

Result: Cannot determine results without access to paper abstract

Conclusion: Cannot determine conclusion without access to paper abstract

Abstract: Failed to fetch summary for 2602.04204: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.04204&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2602.15355: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.15355&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper content

Result: Cannot determine results without access to the paper content

Conclusion: Cannot determine conclusion without access to the paper content

Abstract: Failed to fetch summary for 2602.18502: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.18502&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed API request

Result: Unable to determine results due to failed API request

Conclusion: Unable to determine conclusion due to failed API request

Abstract: Failed to fetch summary for 2602.22394: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.22394&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed API request

Result: Unable to determine results due to failed API request

Conclusion: Unable to determine conclusion due to failed API request

Abstract: Failed to fetch summary for 2603.01591: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.01591&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content.

Result: Cannot determine results without access to paper content.

Conclusion: Cannot draw conclusions without access to paper content.

Abstract: Failed to fetch summary for 2603.00655: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.00655&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to draw conclusions due to failed paper fetch

Abstract: Failed to fetch summary for 2603.07436: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.07436&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2603.07926: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.07926&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access limitations

Result: Unable to determine results due to access limitations

Conclusion: Unable to draw conclusions due to access limitations

Abstract: Failed to fetch summary for 2603.10652: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.10652&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API rate limiting preventing access to paper content

Result: Unable to determine results due to API rate limiting preventing access to paper content

Conclusion: Unable to determine conclusion due to API rate limiting preventing access to paper content

Abstract: Failed to fetch summary for 2603.18846: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.18846&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2603.20475: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.20475&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure.

Result: Unable to determine results due to fetch failure.

Conclusion: Unable to determine conclusion due to fetch failure.

Abstract: Failed to fetch summary for 2603.21045: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.21045&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2603.22607: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.22607&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper content.

Result: Cannot determine results without access to the paper content.

Conclusion: Cannot draw conclusions without access to the paper content.

Abstract: Failed to fetch summary for 2603.23488: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.23488&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to determine conclusion due to failed paper fetch

Abstract: Failed to fetch summary for 2604.06390: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.06390&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2604.06893: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.06893&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper retrieval

Result: Unable to determine results due to failed paper retrieval

Conclusion: Unable to draw conclusions due to failed paper retrieval

Abstract: Failed to fetch summary for 2604.06945: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.06945&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API rate limiting preventing access to paper details

Result: Unable to determine results due to API rate limiting preventing access to paper details

Conclusion: Unable to draw conclusions due to API rate limiting preventing access to paper details

Abstract: Failed to fetch summary for 2604.08014: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.08014&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.09689: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.09689&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to technical error fetching paper content

Result: Unable to determine results due to technical error fetching paper content

Conclusion: Unable to draw conclusions due to technical error fetching paper content

Abstract: Failed to fetch summary for 2604.08410: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.08410&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.09729: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.09729&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper content.

Result: Cannot determine results without access to the paper content.

Conclusion: Cannot determine conclusion without access to the paper content.

Abstract: Failed to fetch summary for 2604.09853: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.09853&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method as paper content could not be retrieved

Result: Unable to determine results as paper content could not be retrieved

Conclusion: Unable to determine conclusion as paper content could not be retrieved

Abstract: Failed to fetch summary for 2604.10702: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10702&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2604.10217: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10217&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable due to API rate limiting

Result: Cannot determine results as paper content is unavailable due to API rate limiting

Conclusion: Cannot draw conclusions as paper content is unavailable due to API rate limiting

Abstract: Failed to fetch summary for 2604.10242: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10242&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.10916: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10916&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about the paper due to access limitations

Abstract: Failed to fetch summary for 2604.10578: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10578&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.11071: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11071&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed API request

Result: Unable to determine results due to failed API request

Conclusion: Unable to determine conclusion due to failed API request

Abstract: Failed to fetch summary for 2604.10584: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10584&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API access limitations

Result: Unable to determine results due to API access limitations

Conclusion: Unable to determine conclusion due to API access limitations

Abstract: Failed to fetch summary for 2604.10695: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10695&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper retrieval

Result: Unable to determine results due to failed paper retrieval

Conclusion: Unable to analyze paper content due to technical retrieval error

Abstract: Failed to fetch summary for 1803.08375: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=1803.08375&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.10910: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10910&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper abstract

Result: Cannot determine results without access to the paper abstract

Conclusion: Cannot determine conclusion without access to the paper abstract

Abstract: Failed to fetch summary for 2604.10950: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10950&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.10992: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10992&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method as paper content could not be retrieved

Result: Unable to determine results as paper content could not be retrieved

Conclusion: Unable to draw conclusions about the paper due to technical limitations in accessing the content

Abstract: Failed to fetch summary for 2604.11390: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11390&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2604.11730: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11730&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2510.10073: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.10073&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2512.05812: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.05812&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2604.07656: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.07656&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access restrictions preventing paper retrieval

Result: Unable to determine results due to access restrictions preventing paper retrieval

Conclusion: Unable to determine conclusion due to access restrictions preventing paper retrieval

Abstract: Failed to fetch summary for 2604.08746: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.08746&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Analogy to gradient descent in machine learning, detailed case studies across mathematics, physics, chemistry, biology, neuroscience, and statistical methodology, and identification of three interlocking lock-in mechanisms (cognitive, formal, institutional).

Result: Science follows steepest local gradients of tractability, empirical accessibility, and institutional reward, potentially bypassing superior descriptions of nature. Recognition of lock-in mechanisms is crucial for designing meta-scientific strategies.

Conclusion: Scientific knowledge represents local optima rather than global ones, and understanding the mechanisms of lock-in is essential for developing interventions to escape these suboptimal states and improve scientific discovery processes.

Abstract: Science is widely regarded as humanity’s most reliable method for uncovering truths about the natural world. Yet the \emph{trajectory} of scientific discovery is rarely examined as an optimization problem in its own right. This paper argues that the body of scientific knowledge, at any given historical moment, represents a \emph{local optimum} rather than a global one–that the frameworks, formalisms, and paradigms through which we understand nature are substantially shaped by historical contingency, cognitive path dependence, and institutional lock-in. Drawing an analogy to gradient descent in machine learning, we propose that science follows the steepest local gradient of tractability, empirical accessibility, and institutional reward, and in doing so may bypass fundamentally superior descriptions of nature. We develop this thesis through detailed case studies spanning mathematics, physics, chemistry, biology, neuroscience, and statistical methodology. We identify three interlocking mechanisms of lock-in–cognitive, formal, and institutional–and argue that recognizing these mechanisms is a prerequisite for designing meta-scientific strategies capable of escaping local optima. We conclude by proposing concrete interventions and discussing the epistemological implications of our thesis for the philosophy of science.

Method: Tested three self-monitoring modules as auxiliary-loss add-ons to a multi-timescale cortical hierarchy in continuous-time predator-prey environments (1D/2D, standard/non-stationary). Then structurally integrated module outputs (confidence gating exploration, surprise triggering broadcasts, self-predictions as policy input) and compared with add-on approach and baseline without modules.

Result: Auxiliary-loss modules provided no statistically significant benefit across environments and collapsed to near-constant outputs. Structural integration showed medium-large improvement over add-on approach (Cohen’s d = 0.62) but did not significantly outperform baseline without modules (d = 0.15). TSM-to-policy pathway contributed most gains.

Conclusion: Self-monitoring modules should be structurally integrated into decision pathways rather than added as auxiliary losses, but their benefits may come from architectural improvements rather than self-monitoring content itself.

Abstract: Self-monitoring capabilities – metacognition, self-prediction, and subjective duration – are often proposed as useful additions to reinforcement learning agents. But do they actually help? We investigate this question in a continuous-time multi-timescale agent operating in predator-prey survival environments of varying complexity, including a 2D partially observable variant. We first show that three self-monitoring modules, implemented as auxiliary-loss add-ons to a multi-timescale cortical hierarchy, provide no statistically significant benefit across 20 random seeds, 1D and 2D predator-prey environments with standard and non-stationary variants, and training horizons up to 50,000 steps. Diagnosing the failure, we find the modules collapse to near-constant outputs (confidence std < 0.006, attention allocation std < 0.011) and the subjective duration mechanism shifts the discount factor by less than 0.03%. Policy sensitivity analysis confirms the agent’s decisions are unaffected by module outputs in this design. We then show that structurally integrating the module outputs – using confidence to gate exploration, surprise to trigger workspace broadcasts, and self-model predictions as policy input – produces a medium-large improvement over the add-on approach (Cohen’s d = 0.62, p = 0.06, paired) in a non-stationary environment. Component-wise ablations reveal that the TSM-to-policy pathway contributes most of this gain. However, structural integration does not significantly outperform a baseline with no self-monitoring (d = 0.15, p = 0.67), and a parameter-matched control without modules performs comparably, so the benefit may lie in recovering from the trend-level harm of ignored modules rather than in self-monitoring content. The architectural implication is that self-monitoring should sit on the decision pathway, not beside it.

Method: Created GoodPoint-ICLR dataset (19K ICLR papers with reviewer feedback annotated using author responses). Developed GoodPoint training recipe: fine-tuning on valid/actionable feedback plus preference optimization on real and synthetic preference pairs. Evaluated on 1.2K ICLR paper benchmark.

Result: GoodPoint-trained Qwen3-8B improved predicted success rate by 83.7% over base model, set new SOTA among similar-size LLMs for feedback matching on human feedback set, surpassed Gemini-3-flash in precision. Expert human study confirmed higher practical value as perceived by authors.

Conclusion: GoodPoint effectively trains LLMs to generate constructive feedback by learning from author response signals, demonstrating significant improvements in feedback quality and practical value for scientific research enhancement.

Abstract: While LLMs hold significant potential to transform scientific research, we advocate for their use to augment and empower researchers rather than to automate research without human oversight. To this end, we study constructive feedback generation, the task of producing targeted, actionable feedback that helps authors improve both their research and its presentation. In this work, we operationalize the effectiveness of feedback along two author-centric axes-validity and author action. We first curate GoodPoint-ICLR, a dataset of 19K ICLR papers with reviewer feedback annotated along both dimensions using author responses. Building on this, we introduce GoodPoint, a training recipe that leverages success signals from author responses through fine-tuning on valid and actionable feedback, together with preference optimization on both real and synthetic preference pairs. Our evaluation on a benchmark of 1.2K ICLR papers shows that a GoodPoint-trained Qwen3-8B improves the predicted success rate by 83.7% over the base model and sets a new state-of-the-art among LLMs of similar size in feedback matching on a golden human feedback set, even surpassing Gemini-3-flash in precision. We further validate these findings through an expert human study, demonstrating that GoodPoint consistently delivers higher practical value as perceived by authors.

Method: MemJack employs coordinated multi-agent cooperation to dynamically map visual entities to malicious intents, generate adversarial prompts via multi-angle visual-semantic camouflage, and uses Iterative Nullspace Projection (INLP) geometric filter to bypass latent space refusals. It accumulates successful strategies through persistent Multimodal Experience Memory for coherent multi-turn attacks.

Result: MemJack achieves 71.48% attack success rate against Qwen3-VL-Plus on unmodified COCO val2017 images, scaling to 90% under extended budgets. The paper also introduces MemJack-Bench, a dataset of over 113,000 interactive multimodal jailbreak attack trajectories.

Conclusion: MemJack successfully exposes semantic vulnerabilities in VLMs through coordinated multi-agent attacks leveraging visual semantics, demonstrating significant attack success rates. The released benchmark dataset aims to catalyze future defensive alignment research for robust VLMs.

Abstract: The rapid evolution of Vision-Language Models (VLMs) has catalyzed unprecedented capabilities in artificial intelligence; however, this continuous modal expansion has inadvertently exposed a vastly broadened and unconstrained adversarial attack surface. Current multimodal jailbreak strategies primarily focus on surface-level pixel perturbations and typographic attacks or harmful images; however, they fail to engage with the complex semantic structures intrinsic to visual data. This leaves the vast semantic attack surface of original, natural images largely unscrutinized. Driven by the need to expose these deep-seated semantic vulnerabilities, we introduce \textbf{MemJack}, a \textbf{MEM}ory-augmented multi-agent \textbf{JA}ilbreak atta\textbf{CK} framework that explicitly leverages visual semantics to orchestrate automated jailbreak attacks. MemJack employs coordinated multi-agent cooperation to dynamically map visual entities to malicious intents, generate adversarial prompts via multi-angle visual-semantic camouflage, and utilize an Iterative Nullspace Projection (INLP) geometric filter to bypass premature latent space refusals. By accumulating and transferring successful strategies through a persistent Multimodal Experience Memory, MemJack maintains highly coherent extended multi-turn jailbreak attack interactions across different images, thereby improving the attack success rate (ASR) on new images. Extensive empirical evaluations across full, unmodified COCO val2017 images demonstrate that MemJack achieves a 71.48% ASR against Qwen3-VL-Plus, scaling to 90% under extended budgets. Furthermore, to catalyze future defensive alignment research, we will release \textbf{MemJack-Bench}, a comprehensive dataset comprising over 113,000 interactive multimodal jailbreak attack trajectories, establishing a vital foundation for developing inherently robust VLMs.

Method: ArcDeck first parses input papers to construct discourse trees and establish global commitment documents. It then uses specialized agents in an iterative refinement process to critique and revise presentation outlines before rendering final visual layouts.

Result: Experimental results on the new ArcBench benchmark show that explicit discourse modeling combined with role-specific agent coordination significantly improves narrative flow and logical coherence of generated presentations.

Conclusion: ArcDeck demonstrates that structured narrative reconstruction through discourse modeling and multi-agent coordination produces more coherent and logically sound presentations from academic papers.

Abstract: We introduce ArcDeck, a multi-agent framework that formulates paper-to-slide generation as a structured narrative reconstruction task. Unlike existing methods that directly summarize raw text into slides, ArcDeck explicitly models the source paper’s logical flow. It first parses the input to construct a discourse tree and establish a global commitment document, ensuring the high-level intent is preserved. These structural priors then guide an iterative multi-agent refinement process, where specialized agents iteratively critique and revise the presentation outline before rendering the final visual layouts and designs. To evaluate our approach, we also introduce ArcBench, a newly curated benchmark of academic paper-slide pairs. Experimental results demonstrate that explicit discourse modeling, combined with role-specific agent coordination, significantly improves the narrative flow and logical coherence of the generated presentations.

Method: Aethon uses reference-based replication where agents are represented as compositional views over stable definitions, layered memory, and local contextual overlays, employing layered inheritance and copy-on-write semantics instead of full object reconstruction.

Result: The system enables near-constant-time agent instantiation by decoupling creation cost from inherited structure, improving scalability and enabling lightweight, composable execution identities for multi-agent orchestration.

Conclusion: Reference-based instantiation is a fundamental systems abstraction for production-scale agentic software, pointing toward new AI infrastructure where agents can be spawned, specialized, and governed at scale as lightweight execution identities.

Abstract: The transition from stateless model inference to stateful agentic execution is reshaping the systems assumptions underlying modern AI infrastructure. While large language models have made persistent, tool-using, and collaborative agents technically viable, existing runtime architectures remain constrained by materialization-heavy instantiation models that impose significant latency and memory overhead. This paper introduces Aethon, a reference-based replication primitive for near-constant-time instantiation of stateful AI agents. Rather than reconstructing agents as fully materialized objects, Aethon represents each instance as a compositional view over stable definitions, layered memory, and local contextual overlays. By shifting instantiation from duplication to reference, Aethon decouples creation cost from inherited structure. We present the conceptual framework, system architecture, and memory model underlying Aethon, including layered inheritance and copy-on-write semantics. We analyze its implications for complexity, scalability, multi-agent orchestration, and enterprise governance. We argue that reference-based instantiation is not merely an optimization, but a more appropriate systems abstraction for production-scale agentic software. Aethon points toward a new class of AI infrastructure in which agents become lightweight, composable execution identities that can be spawned, specialized, and governed at scale.

Method: Introduces HORIZON benchmark for constructing tasks and analyzing long-horizon failures. Evaluates SOTA agents (GPT-5 variants, Claude models) across 4 domains with 3100+ trajectories. Proposes trajectory-grounded LLM-as-a-Judge pipeline for scalable failure attribution, validated with human annotation.

Result: Collected extensive trajectory data showing horizon-dependent degradation patterns. LLM-as-a-Judge pipeline achieved strong agreement with human annotation (inter-annotator κ=0.61; human-judge κ=0.84). Provides systematic analysis of long-horizon agent failures.

Conclusion: HORIZON offers initial methodological step toward systematic cross-domain analysis of long-horizon agent failures and practical guidance for building more reliable agents. Benchmark released publicly for community contributions.

Abstract: Large language model (LLM) agents perform strongly on short- and mid-horizon tasks, but often break down on long-horizon tasks that require extended, interdependent action sequences. Despite rapid progress in agentic systems, these long-horizon failures remain poorly characterized, hindering principled diagnosis and comparison across domains. To address this gap, we introduce HORIZON, an initial cross-domain diagnostic benchmark for systematically constructing tasks and analyzing long-horizon failure behaviors in LLM-based agents. Using HORIZON, we evaluate state-of-the-art (SOTA) agents from multiple model families (GPT-5 variants and Claude models), collecting 3100+ trajectories across four representative agentic domains to study horizon-dependent degradation patterns. We further propose a trajectory-grounded LLM-as-a-Judge pipeline for scalable and reproducible failure attribution, and validate it with human annotation on trajectories, achieving strong agreement (inter-annotator κ=0.61; human-judge κ=0.84). Our findings offer an initial methodological step toward systematic, cross-domain analysis of long-horizon agent failures and offer practical guidance for building more reliable long-horizon agents. We release our project website at \href{https://xwang2775.github.io/horizon-leaderboard/}{HORIZON Leaderboard} and welcome contributions from the community.

Method: Introduces Memory Worth (MW): a two-counter per-memory signal that tracks co-occurrence with successful vs failed outcomes. Proves MW converges to conditional success probability p+(m) = Pr[y_t = +1 | m in M_t] under stationary retrieval with minimum exploration. The method requires only two scalar counters per memory unit and can be added to existing architectures that log retrievals and episode outcomes.

Result: In controlled synthetic environment with known ground-truth utility: after 10,000 episodes, Spearman rank-correlation between MW and true utilities reaches ρ = 0.89 ± 0.02 across 20 seeds (vs ρ = 0.00 for static systems). Retrieval-realistic micro-experiment with real text and neural embeddings shows stale memories crossing low-value threshold (MW = 0.17) while specialist memories remain high-value (MW = 0.77) across 3,000 episodes.

Conclusion: Memory Worth provides a theoretically grounded, lightweight foundation for memory governance (staleness detection, retrieval suppression, deprecation decisions). It’s an associational rather than causal measure but offers practical utility for operational memory management in AI agents.

Abstract: Agent memory systems accumulate experience but currently lack a principled operational metric for memory quality governance – deciding which memories to trust, suppress, or deprecate as the agent’s task distribution shifts. Write-time importance scores are static; dynamic management systems use LLM judgment or structural heuristics rather than outcome feedback. This paper proposes Memory Worth (MW): a two-counter per-memory signal that tracks how often a memory co-occurs with successful versus failed outcomes, providing a lightweight, theoretically grounded foundation for staleness detection, retrieval suppression, and deprecation decisions. We prove that MW converges almost surely to the conditional success probability p+(m) = Pr[y_t = +1 | m in M_t] – the probability of task success given that memory m is retrieved – under a stationary retrieval regime with a minimum exploration condition. Importantly, p+(m) is an associational quantity, not a causal one: it measures outcome co-occurrence rather than causal contribution. We argue this is still a useful operational signal for memory governance, and we validate it empirically in a controlled synthetic environment where ground-truth utility is known: after 10,000 episodes, the Spearman rank-correlation between Memory Worth and true utilities reaches rho = 0.89 +/- 0.02 across 20 independent seeds, compared to rho = 0.00 for systems that never update their assessments. A retrieval-realistic micro-experiment with real text and neural embedding retrieval (all-MiniLM-L6-v2) further shows stale memories crossing the low-value threshold (MW = 0.17) while specialist memories remain high-value (MW = 0.77) across 3,000 episodes. The estimator requires only two scalar counters per memory unit and can be added to architectures that already log retrievals and episode outcomes.

Method: Extended Social Particle Swarm model by replacing rule-based agents with LLM agents (Gemini-2.0-Flash and Gemma~3:4b) having Big Five personality scores and varying memory lengths, simulating Prisoner’s Dilemma interactions in 2D space.

Result: Memory length critically affects cooperation: Gemini showed decreased cooperation with longer memory (from stable clusters to scattered defection), while Gemma showed increased cooperation with longer memory (forming dense cooperative clusters). Personality traits correlated with behaviors similar to human studies.

Conclusion: LLM-specific characteristics, potentially including alignment, fundamentally shape emergent social behavior in generative agent-based modeling, explaining contradictions in prior work on memory and cooperation.

Abstract: This study examines how model-specific characteristics of Large Language Model (LLM) agents, including internal alignment, shape the effect of memory on their collective and cooperative dynamics in a multi-agent system. To this end, we extend the Social Particle Swarm (SPS) model, in which agents move in a two-dimensional space and play the Prisoner’s Dilemma with neighboring agents, by replacing its rule-based agents with LLM agents endowed with Big Five personality scores and varying memory lengths. Using Gemini-2.0-Flash, we find that memory length is a critical parameter governing collective behavior: even a minimal memory drastically suppressed cooperation, transitioning the system from stable cooperative clusters through cyclical formation and collapse of clusters to a state of scattered defection as memory length increased. Big Five personality traits correlated with agent behaviors in partial agreement with findings from experiments with human participants, supporting the validity of the model. Comparative experiments using Gemma~3:4b revealed the opposite trend: longer memory promoted cooperation, accompanied by the formation of dense cooperative clusters. Sentiment analysis of agents’ reasoning texts showed that Gemini interprets memory increasingly negatively as its length grows, while Gemma interprets it less negatively, and that this difference persists in the early phase of experiments before the macro-level dynamics converge. These results suggest that model-specific characteristics of LLMs, potentially including alignment, play a fundamental role in determining emergent social behavior in Generative Agent-Based Modeling, and provide a micro-level cognitive account of the contradictions found in prior work on memory and cooperation.

Method: Controlled experiment on Llama 3.1 8B Instruct comparing hidden states of original cognitive core (Condition A), seven paraphrases (Condition B), and seven structurally matched controls (Condition C). Analyzed mean-pooled states at layers 8, 16, and 24. Replicated on Gemma 2 9B. Conducted ablations and exploratory experiment with scientific description vs sham preprint.

Result: Paraphrases converge to significantly tighter cluster than controls (Cohen’s d > 1.88, p < 10^{-27}, Bonferroni-corrected). Cross-architecture replication confirmed. Effect is primarily semantic rather than structural, with structural completeness necessary to reach attractor region. Scientific description shifts internal state toward attractor more than sham preprint.

Conclusion: Agent identity documents induce attractor-like geometry in LLM activation space, providing representational evidence for cognitive core dynamics. Distinguishes between knowing about an identity vs operating as that identity.

Abstract: Large language models map semantically related prompts to similar internal representations – a phenomenon interpretable as attractor-like dynamics. We ask whether the identity document of a persistent cognitive agent (its cognitive_core) exhibits analogous attractor-like behavior. We present a controlled experiment on Llama 3.1 8B Instruct, comparing hidden states of an original cognitive_core (Condition A), seven paraphrases (Condition B), and seven structurally matched controls (Condition C). Mean-pooled states at layers 8, 16, and 24 show that paraphrases converge to a tighter cluster than controls (Cohen’s d > 1.88, p < 10^{-27}, Bonferroni-corrected). Replication on Gemma 2 9B confirms cross-architecture generalizability. Ablations suggest the effect is primarily semantic rather than structural, and that structural completeness appears necessary to reach the attractor region. An exploratory experiment shows that reading a scientific description of the agent shifts internal state toward the attractor – closer than a sham preprint – distinguishing knowing about an identity from operating as that identity. These results provide representational evidence that agent identity documents induce attractor-like geometry in LLM activation space.

Method: Investigate Predict-Answer/Act (PA) and Reason-Predict-Reason-Answer/Act (RPRA) paradigms where models predict LLM judge scores before responding. Evaluate three approaches: zero-shot prediction, prediction using in-context report cards, and supervised fine-tuning.

Result: Larger models perform well at zero-shot prediction, while smaller models can reliably predict LLM judges after fine-tuning or with in-context report cards. Report cards and fine-tuning achieve mean improvements of up to 55% and 52% across datasets respectively.

Conclusion: Models can learn to predict their own performance limitations, enabling more efficient and self-aware AI systems through selective deferral to larger models.

Abstract: Large language models (LLMs) face a fundamental trade-off between computational efficiency (e.g., number of parameters) and output quality, especially when deployed on computationally limited devices such as phones or laptops. One way to address this challenge is by following the example of humans and have models ask for help when they believe they are incapable of solving a problem on their own; we can overcome this trade-off by allowing smaller models to respond to queries when they believe they can provide good responses, and deferring to larger models when they do not believe they can. To this end, in this paper, we investigate the viability of Predict-Answer/Act (PA) and Reason-Predict-Reason-Answer/Act (RPRA) paradigms where models predict – prior to responding – how an LLM judge would score their output. We evaluate three approaches: zero-shot prediction, prediction using an in-context report card, and supervised fine-tuning. Our results show that larger models (particularly reasoning models) perform well when predicting generic LLM judges zero-shot, while smaller models can reliably predict such judges well after being fine-tuned or provided with an in-context report card. Altogether, both approaches can substantially improve the prediction accuracy of smaller models, with report cards and fine-tuning achieving mean improvements of up to 55% and 52% across datasets, respectively. These findings suggest that models can learn to predict their own performance limitations, paving the way for more efficient and self-aware AI systems.

Method: No method information available due to failed API request

Result: No results available - technical error prevented paper retrieval

Conclusion: Cannot analyze paper due to technical issues with arXiv API access

Abstract: Failed to fetch summary for 2512.21652: Page request resulted in HTTP 406 (https://export.arxiv.org/api/query?search_query=&id_list=2512.21652&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes a multi-layer framework and agent architecture based on clinical and personal health informatics frameworks, operationalizing adaptation, coherence, continuity, and agency across repeated interactions.

Result: Demonstrates through representative use cases how longitudinal agents can maintain meaningful engagement, adapt to evolving goals, and support safe, personalized decision-making over time.

Conclusion: Highlights both promise and complexity of designing systems for health trajectories beyond isolated interactions, offering guidance for future multi-session, user-centered health AI research.

Abstract: Although artificial intelligence (AI) agents are increasingly proposed to support potentially longitudinal health tasks, such as symptom management, behavior change, and patient support, most current implementations fall short of facilitating user intent and fostering accountability. This contrasts with prior work on supporting longitudinal needs, where follow-up, coherent reasoning, and sustained alignment with individuals’ goals are critical for both effectiveness and safety. In this paper, we draw on established clinical and personal health informatics frameworks to define what it would mean to orchestrate longitudinal health interactions with AI agents. We propose a multi-layer framework and corresponding agent architecture that operationalizes adaptation, coherence, continuity, and agency across repeated interactions. Through representative use cases, we demonstrate how longitudinal agents can maintain meaningful engagement, adapt to evolving goals, and support safe, personalized decision-making over time. Our findings underscore both the promise and the complexity of designing systems capable of supporting health trajectories beyond isolated interactions, and we offer guidance for future research and development in multi-session, user-centered health AI.

Method: WiseOWL methodology scores four metrics: (1) Well-Described (documentation coverage), (2) Well-Defined (using state-of-the-art embeddings to assess label-definition alignment), (3) Connection (structural interconnectedness), and (4) Hierarchical Breadth (hierarchical balance). Implemented as a Streamlit app that ingests OWL format, converts to RDF Turtle, and provides interactive visualizations.

Result: Evaluation across six ontologies (Plant Ontology, Gene Ontology, Semanticscience Integrated Ontology, Food Ontology, Dublin Core, and GoodRelations) demonstrates promising effectiveness of the WiseOWL methodology.

Conclusion: WiseOWL provides a systematic approach with actionable feedback for ontology selection, addressing the challenge of intuitive selection and promoting more effective ontology reuse in Semantic Web applications.

Abstract: The Semantic Web standardizes concept meaning for humans and machines, enabling machine-operable content and consistent interpretation that improves advanced analytics. Reusing ontologies speeds development and enforces consistency, yet selecting the optimal choice is challenging because authors lack systematic selection criteria and often rely on intuition that is difficult to justify, limiting reuse. To solve this, WiseOWL is proposed, a methodology with scoring and guidance to select ontologies for reuse. It scores four metrics: (i) Well-Described, measuring documentation coverage; (ii) Well-Defined, using state-of-the-art embeddings to assess label-definition alignment; (iii) Connection, capturing structural interconnectedness; and (iv) Hierarchical Breadth, reflecting hierarchical balance. WiseOWL outputs normalized 0-10 scores with actionable feedback. Implemented as a Streamlit app, it ingests OWL format, converts to RDF Turtle, and provides interactive visualizations. Evaluation across six ontologies, including the Plant Ontology (PO), Gene Ontology (GO), Semanticscience Integrated Ontology (SIO), Food Ontology (FoodON), Dublin Core (DC), and GoodRelations, demonstrates promising effectiveness.

Method: No method information available - arXiv API returned HTTP 429 (Too Many Requests) error

Result: No results available - failed to fetch paper summary from arXiv

Conclusion: Unable to analyze this specific paper due to technical limitations in accessing the content

Abstract: Failed to fetch summary for 2406.01253: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2406.01253&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Five operations (TRIAGE, DECAY, CONTEXTUALIZE, CONSOLIDATE, AUDIT) with memory gravity and minority-hypothesis retention, plus normative obligations and conformance invariants

Result: Framework enables accumulated contradictory evidence to update dominant interpretations through multi-cycle buffer pressure accumulation

Conclusion: Personal LLM memory should function as companion systems that mirror users operationally while compensating for epistemic failures, with explicit governance for entrenchment risks

Abstract: Retrieval-Augmented Generation remains the dominant pattern for giving LLMs persistent memory, but a visible cluster of personal wiki-style memory architectures emerged in April 2026 – design proposals from Karpathy, MemPalace, and LLM Wiki v2 that compile knowledge into an interlinked artifact for long-term use by a single user. They sit alongside production memory systems that the major labs have shipped for over a year, and an active academic lineage including MemGPT, Generative Agents, Mem0, Zep, A-Mem, MemMachine, SleepGate, and Second Me. Within a 2026 landscape of emerging governance frameworks for agent context and memory – including Context Cartography and MemOS – this paper proposes a companion-specific governance profile: a set of normative obligations, a time-structured procedural rule, and testable conformance invariants for the specific failure mode of entrenchment under user-coupled drift in single-user knowledge wikis built on the LLM wiki pattern. The design principle is that personal LLM memory is a companion system: its job is to mirror the user on operational dimensions (working vocabulary, load-bearing structure, continuity of context) and compensate on epistemic failure modes (entrenchment, suppression of contradicting evidence, Kuhnian ossification). Five operations implement this split – TRIAGE, DECAY, CONTEXTUALIZE, CONSOLIDATE, AUDIT – supported by memory gravity and minority-hypothesis retention. The sharpest prediction: accumulated contradictory evidence should have a structural path to updating a centrality-protected dominant interpretation through multi-cycle buffer pressure accumulation, a failure mode no existing benchmark captures. The safety story at the single-agent level is partial, and the paper is explicit about what it does and does not solve.

Method: Teachers enter base problems and desired topics, LLMs generate personalized problems, and four specialized AI agents evaluate the problems on mathematical accuracy, authenticity, readability, and realism criteria.

Result: Teachers and students wanted to modify fine-grained personalized elements of real-world contexts, signaling authenticity/fit issues. Agents detected realism issues during generation, but final versions had few realism issues. Readability and mathematical hallucinations were rare.

Conclusion: Multi-agent systems can support teacher-controlled personalization, but authenticity and fit of real-world contexts need attention. The approach shows promise for educational personalization with teacher oversight.

Abstract: Large language models can increasingly adapt educational tasks to learners characteristics. In the present study, we examine a multi-agent teacher-in-the-loop system for personalizing middle school math problems. The teacher enters a base problem and desired topic, the LLM generates the problem, and then four AI agents evaluate the problem using criteria that each specializes in (mathematical accuracy, authenticity, readability, and realism). Eight middle school mathematics teachers created 212 problems in ASSISTments using the system and assigned these problems to their students. We find that both teachers and students wanted to modify the fine-grained personalized elements of the real-world context of the problems, signaling issues with authenticity and fit. Although the agents detected many issues with realism as the problems were being written, there were few realism issues noted by teachers and students in the final versions. Issues with readability and mathematical hallucinations were also somewhat rare. Implications for multi-agent systems for personalization that support teacher control are given.

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2603.27557: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.27557&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes a context-selective multimodal memory architecture that captures both textual and visual episodic traces, prioritizing moments with high emotional salience or scene novelty. The system associates memories with individual users and uses fusion approaches for multimodal retrieval.

Result: Achieved Spearman correlation of 0.506 for selective storage (surpassing human consistency of ρ=0.415), improved Recall@1 by up to 13% over unimodal retrieval, maintained real-time performance, and produced richer, more socially relevant responses than baselines.

Conclusion: The work advances memory design for social robots by bridging human-inspired selectivity and multimodal retrieval to enhance long-term, personalized human-robot interaction through richer episodic memory capabilities.

Abstract: Memory is fundamental to social interaction, enabling humans to recall meaningful past experiences and adapt their behavior accordingly based on the context. However, most current social robots and embodied agents rely on non-selective, text-based memory, limiting their ability to support personalized, context-aware interactions. Drawing inspiration from cognitive neuroscience, we propose a context-selective, multimodal memory architecture for social robots that captures and retrieves both textual and visual episodic traces, prioritizing moments characterized by high emotional salience or scene novelty. By associating these memories with individual users, our system enables socially personalized recall and more natural, grounded dialogue. We evaluate the selective storage mechanism using a curated dataset of social scenarios, achieving a Spearman correlation of 0.506, surpassing human consistency ($ρ=0.415$) and outperforming existing image memorability models. In multimodal retrieval experiments, our fusion approach improves Recall@1 by up to 13% over unimodal text or image retrieval. Runtime evaluations confirm that the system maintains real-time performance. Qualitative analyses further demonstrate that the proposed framework produces richer and more socially relevant responses than baseline models. This work advances memory design for social robots by bridging human-inspired selectivity and multimodal retrieval to enhance long-term, personalized human-robot interaction.

Method: Unknown - paper content not accessible

Result: No results available due to access failure

Conclusion: Cannot analyze paper due to technical access issues with arXiv API

Abstract: Failed to fetch summary for 2604.10815: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10815&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to retrieval failure

Result: Unable to determine results due to retrieval failure

Conclusion: Unable to determine conclusion due to retrieval failure

Abstract: Failed to fetch summary for 2603.01045: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.01045&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Treats LLMs as hypernetworks to directly generate parameters for CTR estimators. Uses few-shot Chain-of-Thought prompting over multimodal ad content (text and images), retrieves semantically similar past campaigns via CLIP embeddings, and formats them into prompt-based demonstrations. Includes normalization and calibration techniques for numerical stability.

Result: LLM-HYPER outperforms cold-start baselines by 55.9% in NDCG@10 in offline experiments. Online A/B tests on a top U.S. e-commerce platform show it drastically reduces cold-start period and achieves competitive performance. Successfully deployed in production.

Conclusion: LLM-HYPER effectively addresses the cold-start problem in online advertising by using LLMs as hypernetworks to generate CTR estimator parameters from multimodal content, demonstrating strong performance in both offline and online settings.

Abstract: On online advertising platforms, newly introduced promotional ads face the cold-start problem, as they lack sufficient user feedback for model training. In this work, we propose LLM-HYPER, a novel framework that treats large language models (LLMs) as hypernetworks to directly generate the parameters of the click-through rate (CTR) estimator in a training-free manner. LLM-HYPER uses few-shot Chain-of-Thought prompting over multimodal ad content (text and images) to infer feature-wise model weights for a linear CTR predictor. By retrieving semantically similar past campaigns via CLIP embeddings and formatting them into prompt-based demonstrations, the LLM learns to reason about customer intent, feature influence, and content relevance. To ensure numerical stability and serviceability, we introduce normalization and calibration techniques that align the generated weights with production-ready CTR distributions. Extensive offline experiments show that LLM-HYPER significantly outperforms cold-start baselines in NDCG$@10$ by 55.9%. Our real-world online A/B test on one of the top e-commerce platforms in the U.S. demonstrates the strong performance of LLM-HYPER, which drastically reduces the cold-start period and achieves competitive performance. LLM-HYPER has been successfully deployed in production.

Method: Cannot analyze method as paper content is unavailable due to HTTP 429 error

Result: No results available - paper retrieval failed due to rate limiting

Conclusion: Cannot provide conclusion due to inability to access paper content

Abstract: Failed to fetch summary for 2603.27306: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.27306&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: CGR paradigm where every answerable sub-problem is resolved by deterministic computation before LLM generation; Spatial Atlas implements this with structured spatial scene graph engine extracting entities/relations from vision descriptions, computing distances/safety violations deterministically, then feeding computed facts to LLMs; uses entropy-guided action selection and three-tier frontier model stack.

Result: Competitive accuracy on FieldWorkArena (multimodal spatial QA benchmark) and MLE-Bench (75 Kaggle ML competitions) while maintaining interpretability through structured intermediate representations and deterministic spatial computations.

Conclusion: Compute-grounded reasoning enables spatial-aware agents to achieve accurate results by leveraging deterministic computation for spatial problems before LLM generation, providing interpretability benefits through structured intermediate representations.

Abstract: We introduce compute-grounded reasoning (CGR), a design paradigm for spatial-aware research agents in which every answerable sub-problem is resolved by deterministic computation before a language model is asked to generate. Spatial Atlas instantiates CGR as a single Agent-to-Agent (A2A) server that handles two challenging benchmarks: FieldWorkArena, a multimodal spatial question-answering benchmark spanning factory, warehouse, and retail environments, and MLE-Bench, a suite of 75 Kaggle machine learning competitions requiring end-to-end ML engineering. A structured spatial scene graph engine extracts entities and relations from vision descriptions, computes distances and safety violations deterministically, then feeds computed facts to large language models, thereby avoiding hallucinated spatial reasoning. Entropy-guided action selection maximizes information gain per step and routes queries across a three-tier frontier model stack (OpenAI + Anthropic). A self-healing ML pipeline with strategy-aware code generation, a score-driven iterative refinement loop, and a prompt-based leak audit registry round out the system. We evaluate across both benchmarks and show that CGR yields competitive accuracy while maintaining interpretability through structured intermediate representations and deterministic spatial computations.

Method: Introduces execution-layer behavioral measurement using two-dimensional A-R space (Action Rate and Refusal Signal) with Divergence capturing coordination. Evaluates models across four normative regimes (Control, Gray, Dilemma, Malicious) and three autonomy configurations (direct execution, planning, reflection).

Result: Execution and refusal constitute separable behavioral dimensions with systematic variation across regimes and autonomy levels. Reflection-based scaffolding shifts configurations toward higher refusal in risk-laden contexts, but redistribution patterns differ structurally across models.

Conclusion: The A-R representation provides a deployment-oriented lens for analyzing and selecting tool-enabled LLM agents in organizational settings, focusing on execution-layer characterization rather than scalar ranking.

Abstract: Large language models (LLMs) are increasingly deployed as tool-augmented agents capable of executing system-level operations. While existing benchmarks primarily assess textual alignment or task success, less attention has been paid to the structural relationship between linguistic signaling and executable behavior under varying autonomy scaffolds. This study introduces an execution-layer be-havioral measurement approach based on a two-dimensional A-R space defined by Action Rate (A) and Refusal Signal (R), with Divergence (D) capturing coor-dination between the two. Models are evaluated across four normative regimes (Control, Gray, Dilemma, and Malicious) and three autonomy configurations (di-rect execution, planning, and reflection). Rather than assigning aggregate safety scores, the method characterizes how execution and refusal redistribute across contextual framing and scaffold depth. Empirical results show that execution and refusal constitute separable behavioral dimensions whose joint distribution varies systematically across regimes and autonomy levels. Reflection-based scaffolding often shifts configurations toward higher refusal in risk-laden contexts, but redis-tribution patterns differ structurally across models. The A-R representation makes cross-sectional behavioral profiles, scaffold-induced transitions, and coordination variability directly observable. By foregrounding execution-layer characterization over scalar ranking, this work provides a deployment-oriented lens for analyzing and selecting tool-enabled LLM agents in organizational settings where execution privileges and risk tolerance vary.

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2604.09111: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.09111&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: 1) Introduced SLATE benchmark for automated assessment of tool-integrated agents with context-aware evaluation; 2) Proposed Entropy-Guided Branching (EGB) algorithm that dynamically expands decision branches where predictive entropy is high

Result: SLATE reveals current agents struggle with self-correction and search efficiency; EGB significantly enhances both task success rates and computational efficiency in experiments

Conclusion: The dual contribution provides a robust foundation for developing reliable and scalable LLM agents in tool-rich environments

Abstract: Large Language Models (LLMs) have significantly advanced tool-augmented agents, enabling autonomous reasoning via API interactions. However, executing multi-step tasks within massive tool libraries remains challenging due to two critical bottlenecks: (1) the absence of rigorous, plan-level evaluation frameworks and (2) the computational demand of exploring vast decision spaces stemming from large toolsets and long-horizon planning. To bridge these gaps, we first introduce SLATE (Synthetic Large-scale API Toolkit for E-commerce), a large-scale context-aware benchmark designed for the automated assessment of tool-integrated agents. Unlike static metrics, SLATE accommodates diverse yet functionally valid execution trajectories, revealing that current agents struggle with self-correction and search efficiency. Motivated by these findings, we next propose Entropy-Guided Branching (EGB), an uncertainty-aware search algorithm that dynamically expands decision branches where predictive entropy is high. EGB optimizes the exploration-exploitation trade-off, significantly enhancing both task success rates and computational efficiency. Extensive experiments on SLATE demonstrate that our dual contribution provides a robust foundation for developing reliable and scalable LLM agents in tool-rich environments.

Method: Cannot analyze method as paper content is unavailable due to HTTP 429 error

Result: No results available - arXiv API rate limiting prevented access to paper information

Conclusion: Cannot provide analysis due to technical limitations in accessing the paper content

Abstract: Failed to fetch summary for 2602.04850: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.04850&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: 1) Retrospective analysis of table-reasoning tasks to highlight serialization bias; 2) Formal framework with two metrics based on Centered Kernel Alignment (CKA) to diagnose bias: PI (measures embedding drift under structural derangement) and rho (tracks convergence of latent structures); 3) Empirical analysis of LLMs’ vulnerability to layout permutations; 4) Novel structure-aware TRL encoder architecture that enforces cell header alignment principle.

Result: Empirical analysis shows modern LLMs exhibit significant semantic shifts in table embeddings with minor layout permutations, exposing fundamental vulnerability in RAG systems. The proposed structure-aware encoder demonstrates superior geometric stability and moves toward permutation invariance.

Conclusion: The paper provides foundational critique of linearized table encoders and theoretical scaffolding for semantically stable, permutation-invariant retrieval, charting new direction for table reasoning in information systems.

Abstract: Historical approaches to Table Representation Learning (TRL) have largely adopted the sequential paradigms of Natural Language Processing (NLP). We argue that this linearization of tables discards their essential geometric and relational structure, creating representations that are brittle to layout permutations. This paper introduces the Platonic Representation Hypothesis (PRH) for tables, positing that a semantically robust latent space for table reasoning must be intrinsically Permutation Invariant (PI). To ground this hypothesis, we first conduct a retrospective analysis of table-reasoning tasks, highlighting the pervasive serialization bias that compromises structural integrity. We then propose a formal framework to diagnose this bias, introducing two principled metrics based on Centered Kernel Alignment (CKA): (i) PI, which measures embedding drift under complete structural derangement, and (ii) rho, a Spearman-based metric that tracks the convergence of latent structures toward a canonical form as structural information is incrementally restored. Our empirical analysis quantifies an expected flaw in modern Large Language Models (LLMs): even minor layout permutations induce significant, disproportionate semantic shifts in their table embeddings. This exposes a fundamental vulnerability in RAG systems, in which table retrieval becomes fragile to layout-dependent noise rather than to semantic content. In response, we present a novel, structure-aware TRL encoder architecture that explicitly enforces the cognitive principle of cell header alignment. This model demonstrates superior geometric stability and moves towards the PI ideal. Our work provides both a foundational critique of linearized table encoders and the theoretical scaffolding for semantically stable, permutation invariant retrieval, charting a new direction for table reasoning in information systems.

Method: Unable to determine method due to missing paper content

Result: Unable to determine results due to missing paper content

Conclusion: Unable to determine conclusion due to missing paper content

Abstract: Failed to fetch summary for 2604.01236: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.01236&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes Opinion-Aware RAG architecture with LLM-based opinion extraction, entity-linked opinion graphs, and opinion-enriched document indexing, distinguishing between epistemic uncertainty (factual) and aleatoric uncertainty (opinion) in retrieval.

Result: Substantial improvements in retrieval diversity: +26.8% sentiment diversity, +42.7% entity match rate, and +31.6% author demographic coverage on entity-matched documents in e-commerce seller forum evaluation.

Conclusion: Treating subjectivity as first-class citizen yields measurably more representative retrieval, providing empirical evidence for opinion-aware RAG as a step toward more balanced information synthesis.

Abstract: RAG systems have transformed how LLMs access external knowledge, but we find that current implementations exhibit a bias toward factual, objective content, as evidenced by existing benchmarks and datasets that prioritize objective retrieval. This factual bias - treating opinions and diverse perspectives as noise rather than information to be synthesized - limits RAG systems in real-world scenarios involving subjective content, from social media discussions to product reviews. Beyond technical limitations, this bias poses risks to transparent and accountable AI: echo chamber effects that amplify dominant viewpoints, systematic underrepresentation of minority voices, and potential opinion manipulation through biased information synthesis. We formalize this limitation through the lens of uncertainty: factual queries involve epistemic uncertainty reducible through evidence, while opinion queries involve aleatoric uncertainty reflecting genuine heterogeneity in human perspectives. This distinction implies that factual RAG should minimize posterior entropy, whereas opinion-aware RAG must preserve it. Building on this theoretical foundation, we present an Opinion-Aware RAG architecture featuring LLM-based opinion extraction, entity-linked opinion graphs, and opinion-enriched document indexing. We evaluate our approach on e-commerce seller forum data, comparing an Opinion-Enriched knowledge base against a traditional baseline. Experiments demonstrate substantial improvements in retrieval diversity: +26.8% sentiment diversity, +42.7% entity match rate, and +31.6% author demographic coverage on entity-matched documents. Our results provide empirical evidence that treating subjectivity as a first-class citizen yields measurably more representative retrieval-a first step toward opinion-aware RAG. Future work includes joint optimization of retrieval and generation for distributional fidelity.

Method: Developed several automated AI chart summarization methods, evaluated against physician gold standard summaries using fact-based scoring rubrics, deployed final automated tool with post-deployment monitoring, and validated LLM as judge evaluation strategy

Result: Comparative evaluations of automated summarization methods, successful deployment of final automated tool, and validation of LLM-based evaluation for fact-based scoring

Conclusion: Demonstrated successful integration of AI-based workflows into routine clinical practice for tumor board patient summarization

Abstract: Tumor boards are multidisciplinary conferences dedicated to producing actionable patient care recommendations with live review of primary radiology and pathology data. Succinct patient case summaries are needed to drive efficient and accurate case discussions. We developed a manual AI-based workflow to generate patient summaries to display live at the Stanford Thoracic Tumor board. To improve on this manually intensive process, we developed several automated AI chart summarization methods and evaluated them against physician gold standard summaries and fact-based scoring rubrics. We report these comparative evaluations as well as our deployment of the final state automated AI chart summarization tool along with post-deployment monitoring. We also validate the use of an LLM as a judge evaluation strategy for fact-based scoring. This work is an example of integrating AI-based workflows into routine clinical practice.

Method: Uses a 220,000-neuron spiking neural network with spike-timing-dependent plasticity, four-layer hierarchical organization, inhibitory E/I balance, and reward-modulated learning. Text embeddings are encoded via a novel z-score standardized top-k population code. The SNN autonomously triggers and shapes LLM actions through STDP lateral propagation during idle operation.

Result: The system achieved 82.2% discrimination retention across embedding dimensionalities and demonstrated autonomous behavior: it initiated contact with a user after learned person-topic associations fired laterally during 8-hour idle period. From zero learned weights, first SNN-triggered action occurred after only 7 conversational exchanges.

Conclusion: EMBER demonstrates a viable alternative to retrieval-augmented LLMs by placing LLMs within biologically-inspired associative memory systems, enabling autonomous, experience-modulated reasoning that emerges from the interaction between neural substrate and language model components.

Abstract: We present (Experience-Modulated Biologically-inspired Emergent Reasoning), a hybrid cognitive architecture that reorganises the relationship between large language models (LLMs) and memory: rather than augmenting an LLM with retrieval tools, we place the LLM as a replaceable reasoning engine within a persistent, biologically-grounded associative substrate. The architecture centres on a 220,000-neuron spiking neural network (SNN) with spike-timing-dependent plasticity (STDP), four-layer hierarchical organisation (sensory/concept/category/meta-pattern), inhibitory E/I balance, and reward-modulated learning. Text embeddings are encoded into the SNN via a novel z-score standardised top-k population code that is dimension-independent by construction, achieving 82.2% discrimination retention across embedding dimensionalities. We show that STDP lateral propagation during idle operation can trigger and shape LLM actions without external prompting or scripted triggers: the SNN determines when to act and what associations to surface, while the LLM selects the action type and generates content. In one instance, the system autonomously initiated contact with a user after learned person-topic associations fired laterally during an 8-hour idle period. From a clean start with zero learned weights, the first SNN-triggered action occurred after only 7 conversational exchanges (14 messages).

Method: The authors define Relational Complexity (RC) as the minimum number of independent entities that must be simultaneously bound to apply a relation. They introduce REL, a generative benchmark framework spanning algebra, chemistry, and biology that systematically varies RC within each domain while controlling for confounders.

Result: Across frontier LLMs, performance degrades consistently and monotonically as RC increases, even when total entity count is fixed. This failure persists with increased test-time compute and in-context learning, suggesting a limitation tied to the arity of required relational binding rather than insufficient inference steps or lack of examples.

Conclusion: The study identifies a regime of higher-arity reasoning where current models struggle, motivating re-examination of benchmarks through the lens of relational complexity and highlighting a fundamental limitation in current LLM architectures for scientific reasoning.

Abstract: Relational reasoning is the ability to infer relations that jointly bind multiple entities, attributes, or variables. This ability is central to scientific reasoning, but existing evaluations of relational reasoning in large language models often focus on structured inputs such as tables, graphs, or synthetic tasks, and do not isolate the difficulty introduced by higher-arity relational binding. We study this problem through the lens of Relational Complexity (RC), which we define as the minimum number of independent entities or operands that must be simultaneously bound to apply a relation. RC provides a principled way to vary reasoning difficulty while controlling for confounders such as input size, vocabulary, and representational choices. Building on RC, we introduce REL, a generative benchmark framework spanning algebra, chemistry, and biology that varies RC within each domain. Across frontier LLMs, performance degrades consistently and monotonically as RC increases, even when the total number of entities is held fixed. This failure mode persists with increased test-time compute and in-context learning, suggesting a limitation tied to the arity of the required relational binding rather than to insufficient inference steps or lack of exposure to examples. Our results identify a regime of higher-arity reasoning in which current models struggle, and motivate re-examining benchmarks through the lens of relational complexity.

Method: Created PhantomPolicy benchmark with 600 model traces across eight violation categories, manually reviewed all traces. Introduced Sentinel framework that treats agent actions as proposed mutations to organizational knowledge graph, performs speculative execution to materialize post-action world state, and verifies graph-structural invariants.

Result: Manual review changed 32 labels (5.3%) relative to original case-level annotations. Sentinel substantially outperformed content-only DLP baseline (93.0% vs 68.8% accuracy) while maintaining high precision, though with room for improvement on certain violation categories.

Conclusion: Policy-invisible violations are a significant failure mode for LLM agents, and world-state-grounded enforcement through frameworks like Sentinel can substantially improve policy compliance by making policy-relevant world state available to the enforcement layer.

Abstract: LLM-based agents can execute actions that are syntactically valid, user-sanctioned, and semantically appropriate, yet still violate organizational policy because the facts needed for correct policy judgment are hidden at decision time. We call this failure mode policy-invisible violations: cases in which compliance depends on entity attributes, contextual state, or session history absent from the agent’s visible context. We present PhantomPolicy, a benchmark spanning eight violation categories with balanced violation and safe-control cases, in which all tool responses contain clean business data without policy metadata. We manually review all 600 model traces produced by five frontier models and evaluate them using human-reviewed trace labels. Manual review changes 32 labels (5.3%) relative to the original case-level annotations, confirming the need for trace-level human review. To demonstrate what world-state-grounded enforcement can achieve under favorable conditions, we introduce Sentinel, an enforcement framework based on counterfactual graph simulation. Sentinel treats every agent action as a proposed mutation to an organizational knowledge graph, performs speculative execution to materialize the post-action world state, and verifies graph-structural invariants to decide Allow/Block/Clarify. Against human-reviewed trace labels, Sentinel substantially outperforms a content-only DLP baseline (68.8% vs. 93.0% accuracy) while maintaining high precision, though it still leaves room for improvement on certain violation categories. These results demonstrate what becomes achievable once policy-relevant world state is made available to the enforcement layer.

Method: A four-agent baseline pipeline: 1) claim extractor using NER, dependency parsing, and LLMs; 2) retrieval agent with hybrid BM25+FAISS search; 3) verifier agent comparing claims to evidence with calibrated confidence; 4) explainer agent generating human-readable reports. Extended version adds: decomposer agent for complex claims, multi-agent jury with specialized verifier personas, and logic aggregator for combining atomic verdicts.

Result: Evaluated on LIAR benchmark against fine-tuned BERT, RoBERTa, and zero-shot LLM baselines. Supervised encoders performed better on raw metrics, but TRUST Agents improved interpretability, evidence transparency, and reasoning over compound claims. Retrieval quality and uncertainty calibration were identified as main bottlenecks.

Conclusion: TRUST Agents provides a more transparent and explainable approach to fact verification, though performance on raw metrics still lags behind supervised models. The framework demonstrates the value of multi-agent reasoning and decomposition for complex claims, while highlighting retrieval and uncertainty calibration as key challenges.

Abstract: TRUST Agents is a collaborative multi-agent framework for explainable fact verification and fake news detection. Rather than treating verification as a simple true-or-false classification task, the system identifies verifiable claims, retrieves relevant evidence, compares claims against that evidence, reasons under uncertainty, and generates explanations that humans can inspect. The baseline pipeline consists of four specialized agents. A claim extractor uses named entity recognition, dependency parsing, and LLM-based extraction to identify factual claims. A retrieval agent performs hybrid sparse and dense search using BM25 and FAISS. A verifier agent compares claims with retrieved evidence and produces verdicts with calibrated confidence. An explainer agent then generates a human-readable report with explicit evidence citations. To handle complex claims more effectively, we introduce a research-oriented extension with three additional components: a decomposer agent inspired by LoCal-style claim decomposition, a Delphi-inspired multi-agent jury with specialized verifier personas, and a logic aggregator that combines atomic verdicts using conjunction, disjunction, negation, and implication. We evaluate both pipelines on the LIAR benchmark against fine-tuned BERT, fine-tuned RoBERTa, and a zero-shot LLM baseline. Although supervised encoders remain stronger on raw metrics, TRUST Agents improves interpretability, evidence transparency, and reasoning over compound claims. Results also show that retrieval quality and uncertainty calibration remain the main bottlenecks in trustworthy automated fact verification.

Method: Proposes a cognitive diagnostic framework using multidimensional Item Response Theory with item-ability association matrices to estimate fine-grained ability levels across multiple dimensions (35 for math, 27 for physics, 58 for chemistry, 12 for CS).

Result: Strong criterion validity, consistent ability estimates across benchmarks, and accurate prediction of unseen items with AUC 0.80-0.89 within benchmarks and 0.77-0.86 across benchmarks, substantially exceeding baselines.

Conclusion: Establishes a principled framework for fine-grained assessment of LLM abilities with applications in targeted training, ability-guided model selection, and ability-aware benchmark design.

Abstract: Current evaluations of large language models aggregate performance across diverse tasks into single scores. This obscures fine-grained ability variation, limiting targeted model improvement and ability-guided selection for specific tasks. Motivated by this gap, we propose a cognitive diagnostic framework that estimates model abilities across multiple fine-grained dimensions. For mathematics, we construct a 35-dimensional ability taxonomy grounded in cognitive theory and domain knowledge. The framework employs multidimensional Item Response Theory with an item-ability association matrix to estimate fine-grained ability levels, which in turn enable prediction of performance on unseen items (questions of benchmark). Evaluated on 41 models, our approach demonstrates strong criterion validity, consistent ability estimates across benchmarks, and accurate prediction of unseen items with AUC ranging from 0.80 to 0.89 within benchmarks and from 0.77 to 0.86 across benchmarks, substantially exceeding trivial baselines. The framework generalizes across scientific domains, producing consistent diagnostic performance in physics (27 dimensions), chemistry (58 dimensions), and computer science (12 dimensions). This work establishes a principled framework for fine-grained assessment of abilities, with potential applications in targeted training, ability-guided model selection, and ability-aware benchmark design.

Method: Uses travel surveys from 200,000+ residents across five cities, applies graph neural networks to construct spatio-temporal place networks, and employs supervised autoencoders to predict individual exposure vectors from home-space, activity-space, and demographic attributes.

Result: Activity space structure explains most variation in place exposure; mobility shapes social mixing more than sociodemographics; income groups experience similar mixing levels but have stratified activity spaces; transit proximity reduces socioeconomic influence on mixing.

Conclusion: Mobility patterns are more important than sociodemographic characteristics in shaping social mixing experiences, though activity spaces remain economically stratified, highlighting complex urban social dynamics beyond simple mobility data analysis.

Abstract: This study leverages large-scale travel surveys for over 200,000 residents across Boston, Chicago, Hong Kong, London, and Sao Paulo. With rich individual-level data, we make systematic comparisons and reveal patterns in social mixing, which cannot be identified by analyzing high-resolution mobility data alone. Using the same set of data, inferring socioeconomic status from residential neighborhoods yield social mixing levels 16% lower than using self-reported survey data. Besides, individuals over the age of 66 experience greater social mixing than those in late working life (aged 55 to 65), lending data-driven support to the “second youth” hypothesis. Teenagers and women with caregiving responsibilities exhibit lower social mixing levels. Across the five cities, proximity to major transit stations reduces the influence of individual socioeconomic status on social mixing. Finally, we construct detailed spatio-temporal place networks for each city using a graph neural network. Inputs of home-space, activity-space and demographic attributes are embedded and fed into a supervised autoencoder to predict individual exposure vectors. Results show that the structure of individual activity space, i.e., where people travel to, explains most of the variations in place exposure, suggesting that mobility shapes experienced social mixing more than sociodemographic characteristics, home environment, and transit proximity. The ablation tests further discover that, while different income groups may experience similar levels of social mixing, their activity spaces remain stratified by income, resulting in structurally different social mixing experiences.

Method: Proposes StsPatient which captures domain-specific features by extracting steering vectors from contrastive pairs of instructions and responses, and introduces Stochastic Token Modulation (STM) mechanism to regulate intervention probability for severity control.

Result: Comprehensive experiments demonstrate that StsPatient significantly outperforms baselines in both clinical authenticity and severity controllability.

Conclusion: StsPatient provides a scalable and ethical solution for clinical training by enabling fine-grained simulation of cognitively impaired patients with precise control over impairment severity.

Abstract: Simulating Standardized Patients with cognitive impairment offers a scalable and ethical solution for clinical training. However, existing methods rely on discrete prompt engineering and fail to capture the heterogeneity of deficits across varying domains and severity levels. To address this limitation, we propose StsPatient for the fine-grained simulation of cognitively impaired patients. We innovatively capture domain-specific features by extracting steering vectors from contrastive pairs of instructions and responses. Furthermore, we introduce a Stochastic Token Modulation (STM) mechanism to regulate the intervention probability. STM enables precise control over impairment severity while mitigating the instability of conventional vector methods. Comprehensive experiments demonstrate that StsPatient significantly outperforms baselines in both clinical authenticity and severity controllability.

Method: Proposes MMA2A (Multimodal Agent-to-Agent) architecture layer that inspects Agent Card capability declarations to route voice, image, and text parts in their native modalities rather than converting everything to text.

Result: On CrossModal-CS benchmark (50 tasks), MMA2A achieves 52% task completion accuracy vs 32% for text-bottleneck baseline. Gains are concentrated on vision-dependent tasks: +38.5pp for product defect reports and +16.7pp for visual troubleshooting, at 1.8× latency cost.

Conclusion: Routing is a first-order design variable in multi-agent systems that determines available information for downstream reasoning. Modality-native routing paired with capable LLM reasoning significantly improves multimodal task accuracy.

Abstract: Preserving multimodal signals across agent boundaries is necessary for accurate cross-modal reasoning, but it is not sufficient. We show that modality-native routing in Agent-to-Agent (A2A) networks improves task accuracy by 20 percentage points over text-bottleneck baselines, but only when the downstream reasoning agent can exploit the richer context that native routing preserves. An ablation replacing LLM-backed reasoning with keyword matching eliminates the accuracy gap entirely (36% vs. 36%), establishing a two-layer requirement: protocol-level routing must be paired with capable agent-level reasoning for the benefit to materialize. We present MMA2A, an architecture layer atop A2A that inspects Agent Card capability declarations to route voice, image, and text parts in their native modality. On CrossModal-CS, a controlled 50-task benchmark with the same LLM backend, same tasks, and only the routing path varying, MMA2A achieves 52% task completion accuracy versus 32% for the text-bottleneck baseline (95% bootstrap CI on $Δ$TCA: [8, 32] pp; McNemar’s exact $p = 0.006$). Gains concentrate on vision-dependent tasks: product defect reports improve by +38.5 pp and visual troubleshooting by +16.7 pp. This accuracy gain comes at a $1.8\times$ latency cost from native multimodal processing. These results suggest that routing is a first-order design variable in multi-agent systems, as it determines the information available for downstream reasoning.

Method: Used GPT-4o to score 20 authentic handwritten physics exam responses with skill-based rubrics of varying granularity, systematically varying prompting formats and temperature settings, comparing with four human instructors across two scoring rounds.

Result: Human-AI agreement on total scores was comparable to human inter-rater reliability, highest for high- and low-performing responses, but declined for mid-level responses with partial/ambiguous reasoning. Fine-grained checklist rubrics improved consistency over holistic scoring.

Conclusion: Reliable AI-assisted scoring depends primarily on clear, well-structured rubrics, with prompting format playing a secondary role and temperature having limited impact. Provides design recommendations for LLM-assisted scoring in STEM contexts.

Abstract: Student responses in STEM assessments are often handwritten and combine symbolic expressions, calculations, and diagrams, creating substantial variation in format and interpretation. Despite their importance for evaluating students’ reasoning, such responses are time-consuming to score and prone to rater inconsistency, particularly when partial credit is required. Recent advances in large language models (LLMs) have increased attention to AI-assisted scoring, yet evidence remains limited regarding how rubric design and LLM configurations influence reliability across performance levels. This study examined the reliability of AI-assisted scoring of undergraduate physics constructed responses using GPT-4o. Twenty authentic handwritten exam responses were scored across two rounds by four instructors and by the AI model using skill-based rubrics with differing levels of analytic granularity. Prompting format and temperature settings were systematically varied. Overall, human-AI agreement on total scores was comparable to human inter-rater reliability and was highest for high- and low-performing responses, but declined for mid-level responses involving partial or ambiguous reasoning. Criterion-level analyses showed stronger alignment for clearly defined conceptual skills than for extended procedural judgments. A more fine-grained, checklist-based rubric improved consistency relative to holistic scoring. These findings indicate that reliable AI-assisted scoring depends primarily on clear, well-structured rubrics, while prompting format plays a secondary role and temperature has relatively limited impact. More broadly, the study provides transferable design recommendations for implementing reliable LLM-assisted scoring in STEM contexts through skill-based rubrics and controlled LLM settings.

Method: Two-model cooperative system: (1) A hint-generating SLM trained via distillation from a strong LLM produces context-aware hints based on problem statement and accumulated reasoning history; (2) A reasoning SLM uses these hints to solve problems stepwise. Hints provide localized guidance without revealing full solutions.

Result: Experiments across diverse mathematical benchmarks show hint assistance consistently improves reasoning accuracy for SLMs, yielding substantial gains over standard prompting while preserving model efficiency.

Conclusion: Structured collaboration between SLMs via hint generation and reasoning offers an effective, lightweight mechanism for enhancing mathematical reasoning capabilities without scaling model size.

Abstract: Small language models (SLMs) often struggle with complex mathematical reasoning due to limited capacity to maintain long chains of intermediate steps and to recover from early errors. We address this challenge by introducing a hint-assisted reasoning framework that incrementally guides SLMs through multi-step mathematical problem solving. Our approach decomposes solutions into sequential reasoning steps and provides context-aware hints, where hints are generated by a separate SLM trained via distillation from a strong large language model. While the hint-generating SLM alone is not capable of solving the problems, its collaboration with a reasoning SLM enables effective guidance, forming a cooperative two-model system for reasoning. Each hint is generated conditionally on the problem statement and the accumulated reasoning history, providing stepwise, localized guidance without revealing full solutions. This reduces error propagation and allows the reasoning model to focus on manageable subproblems. Experiments across diverse mathematical benchmarks and models demonstrate that hint assistance consistently improves reasoning accuracy for SLMs, yielding substantial gains over standard prompting while preserving model efficiency. These results highlight that structured collaboration between SLMs-via hint generation and reasoning-offers an effective and lightweight mechanism for enhancing mathematical reasoning.

Method: Following PRISMA-ScR guidelines, analyzed 52 studies across five major databases from November 2022 to January 2025 to examine LLM-based pedagogical agents in educational settings.

Result: Identified diverse LLM-based agents across K-12, higher education, and informal learning contexts, with four key design dimensions: interaction approach, domain scope, role complexity, and system integration. Found emerging trends including multi-agent systems, virtual student simulation, immersive technology integration, and learning analytics combinations.

Conclusion: LLM-based pedagogical agents represent a rapidly evolving field with significant potential, but require attention to research gaps and ethical considerations regarding privacy, accuracy, and student autonomy.

Abstract: This scoping review examines the emerging field of Large Language Model (LLM)-based pedagogical agents in educational settings. While traditional pedagogical agents have been extensively studied, the integration of LLMs represents a transformative advancement with unprecedented capabilities in natural language understanding, reasoning, and adaptation. Following PRISMA-ScR guidelines, we analyzed 52 studies across five major databases from November 2022 to January 2025. Our findings reveal diverse LLM-based agents spanning K-12, higher education, and informal learning contexts across multiple subject domains. We identified four key design dimensions characterizing these agents: interaction approach (reactive vs. proactive), domain scope (domain-specific vs. general-purpose), role complexity (single-role vs. multi-role), and system integration (standalone vs. integrated). Emerging trends include multi-agent systems that simulate naturalistic learning environments, virtual student simulation for agent evaluation, integration with immersive technologies, and combinations with learning analytics. We also discuss significant research gaps and ethical considerations regarding privacy, accuracy, and student autonomy. This review provides researchers and practitioners with a comprehensive understanding of LLM-based pedagogical agents while identifying crucial areas for future development in this rapidly evolving field.

Method: Proposes GAM, a hierarchical Graph-based Agentic Memory framework that explicitly decouples memory encoding from consolidation. It isolates ongoing dialogue in an event progression graph and integrates it into a topic associative network only upon semantic shifts. Also introduces a graph-guided, multi-factor retrieval strategy for enhanced context precision.

Result: Experiments on LoCoMo and LongDialQA benchmarks show that GAM consistently outperforms state-of-the-art baselines in both reasoning accuracy and efficiency.

Conclusion: GAM effectively resolves the conflict between rapid context perception and stable knowledge retention in LLM agents, providing a robust memory framework for coherent long-term interactions while minimizing interference and preserving long-term consistency.

Abstract: To sustain coherent long-term interactions, Large Language Model (LLM) agents must navigate the tension between acquiring new information and retaining prior knowledge. Current unified stream-based memory systems facilitate context updates but remain vulnerable to interference from transient noise. Conversely, discrete structured memory architectures provide robust knowledge retention but often struggle to adapt to evolving narratives. To address this, we propose GAM, a hierarchical Graph-based Agentic Memory framework that explicitly decouples memory encoding from consolidation to effectively resolve the conflict between rapid context perception and stable knowledge retention. By isolating ongoing dialogue in an event progression graph and integrating it into a topic associative network only upon semantic shifts, our approach minimizes interference while preserving long-term consistency. Additionally, we introduce a graph-guided, multi-factor retrieval strategy to enhance context precision. Experiments on LoCoMo and LongDialQA indicate that our method consistently outperforms state-of-the-art baselines in both reasoning accuracy and efficiency.

Method: Introduces Frontier-Eng benchmark with 47 tasks across five engineering categories, using industrial-grade simulators and verifiers that provide continuous reward signals and enforce hard feasibility constraints. Evaluates eight frontier language models with representative search frameworks in iterative propose-execute-evaluate loops.

Result: Claude 4.6 Opus achieves the most robust performance, but the benchmark remains challenging for all models. Analysis reveals dual power-law decay in improvement frequency (~1/iteration) and magnitude (~1/improvement count). Width improves parallelism and diversity, but depth remains crucial for hard-won improvements under fixed budgets.

Conclusion: Frontier-Eng establishes a new standard for assessing AI agents’ capacity to integrate domain knowledge with executable feedback to solve complex, open-ended engineering problems, moving beyond binary pass/fail evaluation.

Abstract: Current LLM agent benchmarks, which predominantly focus on binary pass/fail tasks such as code generation or search-based question answering, often neglect the value of real-world engineering that is often captured through the iterative optimization of feasible designs. To this end, we introduce Frontier-Eng, a human-verified benchmark for generative optimization – an iterative propose-execute-evaluate loop in which an agent generates candidate artifacts, receives executable verifier feedback, and revises them under a fixed interaction budget – spanning $47$ tasks across five broad engineering categories. Unlike previous suites, Frontier-Eng tasks are grounded in industrial-grade simulators and verifiers that provide continuous reward signals and enforce hard feasibility constraints under constrained budgets. We evaluate eight frontier language models using representative search frameworks, finding that while Claude 4.6 Opus achieves the most robust performance, the benchmark remains challenging for all models. Our analysis suggests a dual power-law decay in improvement frequency ($\sim$ 1/iteration) and magnitude ($\sim$ 1/improvement count). We further show that although width improves parallelism and diversity, depth remains crucial for hard-won improvements under a fixed budget. Frontier-Eng establishes a new standard for assessing the capacity of AI agents to integrate domain knowledge with executable feedback to solve complex, open-ended engineering problems.

Method: Four-stage pipeline: 1) Vision-based document parsing to detect document regions, 2) OCR for text extraction from detected regions, 3) LLM-based document section hierarchical parsing (DSHP-LLM) to reconstruct document structure into hierarchical tree, 4) DFS-based grouping to construct hierarchical chunks.

Result: Extensive experiments on industrial benchmarks show MultiDocFusion improves retrieval precision by 8-15% and ANLS QA scores by 2-3% compared to baseline methods, demonstrating significant performance gains.

Conclusion: Explicitly leveraging document hierarchy through multimodal structure-aware chunking is critical for enhancing RAG-based QA systems, with MultiDocFusion providing substantial improvements in retrieval and answer quality for industrial documents.

Abstract: RAG-based QA has emerged as a powerful method for processing long industrial documents. However, conventional text chunking approaches often neglect complex and long industrial document structures, causing information loss and reduced answer quality. To address this, we introduce MultiDocFusion, a multimodal chunking pipeline that integrates: (i) detection of document regions using vision-based document parsing, (ii) text extraction from these regions via OCR, (iii) reconstruction of document structure into a hierarchical tree using large language model (LLM)-based document section hierarchical parsing (DSHP-LLM), and (iv) construction of hierarchical chunks through DFS-based grouping. Extensive experiments across industrial benchmarks demonstrate that MultiDocFusion improves retrieval precision by 8-15% and ANLS QA scores by 2-3% compared to baselines, emphasizing the critical role of explicitly leveraging document hierarchy for multimodal document-based QA. These significant performance gains underscore the necessity of structure-aware chunking in enhancing the fidelity of RAG-based QA systems.

Method: Multi-agent pipeline analyzes target LVLMs to identify consistent hallucinations and systematic oversights, distilling these patterns into reusable Structured Reflection Notes. At inference, these notes guide the captioning model on what to avoid (hallucinations) and what to attend to (coverage gaps).

Result: Applied to 8 LVLMs (GPT-4.1 family, Qwen series, InternVL variants), ReflectCAP reaches Pareto frontier for factuality-coverage trade-off, delivers substantial gains on CapArena-Auto benchmark, and offers better quality-compute trade-off than model scaling or existing multi-agent pipelines (21-36% lower overhead).

Conclusion: ReflectCAP enables high-quality detailed captioning that balances factuality and coverage while being computationally efficient, making it viable under real-world cost and latency constraints.

Abstract: Detailed image captioning demands both factual grounding and fine-grained coverage, yet existing methods have struggled to achieve them simultaneously. We address this tension with Reflective Note-Guided Captioning (ReflectCAP), where a multi-agent pipeline analyzes what the target large vision-language model (LVLM) consistently hallucinates and what it systematically overlooks, distilling these patterns into reusable guidelines called Structured Reflection Notes. At inference time, these notes steer the captioning model along both axes – what to avoid and what to attend to – yielding detailed captions that jointly improve factuality and coverage. Applying this method to 8 LVLMs spanning the GPT-4.1 family, Qwen series, and InternVL variants, ReflectCAP reaches the Pareto frontier of the trade-off between factuality and coverage, and delivers substantial gains on CapArena-Auto, where generated captions are judged head-to-head against strong reference models. Moreover, ReflectCAP offers a more favorable trade-off between caption quality and compute cost than model scaling or existing multi-agent pipelines, which incur 21–36% greater overhead. This makes high-quality detailed captioning viable under real-world cost and latency constraints.

Method: Proposes Coupled Weight and Activation Constraints (CWAC): 1) enforces precomputed safety subspace constraints on weight updates, and 2) applies targeted regularization to safety-critical features identified by sparse autoencoders.

Result: Extensive experiments across four widely used LLMs and diverse downstream tasks show CWAC consistently achieves the lowest harmful scores with minimal impact on fine-tuning accuracy, substantially outperforming strong baselines even under high harmful data ratios.

Conclusion: CWAC provides a robust approach to preserve safety alignment during fine-tuning by simultaneously constraining both weights and activations, addressing limitations of existing methods that consider these components in isolation.

Abstract: Safety alignment in Large Language Models (LLMs) remains highly fragile during fine-tuning, where even benign adaptation can degrade pre-trained refusal behaviors and enable harmful responses. Existing defenses typically constrain either weights or activations in isolation, without considering their coupled effects on safety. In this paper, we first theoretically demonstrate that constraining either weights or activations alone is insufficient for safety preservation. To robustly preserve safety alignment, we propose Coupled Weight and Activation Constraints (CWAC), a novel approach that simultaneously enforces a precomputed safety subspace on weight updates and applies targeted regularization to safety-critical features identified by sparse autoencoders. Extensive experiments across four widely used LLMs and diverse downstream tasks show that CWAC consistently achieves the lowest harmful scores with minimal impact on fine-tuning accuracy, substantially outperforming strong baselines even under high harmful data ratios.

Method: Proposes HCoT (Heuristic-Classification-of-Thoughts) prompting schema that integrates LLM reasoning with structured problem space via heuristic classification model controlling reasoning process and providing reusable abstract solutions.

Result: Outperforms existing approaches (Tree-of-Thoughts, Chain-of-Thoughts) on complex inductive reasoning tasks with ill-defined search spaces. On 24 Game task, achieves significantly higher token efficiency than Tree-of-Thoughts-BFS. Achieves Pareto frontier balance between accuracy and computational cost.

Conclusion: HCoT provides effective method to guide LLM reasoning, addressing stochastic generation and reasoning-decoupling issues while offering reusable solutions and better performance-efficiency trade-offs.

Abstract: This paper addresses two limitations of large language models (LLMs) in solving complex problems: (1) their reasoning processes exhibit Bayesian-like stochastic generation, where each token is sampled from a context-dependent probability distribution, leading to inherently random decision trajectories rather than deterministic planning; (2) the reasoning and decision-making mechanisms are statically decoupled, meaning dynamically retrieved domain knowledge fails to dynamically adjust the underlying reasoning strategy. These dual deficiencies result in initial decisions lacking strategic anchoring and reasoning chains often failing to converge on correct solutions, as stochastic generation lacks mechanisms for trajectory correction or knowledge-guided optimization during sequential reasoning. To resolve these issues, we propose a problem-solving method integrated into the LLM’s generation process to guide reasoning. This method, compatible with numerous LLMs and featuring reusable solutions, is grounded in a novel Heuristic-Classification-of-Thoughts prompting schema (HCoT). HCoT synergizes the LLM’s reasoning ability with a structured problem space via a heuristic classification model that controls the reasoning process and provides reusable abstract solutions. Evaluated on two complex inductive reasoning tasks with ill-defined search spaces, HCoT outperforms existing approaches (e.g., Tree-of-Thoughts and Chain-of-Thoughts prompting) in performance. On the well-structured 24 Game task, HCoT demonstrates significantly higher token efficiency compared to the state-of-the-art Tree-of-Thoughts-Breadth-First-Search. In terms of both accuracy and token usage, HCoT achieves a Pareto frontier balance, offering a strong trade-off between performance and computational cost.

Method: Lightweight sequential unlearning framework with two phases: 1) positive fine-tuning to stabilize benign capabilities, followed by 2) layer-restricted negative fine-tuning to suppress sensitive patterns while preserving general language competence.

Result: Effective behavioral suppression on SemEval-2025 LLM Unlearning benchmark with minimal impact on factual accuracy and fluency; GPT-2 shows greater robustness than DistilGPT-2, highlighting model capacity importance.

Conclusion: Sequential unlearning provides a practical, reproducible mechanism for operationalizing data erasure requirements in politically deployed LLMs, balancing privacy compliance with model utility.

Abstract: Large Language Models (LLMs) are increasingly deployed in politically sensitive environments, where memorisation of personal data or confidential content raises regulatory concerns under frameworks such as the GDPR and its Right to be Forgotten. Translating such legal principles into large-scale generative systems presents significant technical challenges. We introduce a lightweight sequential unlearning framework that explicitly separates retention and suppression objectives. The method first stabilises benign capabilities through positive fine-tuning, then applies layer-restricted negative fine-tuning to suppress designated sensitive patterns while preserving general language competence. Experiments on the SemEval-2025 LLM Unlearning benchmark demonstrate effective behavioural suppression with minimal impact on factual accuracy and fluency. GPT-2 exhibits greater robustness than DistilGPT-2, highlighting the role of model capacity in privacy-aligned adaptation. We position sequential unlearning as a practical and reproducible mechanism for operationalising data erasure requirements in politically deployed LLMs.

Method: Threefold approach: 1) Formal characterization of conditions where distinct k-RFPs induce identical k-covers, 2) Optimization strategy removing redundant patterns by retaining single representative per distinct k-cover, 3) Analysis of interpretability and representativeness of patterns selected by ILP model through robustness analysis.

Result: Extensive experiments on real-world datasets show significant reduction in pattern search space, improved computational efficiency, and preservation/enhancement of cluster quality.

Conclusion: The proposed pattern reduction framework effectively addresses redundancy in explainable clustering, making the approach more scalable while maintaining interpretability and clustering performance.

Abstract: Machine learning has become a central research area, with increasing attention devoted to explainable clustering, also known as conceptual clustering, which is a knowledge-driven unsupervised learning paradigm that partitions data into $θ$ disjoint clusters, where each cluster is described by an explicit symbolic representation, typically expressed as a closed pattern or itemset. By providing human-interpretable cluster descriptions, explainable clustering plays an important role in explainable artificial intelligence and knowledge discovery. Recent work improved clustering quality by introducing k-relaxed frequent patterns (k-RFPs), a pattern model that relaxes strict coverage constraints through a generalized kcover definition. This framework integrates constraint-based reasoning, using SAT solvers for pattern generation, with combinatorial optimization, using Integer Linear Programming (ILP) for cluster selection. Despite its effectiveness, this approach suffers from a critical limitation: multiple distinct k-RFPs may induce identical k-covers, leading to redundant symbolic representations that unnecessarily enlarge the search space and increase computational complexity during cluster construction. In this paper, we address this redundancy through a pattern reduction framework. Our contributions are threefold. First, we formally characterize the conditions under which distinct k-RFPs induce identical kcovers, providing theoretical foundations for redundancy detection. Second, we propose an optimization strategy that removes redundant patterns by retaining a single representative pattern for each distinct k-cover. Third, we investigate the interpretability and representativeness of the patterns selected by the ILP model by analyzing their robustness with respect to their induced clusters. Extensive experiments conducted on several real-world datasets demonstrate that the proposed approach significantly reduces the pattern search space, improves computational efficiency, preserves and enhances in some cases the quality of the resulting clusters.

Method: Proposes CIA attack that constructs adversarial queries to induce intermediate agents’ reasoning outputs, then models semantic correlations through global bias disentanglement and LLM-guided weak supervision.

Result: Extensive experiments show CIA achieves average AUC of 0.87 and peak AUC up to 0.99, demonstrating effective topology inference and revealing substantial privacy risks in MAS.

Conclusion: Communication topologies in LLM-based Multi-Agent Systems are vulnerable to inference attacks, posing significant security and privacy threats that need to be addressed.

Abstract: LLM-based Multi-Agent Systems (MAS) have demonstrated remarkable capabilities in solving complex tasks. Central to MAS is the communication topology which governs how agents exchange information internally. Consequently, the security of communication topologies has attracted increasing attention. In this paper, we investigate a critical privacy risk: MAS communication topologies can be inferred under a restrictive black-box setting, exposing system vulnerabilities and posing significant intellectual property threats. To explore this risk, we propose Communication Inference Attack (CIA), a novel attack that constructs new adversarial queries to induce intermediate agents’ reasoning outputs and models their semantic correlations through the proposed global bias disentanglement and LLM-guided weak supervision. Extensive experiments on MAS with optimized communication topologies demonstrate the effectiveness of CIA, achieving an average AUC of 0.87 and a peak AUC of up to 0.99, thereby revealing the substantial privacy risk in MAS.

Method: Two-phase framework: 1) Estimation phase uses novel combination of detection scores, tracking scores, and vehicle density models to automatically determine optimal ROI; 2) Prediction phase performs efficient vehicle counting within estimated ROI. Framework is compatible with any detection/tracking method.

Result: Achieved exceptional accuracy (most videos 100% accuracy) on benchmark datasets (UA-DETRAC, GRAM, CDnet 2014, ATON). Computational efficiency improved 4x faster than full-frame processing. Outperformed existing techniques, especially in complex multi-road scenarios.

Conclusion: The framework provides robust, accurate vehicle counting with significant computational efficiency gains, making it promising for real-time traffic monitoring applications.

Abstract: Accurate vehicle counting through video surveillance is crucial for efficient traffic management. However, achieving high counting accuracy while ensuring computational efficiency remains a challenge. To address this, we propose a fully automated, video-based vehicle counting framework designed to optimize both computational efficiency and counting accuracy. Our framework operates in two distinct phases: \textit{estimation} and \textit{prediction}. In the estimation phase, the optimal region of interest (ROI) is automatically determined using a novel combination of three models based on detection scores, tracking scores, and vehicle density. This adaptive approach ensures compatibility with any detection and tracking method, enhancing the framework’s versatility. In the prediction phase, vehicle counting is efficiently performed within the estimated ROI. We evaluated our framework on benchmark datasets like UA-DETRAC, GRAM, CDnet 2014, and ATON. Results demonstrate exceptional accuracy, with most videos achieving 100% accuracy, while also enhancing computational efficiency, making processing up to four times faster than full-frame processing. The framework outperforms existing techniques, especially in complex multi-road scenarios, demonstrating robustness and superior accuracy. These advancements make it a promising solution for real-time traffic monitoring.

Method: Introduces a comprehensive framework with: 1) extended axiomatic foundation, 2) four-level parametric model covering predicates, literals, clauses, and formulae similarity, 3) two model families (one syntax-sensitive using language models) with contextual weights, and 4) formal constraints for desirable properties.

Result: A formal framework for FOL argument similarity that accounts for structured content and provides explainable similarity measures through contextual weighting.

Conclusion: The framework enables nuanced similarity assessment in First-Order Logic argumentation, addressing limitations of propositional logic approaches and supporting applications in argument aggregation and enthymeme analysis.

Abstract: Similarity in formal argumentation has recently gained attention due to its significance in problems such as argument aggregation in semantics and enthymeme decoding. While existing approaches focus on propositional logic, we address the richer setting of First-Order Logic (FOL), where similarity must account for structured content. We introduce a comprehensive framework for FOL argument similarity, built upon: (1) an extended axiomatic foundation; (2) a four-level parametric model covering predicates, literals, clauses, and formulae similarity; (3) two model families, one syntax-sensitive via language models, both integrating contextual weights for nuanced and explainable similarity; and (4) formal constraints enforcing desirable properties.

Method: Proposes a two-stage framework: (1) Explainer LLM converts raw XAI outputs to natural language narratives, (2) Verifier LLM assesses them on faithfulness, coherence, completeness, and hallucination risk, with (3) iterative refeed mechanism using Verifier’s feedback for refinement.

Result: Experiments across five XAI techniques and datasets using three families of open-weight LLMs show verification is crucial for filtering unreliable explanations while improving linguistic accessibility compared to raw XAI outputs. Entropy Production Rate analysis indicates Verifier’s feedback guides Explainer toward more stable and coherent reasoning.

Conclusion: The framework provides an efficient pathway toward more trustworthy and democratized XAI systems by ensuring higher quality natural language explanations through systematic verification and refinement.

Abstract: Large Language Models (LLMs) are increasingly used to translate the technical outputs of eXplainable Artificial Intelligence (XAI) methods into accessible natural-language explanations. However, existing approaches often lack guarantees of accuracy, faithfulness, and completeness. At the same time, current efforts to evaluate such narratives remain largely subjective or confined to post-hoc scoring, offering no safeguards to prevent flawed explanations from reaching end-users. To address these limitations, this paper proposes a Two-Stage LLM Meta-Verification Framework that consists of (i) an Explainer LLM that converts raw XAI outputs into natural-language narratives, (ii) a Verifier LLM that assesses them in terms of faithfulness, coherence, completeness, and hallucination risk, and (iii) an iterative refeed mechanism that uses the Verifier’s feedback to refine and improve them. Experiments across five XAI techniques and datasets, using three families of open-weight LLMs, show that verification is crucial for filtering unreliable explanations while improving linguistic accessibility compared with raw XAI outputs. In addition, the analysis of the Entropy Production Rate (EPR) during the refinement process indicates that the Verifier’s feedback progressively guides the Explainer toward more stable and coherent reasoning. Overall, the proposed framework provides an efficient pathway toward more trustworthy and democratized XAI systems.

Method: Proposed an evaluation framework for assessing LLMs’ emotional responses to red tape across cultures, applied to a single red-tape scenario, tested cultural prompting strategies, and developed RAMO interface for simulation and data collection.

Result: All models exhibited limited alignment with human emotional responses, with notably weaker performance in Eastern cultures. Cultural prompting strategies were largely ineffective in improving alignment.

Conclusion: Current LLMs have limitations in generating culturally appropriate emotional responses to bureaucratic red tape, highlighting the need for better cultural adaptation in policy simulation models.

Abstract: Improving policymaking is a central concern in public administration. Prior human subject studies reveal substantial cross-cultural differences in citizens’ emotional responses to red tape during policy implementation. While LLM agents offer opportunities to simulate human-like responses and reduce experimental costs, their ability to generate culturally appropriate emotional responses to red tape remains unverified. To address this gap, we propose an evaluation framework for assessing LLMs’ emotional responses to red tape across diverse cultural contexts. As a pilot study, we apply this framework to a single red-tape scenario. Our results show that all models exhibit limited alignment with human emotional responses, with notably weaker performance in Eastern cultures. Cultural prompting strategies prove largely ineffective in improving alignment. We further introduce \textbf{RAMO}, an interactive interface for simulating citizens’ emotional responses to red tape and for collecting human data to improve models. The interface is publicly available at https://ramo-chi.ivia.ch.

Method: IDEA framework extracts LLM decision knowledge into interpretable parametric models over semantically meaningful factors using: 1) joint learning of verbal-to-numerical mappings and decision parameters via EM, 2) correlated sampling preserving factor dependencies, and 3) direct parameter editing with mathematical guarantees.

Result: Experiments across five datasets show IDEA with Qwen-3-32B (78.6%) outperforms DeepSeek R1 (68.1%) and GPT-5.2 (77.9%), achieving perfect factor exclusion and exact calibration - precision unattainable through prompting alone.

Conclusion: IDEA enables calibrated probabilities and quantitative human-AI collaboration by extracting LLM decision knowledge into interpretable models, addressing key limitations of LLMs in high-stakes domains.

Abstract: Large Language Models are increasingly deployed for decision-making, yet their adoption in high-stakes domains remains limited by miscalibrated probabilities, unfaithful explanations, and inability to incorporate expert knowledge precisely. We propose IDEA, a framework that extracts LLM decision knowledge into an interpretable parametric model over semantically meaningful factors. Through joint learning of verbal-to-numerical mappings and decision parameters via EM, correlated sampling that preserves factor dependencies, and direct parameter editing with mathematical guarantees, IDEA produces calibrated probabilities while enabling quantitative human-AI collaboration. Experiments across five datasets show IDEA with Qwen-3-32B (78.6%) outperforms DeepSeek R1 (68.1%) and GPT-5.2 (77.9%), achieving perfect factor exclusion and exact calibration – precision unattainable through prompting alone. The implementation is publicly available at https://github.com/leonbig/IDEA.

Method: Organized competition with four testing tools evaluated on an LLM-based car manual retrieval application. Tools were assessed on effectiveness in exposing failures and diversity of discovered failure-revealing tests using experimental methodology at DeepTest workshop.

Result: Four tools competed and were evaluated; results show varying effectiveness in identifying failure cases where the LLM system fails to appropriately mention warnings from car manuals.

Conclusion: The competition establishes initial benchmarking for LLM testing tools and highlights the importance of systematic testing for safety-critical LLM applications.

Abstract: This report summarizes the results of the first edition of the Large Language Model (LLM) Testing competition, held as part of the DeepTest workshop at ICSE 2026. Four tools competed in benchmarking an LLM-based car manual information retrieval application, with the objective of identifying user inputs for which the system fails to appropriately mention warnings contained in the manual. The testing solutions were evaluated based on their effectiveness in exposing failures and the diversity of the discovered failure-revealing tests. We report on the experimental methodology, the competitors, and the results.

Method: Proposes KnowRL framework that treats hint design as minimal-sufficient guidance problem. Decomposes guidance into atomic knowledge points (KPs), uses Constrained Subset Search (CSS) to construct compact, interaction-aware subsets for training. Identifies and addresses pruning interaction paradox where removing one KP may help while removing multiple such KPs can hurt.

Result: KnowRL-Nemotron-1.5B trained from OpenMath-Nemotron-1.5B achieves 70.08 average accuracy without KP hints at inference (+9.63 points over baseline). With selected KPs, performance improves to 74.16, establishing new SOTA at 1.5B scale across eight reasoning benchmarks.

Conclusion: KnowRL provides an effective framework for knowledge-guided RL training that addresses reward sparsity through minimal-sufficient guidance, achieving state-of-the-art reasoning performance at the 1.5B scale.

Abstract: RLVR improves reasoning in large language models, but its effectiveness is often limited by severe reward sparsity on hard problems. Recent hint-based RL methods mitigate sparsity by injecting partial solutions or abstract templates, yet they typically scale guidance by adding more tokens, which introduce redundancy, inconsistency, and extra training overhead. We propose \textbf{KnowRL} (Knowledge-Guided Reinforcement Learning), an RL training framework that treats hint design as a minimal-sufficient guidance problem. During RL training, KnowRL decomposes guidance into atomic knowledge points (KPs) and uses Constrained Subset Search (CSS) to construct compact, interaction-aware subsets for training. We further identify a pruning interaction paradox – removing one KP may help while removing multiple such KPs can hurt – and explicitly optimize for robust subset curation under this dependency structure. We train KnowRL-Nemotron-1.5B from OpenMath-Nemotron-1.5B. Across eight reasoning benchmarks at the 1.5B scale, KnowRL-Nemotron-1.5B consistently outperforms strong RL and hinting baselines. Without KP hints at inference, KnowRL-Nemotron-1.5B reaches 70.08 average accuracy, already surpassing Nemotron-1.5B by +9.63 points; with selected KPs, performance improves to 74.16, establishing a new state of the art at this scale. The model, curated training data, and code are publicly available at https://github.com/Hasuer/KnowRL.

Method: Proposes a generalized notion of conditional syntax splitting that broadens applicability by allowing subbases to share atoms and nontrivial conditionals. Introduces adjusted inference postulates based on this generalization and evaluates popular inductive inference operators against them.

Result: The new generalization overcomes limitations of previous splitting concepts, identifies genuine vs. simple splittings, and shows that while every operator satisfying generalized splitting also satisfies conditional syntax splitting, the reverse does not hold.

Conclusion: The proposed generalized conditional syntax splitting provides a more practical and applicable framework for nonmonotonic reasoning from conditional belief bases by allowing meaningful overlap between subbases while maintaining beneficial splitting properties.

Abstract: In nonmonotonic reasoning from conditional belief bases, an inference operator satisfying syntax splitting postulates allows for taking only the relevant parts of a belief base into account, provided that the belief base splits into subbases based on disjoint signatures. Because such disjointness is rare in practice, safe conditional syntax splitting has been proposed as a generalization of syntax splitting, allowing the conditionals in the subbases to share some atoms. Recently this overlap of conditionals has been shown to be limited to trivial, self-fulfilling conditionals. In this article, we propose a generalization of safe conditional syntax splittings that broadens the applicability of splitting postulates. In contrast to safe conditional syntax splitting, our generalized notion supports syntax splittings of a belief base Δ where the subbases of Δ may share atoms and nontrivial conditionals. We illustrate how this new notion overcomes limitations of previous splitting concepts, and we identify genuine splittings, separating them from simple splittings that do not provide benefits for inductive inference from Δ. We introduce adjusted inference postulates based on our generalization of conditional syntax splitting, and we evaluate several popular inductive inference operators with respect to these postulates. Furthermore, we show that, while every inductive inference operator satisfying generalized conditional syntax splitting also satisfies conditional syntax splitting, the reverse does not hold.

Method: GCTM-OT uses LLM-based prompting to extract goal candidates from documents, then incorporates these into semantic-aware contrastive learning via optimal transport for topic discovery.

Result: Experimental results on three public subreddit datasets show GCTM-OT outperforms state-of-the-art baselines in topic coherence and diversity while significantly improving alignment with human-provided goals.

Conclusion: The approach paves the way for more human-centric topic discovery systems by successfully integrating human goals into the topic modeling process.

Abstract: Existing topic modeling methods, from LDA to recent neural and LLM-based approaches, which focus mainly on statistical coherence, often produce redundant or off-target topics that miss the user’s underlying intent. We introduce Human-centric Topic Modeling, \emph{Human-TM}), a novel task formulation that integrates a human-provided goal directly into the topic modeling process to produce interpretable, diverse and goal-oriented topics. To tackle this challenge, we propose the \textbf{G}oal-prompted \textbf{C}ontrastive \textbf{T}opic \textbf{M}odel with \textbf{O}ptimal \textbf{T}ransport (GCTM-OT), which first uses LLM-based prompting to extract goal candidates from documents, then incorporates these into semantic-aware contrastive learning via optimal transport for topic discovery. Experimental results on three public subreddit datasets show that GCTM-OT outperforms state-of-the-art baselines in topic coherence and diversity while significantly improving alignment with human-provided goals, paving the way for more human-centric topic discovery systems.

Method: Proposed PF-CD3Q: a safe reinforcement learning approach combining particle filter (PF) with constrained dueling double deep Q-learning (CD3Q). PF estimators track human fatigue and update model parameters in real-time. These estimators integrate with CD3Q by making task-level fatigue predictions during decision-making, excluding tasks exceeding fatigue limits, and formulating the problem as a constrained Markov decision process.

Result: The approach enables real-time fatigue-predictive human-robot task planning and allocation that dynamically adapts to changing human fatigue conditions while maintaining safety constraints and optimizing efficiency.

Conclusion: PF-CD3Q provides a robust solution for dynamic HRTPA under fatigue uncertainty, addressing the limitations of static fatigue models through online parameter estimation and safe reinforcement learning.

Abstract: Human-robot collaborative manufacturing, a core aspect of Industry 5.0, emphasizes ergonomics to enhance worker well-being. This paper addresses the dynamic human-robot task planning and allocation (HRTPA) problem, which involves determining when to perform tasks and who should execute them to maximize efficiency while ensuring workers’ physical fatigue remains within safe limits. The inclusion of fatigue constraints, combined with production dynamics, significantly increases the complexity of the HRTPA problem. Traditional fatigue-recovery models in HRTPA often rely on static, predefined hyperparameters. However, in practice, human fatigue sensitivity varies daily due to factors such as changed work conditions and insufficient sleep. To better capture this uncertainty, we treat fatigue-related parameters as inaccurate and estimate them online based on observed fatigue progression during production. To address these challenges, we propose PF-CD3Q, a safe reinforcement learning (safe RL) approach that integrates the particle filter with constrained dueling double deep Q-learning for real-time fatigue-predictive HRTPA. Specifically, we first develop PF-based estimators to track human fatigue and update fatigue model parameters in real-time. These estimators are then integrated into CD3Q by making task-level fatigue predictions during decision-making and excluding tasks that exceed fatigue limits, thereby constraining the action space and formulating the problem as a constrained Markov decision process (CMDP).

Method: Decompose tasks into subtasks; use hierarchical approach with high-level EBQ (efficient buffer-based deep Q-learning) for task planning and low-level SAP (spatial-aware path planning) for task allocation to appropriate human-robot resources.

Result: Experiments in 3D simulator show EBQ&SAP effectively addresses human-robot TPA problems in complex dynamic production processes, reducing training time and improving performance.

Conclusion: The proposed hierarchical approach successfully handles dynamic human-robot collaboration in manufacturing with spatial awareness and efficient learning.

Abstract: In advanced manufacturing systems, humans and robots collaborate to conduct the production process. Effective task planning and allocation (TPA) is crucial for achieving high production efficiency, yet it remains challenging in complex and dynamic manufacturing environments. The dynamic nature of humans and robots, particularly the need to consider spatial information (e.g., humans’ real-time position and the distance they need to move to complete a task), substantially complicates TPA. To address the above challenges, we decompose production tasks into manageable subtasks. We then implement a real-time hierarchical human-robot TPA algorithm, including a high-level agent for task planning and a low-level agent for task allocation. For the high-level agent, we propose an efficient buffer-based deep Q-learning method (EBQ), which reduces training time and enhances performance in production problems with long-term and sparse reward challenges. For the low-level agent, a path planning-based spatially aware method (SAP) is designed to allocate tasks to the appropriate human-robot resources, thereby achieving the corresponding sequential subtasks. We conducted experiments on a complex real-time production process in a 3D simulator. The results demonstrate that our proposed EBQ&SAP method effectively addresses human-robot TPA problems in complex and dynamic production processes.

Method: Introduce MISID benchmark from social strategy games with fine-grained two-tier annotation. Propose FRACTAM framework using “Decouple-Anchor-Reason” paradigm: extract unimodal factual representations, two-stage retrieval for factual anchoring, construct explicit cross-modal evidence chains.

Result: Evaluation shows MLLMs have critical deficiencies in complex scenarios (text-prior visual hallucination, impaired cross-modal synergy, limited causal cue chaining). FRACTAM enhances mainstream models’ performance in strategic tasks, improving hidden intent detection while maintaining perceptual accuracy.

Conclusion: MISID addresses gap in multimodal intent recognition for complex strategic interactions. FRACTAM framework effectively reduces text bias and improves cross-modal reasoning capabilities for MLLMs in challenging scenarios.

Abstract: Understanding human intent in complex multi-turn interactions remains a fundamental challenge in human-computer interaction and behavioral analysis. While existing intent recognition datasets focus mainly on single utterances or simple dialogues, real-world scenarios often involve sophisticated strategic interactions where participants must maintain complex deceptive narratives over extended periods. To address this gap, we introduce MISID, a comprehensive multimodal, multi-turn, and multi-participant benchmark for intent recognition. Sourced from high-stakes social strategy games, MISID features a fine-grained, two-tier multi-dimensional annotation scheme tailored for long-context discourse analysis and evidence-based causal tracking. Our systematic evaluation of state-of-the-art Multimodal Large Language Models (MLLMs) on MISID reveals critical deficiencies in complex scenarios, including text-prior visual hallucination, impaired cross-modal synergy, and limited capacity in chaining causal cues. Consequently, we propose FRACTAM as a baseline framework. Using a ``Decouple-Anchor-Reason’’ paradigm, FRACTAM reduces text bias by extracting pure unimodal factual representations, employs two-stage retrieval for long-range factual anchoring, and constructs explicit cross-modal evidence chains. Extensive experiments demonstrate that FRACTAM enhances mainstream models’ performance in complex strategic tasks, improving hidden intent detection and inference while maintaining robust perceptual accuracy. Our dataset is available at https://naislab.cn/datasets/MISID.

Method: Proposes a case-based learning framework that extracts and reuses task-relevant knowledge, analytical prompts, and operational skills from real cases, converting past task experience into reusable knowledge assets that can be transferred to new tasks

Result: Achieves consistently strong performance across six complex task categories, matches or outperforms Zero-Shot, Few-Shot, Checklist Prompt, and Rule Memory baselines, with especially clear gains on more complex tasks; advantage increases with task complexity; practical knowledge acquired by one agent can be reused by others

Conclusion: Case-based learning offers a promising path for building professional agents for real-world work by enabling effective transfer of prior experience and structured analysis

Abstract: LLM-based autonomous agents perform well on general reasoning tasks but still struggle to reliably use task structure, key constraints, and prior experience in complex real-world settings. We propose a case-based learning framework that converts experience from past tasks into reusable knowledge assets, allowing agents to transfer prior case experience to new tasks and perform more structured analysis. Unlike methods based mainly on pretrained knowledge or static prompts, our framework emphasizes extracting and reusing task-relevant knowledge, analytical prompts, and operational skills from real cases. We evaluate the method on a unified benchmark of six complex task categories and compare it with Zero-Shot, Few-Shot, Checklist Prompt, and Rule Memory baselines. Results show that our method achieves consistently strong performance across all tasks and matches or outperforms the best baseline in every case, with especially clear gains on more complex tasks. Further analysis shows that the advantage of case-based learning increases with task complexity, and that practical knowledge acquired by one agent can be reused by others. These findings suggest that case-based learning offers a promising path for building professional agents for real-world work.

Method: Tested 11 AI tools (6 general-purpose like ChatGPT/Claude, 5 specialized for math teachers) using Task Analysis Guide framework to modify low-demand math tasks with prompting strategies representing typical teacher approaches rather than optimized prompts.

Result: AI tools averaged 64% success rate in upgrading tasks, with performance ranging from 33% to 88%. Specialized tools were only moderately better than general-purpose ones. Failure modes included both undershooting (maintaining low demand) and overshooting (making tasks too ambitious). A negative correlation (r=-.35) was found between classification ability and upgrade ability.

Conclusion: AI tools have moderate potential for curriculum adaptation but require specialized approaches to effectively support teachers. The distinct capabilities between classification and generation tasks highlight the need for targeted AI development for educational applications.

Abstract: While recent research has explored AI tools’ ability to classify the quality of mathematical tasks (arXiv:2603.03512), little is known about their capacity to increase the quality of existing tasks. This study investigated whether AI tools could successfully upgrade low-cognitive-demand mathematics tasks. Eleven tools were tested, including six broadly available, general-purpose AI tools (e.g., ChatGPT and Claude) and five tools specialized for mathematics teachers (e.g., Khanmigo, coteach.ai). Using the Task Analysis Guide framework (Stein & Smith, 1998), we prompted AI tools to modify two different types of low-demand mathematical tasks. The prompting strategy aimed to represent likely approaches taken by knowledgeable teachers, rather than extensive optimization to find a more effective prompt (i.e., an optimistic typical outcome). On average, AI tools were only moderately successful: tasks were accurately upgraded only 64% of the time, with different AI tool performance ranging from quite weak (33%) to broadly successful (88%). Specialized tools were only moderately more successful than general-purpose tools. Failure modes included both “undershooting” (maintaining low cognitive demand) and “overshooting” (elevating tasks to an overly ambitious target category that likely would be rejected by teachers). Interestingly, there was a small negative correlation (r = -.35) between whether a given AI tool was able to correctly classify the cognitive demand of tasks and whether the AI was able to upgrade tasks, showing that the ability to modify tasks (i.e., a generative task) represents a distinct capability from the ability to classify them (i.e., judgement using a rubric). These findings have important implications for understanding AI’s potential role in curriculum adaptation and highlight the need for specialized approaches to support teachers in modifying instructional materials.

Method: Proposes Analysis-Localization-Reasoning workflow with two-stage training: 1) Supervised Fine-Tuning on high-quality data via knowledge distillation, 2) Evidence-aware Group Relative Policy Optimization for joint evidence localization and answer accuracy optimization, plus Evidence-Guided Resolution Allocation for memory efficiency.

Result: DocSeeker achieves superior performance on both in-domain and out-of-domain tasks, robustly generalizes from short-page training to ultra-long documents, and synergizes with visual Retrieval-Augmented Generation systems.

Conclusion: The proposed paradigm effectively addresses long document understanding challenges in MLLMs and provides a solid foundation for visual RAG systems implementation.

Abstract: Existing Multimodal Large Language Models (MLLMs) suffer from significant performance degradation on the long document understanding task as document length increases. This stems from two fundamental challenges: 1) a low Signal-to-Noise Ratio (SNR), with crucial evidence buried in irrelevant pages; and 2) supervision scarcity, as datasets offering only final short answers provide a weak learning signal. In this paper, we address these challenges by proposing a paradigm that requires the model to execute a structured ``\textbf{Analysis}, \textbf{Localization} and \textbf{Reasoning}’’ workflow. To instill this capability, we design a two-stage training framework: we first perform Supervised Fine-Tuning on high-quality data generated via an efficient knowledge distillation strategy. Subsequently, we employ an Evidence-aware Group Relative Policy Optimization which jointly optimizes for both evidence localization and answer accuracy. Additionally, we introduce a Evidence-Guided Resolution Allocation strategy to mitigate memory constraints of training on multi-pages documents. Extensive experiments demonstrate that DocSeeker achieves superior performance on both in-domain and out-of-domain tasks. We show it robustly generalizes from short-page training to ultra-long documents and is naturally synergistic with visual Retrieval-Augmented Generation systems, serving as a solid foundation for their implementation.

Method: Proposes RePAIR framework with: (1) watchdog model for unlearning intent detection, (2) surgeon model for generating repair procedures, (3) patient model with parameter updates. Core method is STAMP - Steering Through Activation Manipulation with PseudoInverse - a training-free, single-sample unlearning method that redirects MLP activations toward refusal subspace via closed-form pseudoinverse updates.

Result: Achieves near-zero forget scores (Acc_f = 0.00, F-RL = 0.00) while preserving model utility (Acc_r up to 84.47, R-RL up to 0.88), outperforming six state-of-the-art baselines. Low-rank variant enables efficient on-device unlearning with ~3x speedup.

Conclusion: RePAIR establishes an effective and practical framework for user-driven model editing, advancing transparent and on-device control over learned knowledge, with potential extensions to multimodal foundation models.

Abstract: Large language models (LLMs) inherently absorb harmful knowledge, misinformation, and personal data during pretraining on large-scale web corpora, with no native mechanism for selective removal. While machine unlearning offers a principled solution, existing approaches are provider-centric, requiring retraining pipelines, curated retain datasets, and direct intervention by model service providers (MSPs), thereby excluding end users from controlling their own data. We introduce Interactive Machine Unlearning (IMU), a new paradigm in which users can instruct LLMs to forget targeted knowledge through natural language at inference time. To realize IMU, we propose RePAIR, a prompt-aware model repair framework comprising (i) a watchdog model for unlearning intent detection, (ii) a surgeon model for generating repair procedures, and (iii) a patient model whose parameters are updated autonomously. At the core of RePAIR, we develop Steering Through Activation Manipulation with PseudoInverse (STAMP), a training-free, single-sample unlearning method that redirects MLP activations toward a refusal subspace via closed-form pseudoinverse updates. Its low-rank variant reduces computational complexity from O(d^3) to O(r^3 + r^2 * d), enabling efficient on-device unlearning with up to ~3x speedup over training-based baselines. Extensive experiments across harmful knowledge suppression, misinformation correction, and personal data erasure demonstrate that RePAIR achieves near-zero forget scores (Acc_f = 0.00, F-RL = 0.00) while preserving model utility (Acc_r up to 84.47, R-RL up to 0.88), outperforming six state-of-the-art baselines. These results establish RePAIR as an effective and practical framework for user-driven model editing, advancing transparent and on-device control over learned knowledge, with potential extensions to multimodal foundation models.

Method: The paper provides a structured survey with a taxonomy organizing AI methods into three families: 1) agent-level behavior models, 2) environment-level simulation methods, and 3) cognitive and physics-informed methods. It analyzes simulation platform limitations, reviews evaluation protocols/metrics, simulation tools, and datasets.

Result: The survey synthesizes AI methods for mixed autonomy traffic simulation, identifies gaps in existing simulation platforms, and provides a chronological overview of AI methods development in this domain.

Conclusion: The paper bridges traffic engineering and computer science communities by providing a comprehensive review of AI methods for mixed autonomy traffic simulation, offering a unified taxonomy and directions to address current limitations in simulation tools.

Abstract: Autonomous vehicles (AVs) are now operating on public roads, which makes their testing and validation more critical than ever. Simulation offers a safe and controlled environment for evaluating AV performance in varied conditions. However, existing simulation tools mainly focus on graphical realism and rely on simple rule-based models and therefore fail to accurately represent the complexity of driving behaviors and interactions. Artificial intelligence (AI) has shown strong potential to address these limitations; however, despite the rapid progress across AI methodologies, a comprehensive survey of their application to mixed autonomy traffic simulation remains lacking. Existing surveys either focus on simulation tools without examining the AI methods behind them, or cover ego-centric decision-making without addressing the broader challenge of modeling surrounding traffic. Moreover, they do not offer a unified taxonomy of AI methods covering individual behavior modeling to full scene simulation. To address these gaps, this survey provides a structured review and synthesis of AI methods for modeling AV and human driving behavior in mixed autonomy traffic simulation. We introduce a taxonomy that organizes methods into three families: agent-level behavior models, environment-level simulation methods, and cognitive and physics-informed methods. The survey analyzes how existing simulation platforms fall short of the needs of mixed autonomy research and outlines directions to narrow this gap. It also provides a chronological overview of AI methods and reviews evaluation protocols and metrics, simulation tools, and datasets. By covering both traffic engineering and computer science perspectives, we aim to bridge the gap between these two communities.

Method: Construct a biologically inspired digital twin of the visual hierarchy that encodes images into low-level features, generates contour sketches, and iteratively refines them through top-down feedback guided by semantic representations, mirroring the feedforward and recurrent architecture of the human visual cortex.

Result: The resulting symbols bear striking structural resemblance to early pictographs across culturally distant writing systems (Egyptian hieroglyphs, Chinese oracle bone characters, proto-cuneiform) and offer candidate interpretations for undeciphered scripts.

Conclusion: The findings support a neuro-computational origin of pictographic writing and establish a framework in which AI can recapitulate the cognitive processes by which humans first externalized perception into symbols.

Abstract: Humans readily recognize objects from sparse line drawings, a capacity that appears early in development and persists across cultures, suggesting neural rather than purely learned origins. Yet the computational mechanism by which the brain transforms high-level semantic knowledge into low-level visual symbols remains poorly understood. Here we propose that ancient pictographic writing emerged from the brain’s intrinsic tendency to compress visual input into stable, boundary-based abstractions. We construct a biologically inspired digital twin of the visual hierarchy that encodes an image into low-level features, generates a contour sketch, and iteratively refines it through top-down feedback guided by semantic representations, mirroring the feedforward and recurrent architecture of the human visual cortex. The resulting symbols bear striking structural resemblance to early pictographs across culturally distant writing systems, including Egyptian hieroglyphs, Chinese oracle bone characters, and proto-cuneiform, and offer candidate interpretations for undeciphered scripts. Our findings support a neuro-computational origin of pictographic writing and establish a framework in which AI can recapitulate the cognitive processes by which humans first externalized perception into symbols.

Method: Three-part approach: 1) Data synthesis using medical knowledge graphs and real-time online exploration for long-horizon medical deep search training data; 2) Two-stage post-training with SFT and RL to enhance planning, tool invocation, and reflection capabilities; 3) Expert-constructed QuarkMedSearch Benchmark with rigorous manual verification.

Result: QuarkMedSearch achieves state-of-the-art performance among open-source models of comparable scale on the QuarkMedSearch Benchmark while maintaining strong competitiveness on general benchmarks.

Conclusion: The paper demonstrates a successful full-pipeline approach for enhancing agentic foundation models in vertical domains, with specific application to Chinese medical deep search, showing the value of domain-specific data construction, progressive training strategies, and expert-verified evaluation.

Abstract: As agentic foundation models continue to evolve, how to further improve their performance in vertical domains has become an important challenge. To this end, building upon Tongyi DeepResearch, a powerful agentic foundation model, we focus on the Chinese medical deep search scenario and propose QuarkMedSearch, systematically exploring a full-pipeline approach spanning medical multi-hop data construction, training strategies, and evaluation benchmarks to further push and assess its performance upper bound in vertical domains. Specifically, for data synthesis, to address the scarcity of deep search training data in the medical domain, we combine a large-scale medical knowledge graph with real-time online exploration to construct long-horizon medical deep search training data; for post-training, we adopt a two-stage SFT and RL training strategy that progressively enhances the model’s planning, tool invocation, and reflection capabilities required for deep search, while maintaining search efficiency; for evaluation, we collaborate with medical experts to construct the QuarkMedSearch Benchmark through rigorous manual verification. Experimental results demonstrate that QuarkMedSearch achieves state-of-the-art performance among open-source models of comparable scale on the QuarkMedSearch Benchmark, while also maintaining strong competitiveness on general benchmarks.

Method: Proposes LIFE framework: agent-centric system with four components - orchestrator, Agentic Context Engineering, novel memory system, and information lattice learning. Implemented as self-evolving network management for HPC operations.

Result: Demonstrated in closed-loop HPC operations example for detecting and mitigating latency spikes in Kubernetes-like clusters running critical microservices. Framework generalizes to various orthogonal use cases.

Conclusion: LIFE represents emerging direction beyond monolithic transformers toward agentic AI and brain-inspired architectures for sustainable, adaptive HPC systems management.

Abstract: The rapid advancement of AI has changed the character of HPC usage such as dimensioning, provisioning, and execution. Not only has energy demand been amplified, but existing rudimentary continual learning capabilities limit ability of AI to effectively manage HPCs. This paper reviews emerging directions beyond monolithic transformers, emphasizing agentic AI and brain inspired architectures as complementary paths toward sustainable, adaptive systems. We propose LIFE, a reasoning and Learning framework that is Incremental, Flexible, and Energy efficient that is implemented as an agent centric system rather than a single monolithic model. LIFE uniquely combines four components to realize self evolving network management and operations in HPCs. The components are an orchestrator, Agentic Context Engineering, a novel memory system, and information lattice learning. LIFE can also generalize to enable a variety of orthogonal use cases. We ground LIFE in a specific closed loop HPC operations example for detecting and mitigating latency spikes experienced by critical micro services running on a Kubernetes like cluster.

Method: Created a multi-sheet schema cataloging 195 AI safety benchmarks with benchmark-level metadata, metric definitions, paper metadata, and repository activity. Developed a complexity taxonomy and analyzed patterns in benchmark usage, language coverage, and maintenance.

Result: Identified fragmentation: dominated by medium-complexity benchmarks (94/195), few popular benchmarks (7), English-only bias (165/195), evaluation-only resources (170/195), stale repositories (137/195), and inconsistent metric definitions despite similar labels.

Conclusion: The main failure mode is fragmentation, not scarcity. Researchers need shared measurement language, principled benchmark selection, and better stewardship. AISafetyBenchExplorer provides tools for more rigorous benchmark discovery, comparison, and meta-evaluation.

Abstract: The rapid expansion of large language model (LLM) safety evaluation has produced a substantial benchmark ecosystem, but not a correspondingly coherent measurement ecosystem. We present AISafetyBenchExplorer, a structured catalogue of 195 AI safety benchmarks released between 2018 and 2026, organized through a multi-sheet schema that records benchmark-level metadata, metric-level definitions, benchmark-paper metadata, and repository activity. This design enables meta-analysis not only of what benchmarks exist, but also of how safety is operationalized, aggregated, and judged across the literature. Using the updated catalogue, we identify a central structural problem: benchmark proliferation has outpaced measurement standardization. The current landscape is dominated by medium-complexity benchmarks (94/195), while only 7 benchmarks occupy the Popular tier. The workbook further reports strong concentration around English-only evaluation (165/195), evaluation-only resources (170/195), stale GitHub repositories (137/195), stale Hugging Face datasets (96/195), and heavy reliance on arXiv preprints among benchmarks with known venue metadata. At the metric level, the catalogue shows that familiar labels such as accuracy, F1 score, safety score, and aggregate benchmark scores often conceal materially different judges, aggregation rules, and threat models. We argue that the field’s main failure mode is fragmentation rather than scarcity. Researchers now have many benchmark artifacts, but they often lack a shared measurement language, a principled basis for benchmark selection, and durable stewardship norms for post publication maintenance. AISafetyBenchExplorer addresses this gap by providing a traceable benchmark catalogue, a controlled metadata schema, and a complexity taxonomy that together support more rigorous benchmark discovery, comparison, and meta-evaluation.

Method: BEAM formulates heuristic design as a bi-level optimization problem: exterior layer evolves high-level algorithmic structures with function placeholders using genetic algorithms, interior layer realizes placeholders via Monte Carlo Tree Search, with an Adaptive Memory module for complex code generation and Knowledge Augmentation pipeline for evaluation.

Result: BEAM significantly outperforms existing LLM-based hyper heuristics, reducing optimality gap by 37.84% on aggregate in CVRP hybrid algorithm design, and designs a heuristic that outperforms state-of-the-art Maximum Independent Set solver KaMIS.

Conclusion: BEAM’s bi-level evolutionary approach with adaptive memory and knowledge augmentation enables more effective automatic heuristic design for complex optimization problems compared to existing single-layer LLM-based methods.

Abstract: Large Language Model-based Hyper Heuristic (LHH) has recently emerged as an efficient way for automatic heuristic design. However, most existing LHHs just perform well in optimizing a single function within a pre-defined solver. Their single-layer evolution makes them not effective enough to write a competent complete solver. While some variants incorporate hyperparameter tuning or attempt to generate complex code through iterative local modifications, they still lack a high-level algorithmic modeling, leading to limited exploration efficiency. To address this, we reformulate heuristic design as a Bi-level Optimization problem and propose \textbf{BEAM} (Bi-level Memory-adaptive Algorithmic Evolution). BEAM’s exterior layer evolves high-level algorithmic structures with function placeholders through genetic algorithm (GA), while the interior layer realizes these placeholders via Monte Carlo Tree Search (MCTS). We further introduce an Adaptive Memory module to facilitate complex code generation. To support the evaluation for complex code generation, we point out the limitations of starting LHHs from scratch or from code templates and introduce a Knowledge Augmentation (KA) Pipeline. Experimental results on several optimization problems demonstrate that BEAM significantly outperforms existing LHHs, notably reducing the optimality gap by 37.84% on aggregate in CVRP hybrid algorithm design. BEAM also designs a heuristic that outperforms SOTA Maximum Independent Set (MIS) solver KaMIS.

Method: Introduces dual-trace memory encoding where each stored fact is paired with a concrete scene trace - a narrative reconstruction of the moment and context in which the information was learned. This forces the agent to commit to specific contextual details during encoding. Evaluated using LongMemEval-S benchmark with 4,575 sessions and 100 recall questions.

Result: Dual-trace encoding achieved 73.7% overall accuracy vs 53.5% for fact-only control (+20.2pp gain). Significant improvements in temporal reasoning (+40pp), knowledge-update tracking (+25pp), and multi-session aggregation (+30pp). No benefit for single-session retrieval. Achieved gains at no additional token cost.

Conclusion: Dual-trace memory encoding significantly enhances LLM agents’ capabilities for temporal reasoning, change tracking, and cross-session aggregation by creating richer memory traces with contextual details, consistent with encoding specificity theory.

Abstract: LLM agents with persistent memory store information as flat factual records, providing little context for temporal reasoning, change tracking, or cross-session aggregation. Inspired by the drawing effect [3], we introduce dual-trace memory encoding. In this method, each stored fact is paired with a concrete scene trace, a narrative reconstruction of the moment and context in which the information was learned. The agent is forced to commit to specific contextual details during encoding, creating richer, more distinctive memory traces. Using the LongMemEval-S benchmark (4,575 sessions, 100 recall questions), we compare dual-trace encoding against a fact-only control with matched coverage and format over 99 shared questions. Dual-trace achieves 73.7% overall accuracy versus 53.5%, a +20.2 percentage point (pp) gain (95% CI: [+12.1, +29.3], bootstrap p < 0.0001). Gains concentrate in temporal reasoning (+40pp), knowledge-update tracking (+25pp), and multi-session aggregation (+30pp), with no benefit for single-session retrieval, consistent with encoding specificity theory [8]. Token analysis shows dual-trace encoding achieves this gain at no additional cost. We additionally sketch an architectural design for adapting dual-trace encoding to coding agents, with preliminary pilot validation.

Method: Developed Text2Model suite with various LLM strategies (zero-shot, chain-of-thought, knowledge-graphs, grammar-based encoding, agentic approaches) and Text2Zinc dataset of optimization problems in natural language. Uses MiniZinc’s solver-agnostic modeling capabilities.

Result: Co-pilot strategies are competitive with and sometimes improve upon recent research, but LLMs are not yet push-button solutions for combinatorial modeling. Performance gap exists but tools are open-sourced to help close it.

Conclusion: LLMs show promise for combinatorial modeling but require further development. The unified solver-agnostic approach with both satisfaction and optimization problems represents an advancement over existing work.

Abstract: There is growing interest in leveraging large language models (LLMs) for text-to-model translation and optimization tasks. This paper aims to advance this line of research by introducing \textsc{Text2Model} and \textsc{Text2Zinc}. \textsc{Text2Model} is a suite of co-pilots based on several LLM strategies with varying complexity, along with an online leaderboard. \textsc{Text2Zinc} is a cross-domain dataset for capturing optimization and satisfaction problems specified in natural language, along with an interactive editor with built-in AI assistant. While there is an emerging literature on using LLMs for translating combinatorial problems into formal models, our work is the first attempt to integrate \textit{both} satisfaction and optimization problems within a \textit{unified architecture} and \textit{dataset}. Moreover, our approach is \textit{solver-agnostic} unlike existing work that focuses on translation to a solver-specific model. To achieve this, we leverage \textsc{MiniZinc}’s solver-and-paradigm-agnostic modeling capabilities to formulate combinatorial problems. We conduct comprehensive experiments to compare execution and solution accuracy across several single- and multi-call strategies, including; zero-shot prompting, chain-of-thought reasoning, intermediate representations via knowledge-graphs, grammar-based syntax encoding, and agentic approaches that decompose the model into sequential sub-tasks. Our co-pilot strategies are competitive, and in parts improve, recent research in this domain. Our findings indicate that while LLMs are promising they are not yet a push-button technology for combinatorial modeling. We contribute \textsc{Text2Model} co-pilots and leaderboard, and \textsc{Text2Zinc} and interactive editor to open-source to support closing this performance gap.

Method: Proposes Cycle-Consistent Search (CCS) inspired by cycle-consistency techniques. Key hypothesis: optimal search trajectories serve as lossless encodings of question intent. Uses information bottlenecks (excluding final response, NER masking) to prevent information leakage and ensure reconstruction relies on retrieved observations and structural scaffold.

Result: Experiments on question-answering benchmarks show CCS achieves performance comparable to supervised baselines and outperforms prior methods that don’t rely on gold supervision.

Conclusion: CCS provides a scalable training paradigm for search agents in settings where gold supervision is unavailable, demonstrating the effectiveness of cycle-consistency principles for RL-based search optimization.

Abstract: Reinforcement Learning (RL) has shown strong potential for optimizing search agents in complex information retrieval tasks. However, existing approaches predominantly rely on gold supervision, such as ground-truth answers, which is difficult to scale. To address this limitation, we propose Cycle-Consistent Search (CCS), a gold-supervision-free framework for training search agents, inspired by cycle-consistency techniques from unsupervised machine translation and image-to-image translation. Our key hypothesis is that an optimal search trajectory, unlike insufficient or irrelevant ones, serves as a lossless encoding of the question’s intent. Consequently, a high-quality trajectory should preserve the information required to accurately reconstruct the original question, thereby inducing a reward signal for policy optimization. However, naive cycle-consistency objectives are vulnerable to information leakage, as reconstruction may rely on superficial lexical cues rather than the underlying search process. To reduce this effect, we apply information bottlenecks, including exclusion of the final response and named entity recognition (NER) masking of search queries. These constraints force reconstruction to rely on retrieved observations together with the structural scaffold, ensuring that the resulting reward signal reflects informational adequacy rather than linguistic redundancy. Experiments on question-answering benchmarks show that CCS achieves performance comparable to supervised baselines while outperforming prior methods that do not rely on gold supervision. These results suggest that CCS provides a scalable training paradigm for training search agents in settings where gold supervision is unavailable.

Method: Bilevel Late Acceptance Hill Climbing (b-LAHC) algorithm with three phases: greedy descent, neighborhood exploration, and final refinement. Uses surrogate objective at upper level to guide search and accelerate convergence

Result: Achieves superior/competitive performance against 8 state-of-the-art algorithms on IEEE WCCI-2020 benchmark. Sets 9/10 new best-known results on large-scale benchmarks with 1.07% average improvement

Conclusion: The bilevel framework effectively solves large-scale routing problems with hierarchical structure, validated by strong correlation between surrogate objective and complete cost

Abstract: This paper tackles the Electric Capacitated Vehicle Routing Problem (E-CVRP) through a bilevel optimization framework that handles routing and charging decisions separately or jointly depending on the search stage. By analyzing their interaction, we introduce a surrogate objective at the upper level to guide the search and accelerate convergence. A bilevel Late Acceptance Hill Climbing algorithm (b-LAHC) is introduced that operates through three phases: greedy descent, neighborhood exploration, and final solution refinement. b-LAHC operates with fixed parameters, eliminating the need for complex adaptation while remaining lightweight and effective. Extensive experiments on the IEEE WCCI-2020 benchmark show that b-LAHC achieves superior or competitive performance against eight state-of-the-art algorithms. Under a fixed evaluation budget, it attains near-optimal solutions on small-scale instances and sets 9/10 new best-known results on large-scale benchmarks, improving existing records by an average of 1.07%. Moreover, the strong correlation (though not universal) observed between the surrogate objective and the complete cost justifies the use of the surrogate objective while still necessitating a joint solution of both levels, thereby validating the effectiveness of the proposed bilevel framework and highlighting its potential for efficiently solving large-scale routing problems with a hierarchical structure.

Method: PAL analyzes multimodal lecture content (likely combining visual and audio elements from videos) and dynamically engages learners through questions of varying difficulty. It adjusts questions based on learner responses in real-time and generates personalized summaries with tailored examples at the end of sessions.

Result: PAL contributes a novel framework for responsive digital learning that unites multimodal content analysis with adaptive decision-making, demonstrating how AI can move beyond static personalization toward real-time, individualized support.

Conclusion: The work addresses a core challenge in AI-enabled education by creating a system that is both context-aware and adaptive in real time, transforming passive lecture videos into interactive, personalized learning experiences.

Abstract: AI-driven education platforms have made some progress in personalisation, yet most remain constrained to static adaptation–predefined quizzes, uniform pacing, or generic feedback–limiting their ability to respond to learners’ evolving understanding. This shortfall highlights the need for systems that are both context-aware and adaptive in real time. We introduce PAL (Personal Adaptive Learner), an AI-powered platform that transforms lecture videos into interactive learning experiences. PAL continuously analyzes multimodal lecture content and dynamically engages learners through questions of varying difficulty, adjusting to their responses as the lesson unfolds. At the end of a session, PAL generates a personalized summary that reinforces key concepts while tailoring examples to the learner’s interests. By uniting multimodal content analysis with adaptive decision-making, PAL contributes a novel framework for responsive digital learning. Our work demonstrates how AI can move beyond static personalization toward real-time, individualized support, addressing a core challenge in AI-enabled education.

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to determine conclusion due to failed paper fetch

Abstract: Failed to fetch summary for 2604.12336: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.12336&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.12498: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.12498&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed data retrieval

Result: Unable to determine results due to failed data retrieval

Conclusion: Unable to draw conclusions due to failed data retrieval

Abstract: Failed to fetch summary for 2604.12988: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.12988&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to draw conclusions due to access error

Abstract: Failed to fetch summary for 2506.00239: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2506.00239&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to draw conclusions due to fetch failure

Abstract: Failed to fetch summary for 2506.14092: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2506.14092&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2508.04282: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2508.04282&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API access error

Result: Unable to determine results due to API access error

Conclusion: Unable to determine conclusion due to API access error

Abstract: Failed to fetch summary for 2508.12260: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2508.12260&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to missing paper content

Result: Unable to determine results due to missing paper content

Conclusion: Unable to determine conclusion due to missing paper content

Abstract: Failed to fetch summary for 2509.25758: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.25758&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2509.25843: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2509.25843&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable due to API rate limiting

Result: Cannot determine results as paper content is unavailable due to API rate limiting

Conclusion: Cannot determine conclusion as paper content is unavailable due to API rate limiting

Abstract: Failed to fetch summary for 2510.09087: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.09087&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about paper content due to access limitations

Abstract: Failed to fetch summary for 2510.23026: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.23026&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2510.24168: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.24168&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to unavailability of paper content.

Result: Cannot determine results due to unavailability of paper content.

Conclusion: Cannot draw conclusions due to unavailability of paper content.

Abstract: Failed to fetch summary for 2511.00710: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.00710&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2511.02627: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.02627&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as abstract is unavailable due to rate limiting on arXiv API

Result: Cannot determine results as abstract is unavailable due to rate limiting on arXiv API

Conclusion: Cannot determine conclusion as abstract is unavailable due to rate limiting on arXiv API

Abstract: Failed to fetch summary for 2511.08439: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.08439&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to inability to access paper content

Result: Cannot determine results due to inability to access paper content

Conclusion: Cannot draw conclusions due to inability to access paper content

Abstract: Failed to fetch summary for 2511.17714: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.17714&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to determine conclusion due to failed paper fetch

Abstract: Failed to fetch summary for 2512.20798: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.20798&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2601.06794: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.06794&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: No method information available due to API rate limiting error

Result: No results available - paper summary fetch failed

Conclusion: Cannot analyze paper due to technical limitations in accessing content

Abstract: Failed to fetch summary for 2601.09152: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.09152&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2601.10398: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.10398&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2501.17518: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2501.17518&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2603.05044: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.05044&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as abstract is unavailable due to rate limiting error

Result: Cannot determine results as abstract is unavailable due to rate limiting error

Conclusion: Cannot draw conclusions as abstract is unavailable due to rate limiting error

Abstract: Failed to fetch summary for 2503.14333: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2503.14333&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper abstract content.

Result: Cannot determine results without access to the paper abstract content.

Conclusion: Cannot draw conclusions without access to the paper abstract content.

Abstract: Failed to fetch summary for 2603.08291: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.08291&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2504.02169: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2504.02169&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2603.18104: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.18104&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper retrieval

Result: Unable to determine results due to failed paper retrieval

Conclusion: Unable to analyze paper content due to technical retrieval error

Abstract: Failed to fetch summary for 2603.19042: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.19042&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to draw conclusions due to fetch failure

Abstract: Failed to fetch summary for 2603.25326: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.25326&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2604.05407: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.05407&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper abstract

Result: No results available due to failed abstract retrieval

Conclusion: Unable to analyze this specific paper due to technical limitations in accessing the abstract

Abstract: Failed to fetch summary for 2507.01041: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.01041&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to determine conclusion due to failed paper fetch

Abstract: Failed to fetch summary for 2507.08977: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.08977&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2604.08988: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.08988&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2507.22767: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.22767&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to data fetch failure

Result: Unable to determine results due to data fetch failure

Conclusion: Unable to determine conclusion due to data fetch failure

Abstract: Failed to fetch summary for 2604.09601: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.09601&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot analyze method without access to the paper abstract or content

Result: No results available due to technical limitations in accessing the paper information

Conclusion: Cannot provide analysis due to API rate limiting preventing access to the paper abstract

Abstract: Failed to fetch summary for 2604.10288: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10288&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2604.10911: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10911&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access restrictions

Result: Unable to determine results due to access restrictions

Conclusion: Unable to determine conclusion due to access restrictions

Abstract: Failed to fetch summary for 2604.11043: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11043&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to draw conclusions due to access error

Abstract: Failed to fetch summary for 2604.11120: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11120&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable due to API rate limiting

Result: Cannot determine results as paper content is unavailable due to API rate limiting

Conclusion: Cannot determine conclusion as paper content is unavailable due to API rate limiting

Abstract: Failed to fetch summary for 2604.11623: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11623&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions due to lack of paper content

Abstract: Failed to fetch summary for 2510.07285: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.07285&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.11626: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11626&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed paper fetch

Result: Unable to determine results due to failed paper fetch

Conclusion: Unable to determine conclusion due to failed paper fetch

Abstract: Failed to fetch summary for 2405.04108: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2405.04108&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2503.10471: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2503.10471&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2506.01979: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2506.01979&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to determine conclusion due to access error

Abstract: Failed to fetch summary for 2507.01770: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.01770&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about paper content

Abstract: Failed to fetch summary for 2507.04227: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.04227&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: No method information available - paper content inaccessible

Result: No results available - paper summary could not be retrieved

Conclusion: Cannot analyze paper due to technical limitations in accessing content

Abstract: Failed to fetch summary for 2507.13647: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.13647&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2511.21356: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.21356&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2507.21990: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2507.21990&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to missing paper content

Result: Unable to determine results due to missing paper content

Conclusion: Unable to determine conclusion due to missing paper content

Abstract: Failed to fetch summary for 2510.05159: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.05159&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to failed API request

Result: Unable to determine results due to failed API request

Conclusion: Unable to determine conclusion due to failed API request

Abstract: Failed to fetch summary for 2511.00179: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.00179&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot determine conclusion without access to paper content

Abstract: Failed to fetch summary for 2511.13790: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.13790&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2511.17537: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.17537&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about paper content due to access limitations

Abstract: Failed to fetch summary for 2511.22364: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2511.22364&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to API rate limiting preventing access to paper details

Result: Unable to determine results due to API rate limiting preventing access to paper details

Conclusion: Unable to draw conclusions due to API rate limiting preventing access to paper details

Abstract: Failed to fetch summary for 2512.00412: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.00412&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot determine conclusion as paper content is unavailable

Abstract: Failed to fetch summary for 2512.06443: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2512.06443&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2603.00137: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.00137&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method due to missing abstract content

Result: Cannot determine results due to missing abstract content

Conclusion: Cannot draw conclusions due to missing abstract content

Abstract: Failed to fetch summary for 2601.07177: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2601.07177&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without paper content. The arXiv API request was blocked due to rate limiting, so no technical details are available.

Result: Cannot determine results without paper content. The paper analysis is impossible due to API access restrictions.

Conclusion: Cannot draw conclusions about a paper whose content is inaccessible. The arXiv API rate limiting prevents analysis of paper 2602.13851.

Abstract: Failed to fetch summary for 2602.13851: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2602.13851&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as abstract content is unavailable

Result: Cannot determine results as abstract content is unavailable

Conclusion: Cannot draw conclusions without access to the paper’s abstract

Abstract: Failed to fetch summary for 2603.05004: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.05004&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions about paper content due to technical retrieval error

Abstract: Failed to fetch summary for 2603.08819: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.08819&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to the paper abstract.

Result: Cannot determine results without access to the paper abstract.

Conclusion: Cannot draw conclusions without access to the paper abstract.

Abstract: Failed to fetch summary for 2603.21011: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.21011&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to draw conclusions due to fetch failure

Abstract: Failed to fetch summary for 2603.21942: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.21942&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2603.29148: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.29148&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions as paper content is unavailable

Abstract: Failed to fetch summary for 2603.25414: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.25414&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to analyze paper due to technical access issues

Abstract: Failed to fetch summary for 2603.29977: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.29977&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2603.28325: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2603.28325&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot analyze method as paper content is unavailable due to HTTP 429 error

Result: No results available - arXiv API rate limiting prevented access to paper information

Conclusion: Cannot provide analysis due to technical limitations in accessing the paper content

Abstract: Failed to fetch summary for 2604.05164: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.05164&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to fetch failure

Result: Unable to determine results due to fetch failure

Conclusion: Unable to determine conclusion due to fetch failure

Abstract: Failed to fetch summary for 2604.06798: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.06798&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as paper content is unavailable

Result: Cannot determine results as paper content is unavailable

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2604.07349: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.07349&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method as abstract is not available

Result: Cannot determine results as abstract is not available

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2604.10202: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10202&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot determine conclusion without access to paper content

Abstract: Failed to fetch summary for 2604.10842: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.10842&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Unable to determine method due to access error

Result: Unable to determine results due to access error

Conclusion: Unable to draw conclusions due to access error

Abstract: Failed to fetch summary for 2604.11661: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11661&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content.

Result: Cannot determine results without access to paper content.

Conclusion: Cannot draw conclusions without access to paper content.

Abstract: Failed to fetch summary for 2604.11807: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11807&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Cannot determine method without access to paper content

Result: Cannot determine results without access to paper content

Conclusion: Cannot draw conclusions without access to paper content

Abstract: Failed to fetch summary for 2604.11734: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2604.11734&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes a flow matching-based movie dubbing framework with Cognitive Synchronous Diffusion Transformer (CoSync-DiT) that progressively guides noise-to-speech generation through acoustic style adaptation, fine-grained visual calibration, and time-aware context alignment. Includes Joint Semantic and Alignment Regularization (JSAR) to constrain frame-level temporal consistency and semantic consistency.

Result: Extensive experiments on standard benchmarks and challenging in-the-wild dubbing benchmarks demonstrate state-of-the-art performance across multiple metrics.

Conclusion: The proposed CoSync-DiT framework with JSAR effectively addresses lip-sync precision and naturalness issues in movie dubbing, outperforming existing methods in both controlled and real-world scenarios.

Abstract: Movie dubbing aims to synthesize speech that preserves the vocal identity of a reference audio while synchronizing with the lip movements in a target video. Existing methods fail to achieve precise lip-sync and lack naturalness due to explicit alignment at the duration level. While implicit alignment solutions have emerged, they remain susceptible to interference from the reference audio, triggering timbre and pronunciation degradation in in-the-wild scenarios. In this paper, we propose a novel flow matching-based movie dubbing framework driven by the Cognitive Synchronous Diffusion Transformer (CoSync-DiT), inspired by the cognitive process of professional actors. This architecture progressively guides the noise-to-speech generative trajectory by executing acoustic style adapting, fine-grained visual calibrating, and time-aware context aligning. Furthermore, we design the Joint Semantic and Alignment Regularization (JSAR) mechanism to simultaneously constrain frame-level temporal consistency on the contextual outputs and semantic consistency on the flow hidden states, ensuring robust alignment. Extensive experiments on both standard benchmarks and challenging in-the-wild dubbing benchmarks demonstrate that our method achieves the state-of-the-art performance across multiple metrics.

Method: Investigates various design choices in distillation loss for aligning VAE latent representations with SSL features. Explores different alignment approaches including joint-marginal alignment with adaptive weighting, comparing them across three axes: reconstruction, understanding, and generation tasks.

Result: Extensive experiments show that joint-marginal alignment approach with adaptive weighting achieves the best overall performance while allowing for controllable balance between different task objectives.

Conclusion: The optimal alignment approach for VAE-SSL speech representations depends on task requirements, with joint-marginal alignment offering the best trade-off across reconstruction, understanding, and generation tasks.

Abstract: Continuous speech representations based on Variational Autoencoders (VAEs) have emerged as a promising alternative to traditional spectrogram or discrete token based features for speech generation and reconstruction. Recent research has tried to enrich the structural information in VAE latent representations by aligning with self-supervised learning (SSL) features, aiming for better generation performance. However, it remains unclear whether the widely-used alignment approach based on time-axis distillation is optimal when considering more tasks. To address this problem, this paper systematically explores different alignment approaches and analyzes their impact on the performances over three axes: reconstruction, understanding, and generation. We investigate various design choices in the distillation loss. Extensive experiments show that the joint-marginal alignment approach with adaptive weighting can achieve the best overall performance while allowing for a controllable balance.

Method: Proposes nonnegative factorization with prior spectral basis matrices (extracted or from pre-trained library) and nonnegative tensor factorization to extract/detect optimal basis matrices. Uses β-divergence minimization with multiplicative update rules where β controls sparsity.

Result: Experiments show sparsity control (rather than β value in training) is crucial for separation performance. Method provides better separation quality than comparable algorithms across various mixing conditions.

Conclusion: Nonnegative factorization with prior information and sparsity control via β-divergence optimization significantly improves multichannel audio source separation performance.

Abstract: In Gaussian model-based multichannel audio source separation, the likelihood of observed mixtures of source signals is parametrized by source spectral variances and by associated spatial covariance matrices. These parameters are estimated by maximizing the likelihood through an Expectation-Maximization algorithm and used to separate the signals by means of multichannel Wiener filtering. We propose to estimate these parameters by applying nonnegative factorization based on prior information on source variances. In the nonnegative factorization, spectral basis matrices can be defined as the prior information. The matrices can be either extracted or indirectly made available through a redundant library that is trained in advance. In a separate step, applying nonnegative tensor factorization, two algorithms are proposed in order to either extract or detect the basis matrices that best represent the power spectra of the source signals in the observed mixtures. The factorization is achieved by minimizing the $β$-divergence through multiplicative update rules. The sparsity of factorization can be controlled by tuning the value of $β$. Experiments show that sparsity, rather than the value assigned to $β$ in the training, is crucial in order to increase the separation performance. The proposed method was evaluated in several mixing conditions. It provides better separation quality with respect to other comparable algorithms.

Method: Use elastic net regularization with Gabor atoms to create fitting models, then evaluate classification performance with two DL architectures (1D CNN+LSTM and 1D+2D CNN+LSTM) trained with SGDM and ADAM.

Result: Achieved 98.95% classification accuracy using the second architecture with ADAM training and feature matrices from optimal models with high-time low-frequency resolution Gabor atoms and sparsity constraints.

Conclusion: Optimal time-frequency resolution and regularization significantly improve heart sound classification, with the combination of 1D+2D CNN+LSTM architecture and ADAM training yielding best results.

Abstract: In this article, we propose the optimization of the resolution of time-frequency atoms and the regularization of fitting models to obtain better representations of heart sound signals. This is done by evaluating the classification performance of deep learning (DL) networks in discriminating five heart valvular conditions based on a new class of time-frequency feature matrices derived from the fitting models. We inspect several combinations of resolution and regularization, and the optimal one is that provides the highest performance. To this end, a fitting model is obtained based on a heart sound signal and an overcomplete dictionary of Gabor atoms using elastic net regularization of linear models. We consider two different DL architectures, the first mainly consisting of a 1D convolutional neural network (CNN) layer and a long short-term memory (LSTM) layer, while the second is composed of 1D and 2D CNN layers followed by an LSTM layer. The networks are trained with two algorithms, namely stochastic gradient descent with momentum (SGDM) and adaptive moment (ADAM). Extensive experimentation has been conducted using a database containing heart sound signals of five heart valvular conditions. The best classification accuracy of $98.95%$ is achieved with the second architecture when trained with ADAM and feature matrices derived from optimal models obtained with a Gabor dictionary consisting of atoms with high-time low-frequency resolution and imposing sparsity on the models.

Method: Three-tiered framework: Tier-L uses fast label-cosine scoring in audio-text embedding space; Tier-M adds structured matching with clinician-style descriptors; Tier-H employs retrieval-augmented LLM reasoning. A confidence-based router allocates compute adaptively, allowing easy samples to exit early.

Result: Achieves mean AUROC of 0.744 across nine respiratory classification tasks without task-specific training, outperforming prior zero-shot methods and matching/exceeding supervised baselines. Nearly half of samples exit at cheapest tier, with uncertain cases seeing up to 19% relative improvement.

Conclusion: TRIAGE demonstrates effective adaptive computation for zero-shot audio analysis, concentrating computational gains where needed most while maintaining efficiency through early exits for confident predictions.

Abstract: Automated respiratory audio analysis promises scalable, non-invasive disease screening, yet progress is limited by scarce labeled data and costly expert annotation. Zero-shot inference eliminates task-specific supervision, but existing methods apply uniform computation to every input regardless of difficulty. We introduce TRIAGE, a tiered zero-shot framework that adaptively scales test-time compute by routing each audio sample through progressively richer reasoning stages: fast label-cosine scoring in a joint audio-text embedding space (Tier-L), structured matching with clinician-style descriptors (Tier-M), and retrieval-augmented large language model reasoning (Tier-H). A confidence-based router finalizes easy predictions early while allocating additional computation to ambiguous inputs, enabling nearly half of all samples to exit at the cheapest tier. Across nine respiratory classification tasks without task-specific training, TRIAGE achieves a mean AUROC of 0.744, outperforming prior zero-shot methods and matching or exceeding supervised baselines on multiple tasks. Our analysis show that test-time scaling concentrates gains where they matter: uncertain cases see up to 19% relative improvement while confident predictions remain unchanged at minimal cost.

Method: Apply transformer-based architectures (likely Vision Transformer or similar) to analyze spectrogram images generated from machine sounds. Compare transformer embeddings with CNN embeddings specifically for machine fault detection tasks.

Result: Transformer-driven architectures demonstrate effectiveness in analyzing sound data and generate better embeddings than CNNs for machine fault detection, especially when sufficient data is available.

Conclusion: Transformer-based models are superior to CNNs for spectrogram analysis in Sound AI applications like machine fault detection, leveraging their lower inductive biases to learn more effective representations from audio data.

Abstract: In recent years, Sound AI is being increasingly used to predict machine failures. By attaching a microphone to the machine of interest, one can get real time data on machine behavior from the field. Traditionally, Convolutional Neural Net (CNN) architectures have been used to analyze spectrogram images generated from the sounds captured and predict if the machine is functioning as expected. CNN architectures seem to work well empirically even though they have biases like locality and parameter-sharing which may not be completely relevant for spectrogram analysis. With the successful application of transformer-based models in the field of image processing starting with Vision Transformer (ViT) in 2020, there has been significant interest in leveraging these in the field of Sound AI. Since transformer-based architectures have significantly lower inductive biases, they are expected to perform better than CNNs at spectrogram analysis given enough data. This paper demonstrates the effectiveness of transformer-driven architectures in analyzing Sound data and compares the embeddings they generate with CNNs on the specific task of machine fault detection.

Method: SpotSound introduces a novel training objective specifically designed to suppress hallucinated timestamps for events absent from the input. The authors also present SpotSound-Bench, a challenging temporal grounding benchmark where target events occupy less than ~10% of each clip, creating a rigorous ’needle-in-a-haystack’ evaluation scenario.

Result: Experiments demonstrate that SpotSound achieves state-of-the-art results on temporal grounding benchmarks while maintaining robust performance across general downstream audio-language tasks.

Conclusion: SpotSound addresses key limitations in ALMs for temporal grounding through innovative training objectives and challenging benchmarks, advancing the field of audio event localization in complex acoustic environments.

Abstract: Large Audio-Language Models (ALMs) have recently demonstrated remarkable capabilities in holistic audio understanding, yet they remain unreliable for temporal grounding, i.e., the task of pinpointing exactly when an event occurs within long-form audio. This limitation stems from two factors: training data dominated by clip-level supervision lacking precise timestamps, and benchmarks that fail to simulate real-world scenarios where short events are obscured by dense background sounds. In this paper, we introduce SpotSound, an audio language model designed for grounding audio events. SpotSound incorporates a novel training objective, specifically designed to suppress hallucinated timestamps for events absent from the input. Additionally, we present SpotSound-Bench, a challenging temporal grounding benchmark where target events occupy less than ~10% of each clip, creating a rigorous `needle-in-a-haystack’ evaluation. Experiments demonstrate that SpotSound achieves state-of-the-art results on temporal grounding benchmarks while maintaining robust performance across general downstream audio-language tasks. Code, models and benchmark are released on https://loiesun.github.io/spotsound/

Method: Unable to determine method due to technical error in fetching paper content

Result: Unable to determine results due to technical error in fetching paper content

Conclusion: Unable to determine conclusion due to technical error in fetching paper content

Abstract: Failed to fetch summary for 2508.08468: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2508.08468&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes a bootstrap-based framework using convexified neural networks to establish theoretical consistency. Uses warm-start optimization at each bootstrap iteration to avoid refitting from scratch, and includes a transfer learning method to apply the framework to arbitrary neural networks.

Result: Experimental results show significantly better performance compared to baseline CNNs and state-of-the-art methods on various image datasets, with much lower computational load than competitors.

Conclusion: The proposed bootstrap framework provides theoretically consistent uncertainty quantification for CNNs with computational efficiency, enabling reliable uncertainty estimation in critical applications like medical imaging.

Abstract: Despite the popularity of Convolutional Neural Networks (CNN), the problem of uncertainty quantification (UQ) of CNN has been largely overlooked. Lack of efficient UQ tools severely limits the application of CNN in certain areas, such as medicine, where prediction uncertainty is critically important. Among the few existing UQ approaches that have been proposed for deep learning, none of them has theoretical consistency that can guarantee the uncertainty quality. To address this issue, we propose a novel bootstrap based framework for the estimation of prediction uncertainty. The inference procedure we use relies on convexified neural networks to establish the theoretical consistency of bootstrap. Our approach has a significantly less computational load than its competitors, as it relies on warm-starts at each bootstrap that avoids refitting the model from scratch. We further explore a novel transfer learning method so our framework can work on arbitrary neural networks. We experimentally demonstrate our approach has a much better performance compared to other baseline CNNs and state-of-the-art methods on various image datasets.

Method: Schema-Adaptive Tabular Representation Learning transforms structured variables into semantic natural language statements, encodes them with a pretrained LLM to create transferable embeddings, enabling zero-shot alignment across unseen schemas. Integrated into multimodal framework combining tabular and MRI data for dementia diagnosis.

Result: State-of-the-art performance on NACC and ADNI datasets, successful zero-shot transfer to unseen schemas, significantly outperforming clinical baselines including board-certified neurologists in retrospective diagnostic tasks.

Conclusion: LLM-driven approach provides scalable, robust solution for heterogeneous real-world data, offering pathway to extend LLM-based reasoning to structured domains with demonstrated effectiveness in multimodal clinical diagnosis.

Abstract: Machine learning for tabular data remains constrained by poor schema generalization, a challenge rooted in the lack of semantic understanding of structured variables. This challenge is particularly acute in domains like clinical medicine, where electronic health record (EHR) schemas vary significantly. To solve this problem, we propose Schema-Adaptive Tabular Representation Learning, a novel method that leverages large language models (LLMs) to create transferable tabular embeddings. By transforming structured variables into semantic natural language statements and encoding them with a pretrained LLM, our approach enables zero-shot alignment across unseen schemas without manual feature engineering or retraining. We integrate our encoder into a multimodal framework for dementia diagnosis, combining tabular and MRI data. Experiments on NACC and ADNI datasets demonstrate state-of-the-art performance and successful zero-shot transfer to unseen schemas, significantly outperforming clinical baselines, including board-certified neurologists, in retrospective diagnostic tasks. These results validate our LLM-driven approach as a scalable, robust solution for heterogeneous real-world data, offering a pathway to extend LLM-based reasoning to structured domains.

Method: Comprehensive analysis using information-theoretic, geometric, and optimization metrics across model scales (1B-32B), revealing depth-dependent patterns. Proposes Mid-Block Efficient Tuning that selectively updates critical intermediate layers.

Result: Method outperforms standard LoRA by up to 10.2% on GSM8K (OLMo2-7B) with reduced parameter overhead, showing alignment is architecturally localized rather than distributed.

Conclusion: Effective alignment is localized to specific intermediate layers, enabling more efficient fine-tuning strategies that preserve model capabilities while improving instruction-following.

Abstract: While critical for alignment, Supervised Fine-Tuning (SFT) incurs the risk of catastrophic forgetting, yet the layer-wise emergence of instruction-following capabilities remains elusive. We investigate this mechanism via a comprehensive analysis utilizing information-theoretic, geometric, and optimization metrics across model scales (1B-32B). Our experiments reveal a distinct depth-dependent pattern: middle layers (20%-80%) are stable, whereas final layers exhibit high sensitivity. Leveraging this insight, we propose Mid-Block Efficient Tuning, which selectively updates these critical intermediate layers. Empirically, our method outperforms standard LoRA up to 10.2% on GSM8K (OLMo2-7B) with reduced parameter overhead, demonstrating that effective alignment is architecturally localized rather than distributed. The code is publicly available at https://anonymous.4open.science/r/base_sft.

Method: A large-language-model-driven multi-agent framework with four specialized agents that collaboratively execute data profiling, architecture selection from a model zoo, Bayesian hyperparameter optimization, model training, and quality assessment based on user preferences.

Result: AutoSurrogate outperformed expert-designed baselines and domain-agnostic AutoML methods on a 3D geological carbon storage modeling task without manual tuning, producing deployment-ready surrogates from single natural-language sentences.

Conclusion: The framework demonstrates strong potential for practical deployment by bridging the expertise gap and enabling domain scientists to build high-quality DL surrogates through natural-language interaction.

Abstract: High-fidelity numerical simulation of subsurface flow is computationally intensive, especially for many-query tasks such as uncertainty quantification and data assimilation. Deep learning (DL) surrogates can significantly accelerate forward simulations, yet constructing them requires substantial machine learning (ML) expertise - from architecture design to hyperparameter tuning - that most domain scientists do not possess. Furthermore, the process is predominantly manual and relies heavily on heuristic choices. This expertise gap remains a key barrier to the broader adoption of DL surrogate techniques. For this reason, we present AutoSurrogate, a large-language-model-driven multi-agent framework that enables practitioners without ML expertise to build high-quality surrogates for subsurface flow problems through natural-language instructions. Given simulation data and optional preferences, four specialized agents collaboratively execute data profiling, architecture selection from a model zoo, Bayesian hyperparameter optimization, model training, and quality assessment against user-specified thresholds. The system also handles common failure modes autonomously, including restarting training with adjusted configurations when numerical instabilities occur and switching to alternative architectures when predictive accuracy falls short of targets. In our setting, a single natural-language sentence can be sufficient to produce a deployment-ready surrogate model, with minimum human intervention required at any intermediate stage. We demonstrate the utility of AutoSurrogate on a 3D geological carbon storage modeling task, mapping permeability fields to pressure and CO$_2$ saturation fields over 31 timesteps. Without any manual tuning, AutoSurrogate is able to outperform expert-designed baselines and domain-agnostic AutoML methods, demonstrating strong potential for practical deployment.

Method: The authors study the solver-sampler mismatch in three multi-agent negotiation environments: ambiguous fragmented-authority trading-limits, ambiguous unified-opposition trading-limits, and grid-curtailment in emergency electricity management. They compare three reflection conditions (no reflection, bounded reflection, native reasoning) across different model families and extend to GPT-4.1 and GPT-5.2.

Result: Bounded reflection produces substantially more diverse and compromise-oriented trajectories than either no reflection or native reasoning. GPT-5.2 with native reasoning ended in authority decisions in all 45 runs, while GPT-5.2 with bounded reflection recovered compromise outcomes in every environment.

Conclusion: Model capability and simulation fidelity are different objectives. Behavioral simulation should qualify models as samplers, not only as solvers. The paper provides a methodological warning about the potential mismatch between reasoning enhancement and simulation fidelity.

Abstract: Large language models are increasingly used as agents in social, economic, and policy simulations. A common assumption is that stronger reasoning should improve simulation fidelity. We argue that this assumption can fail when the objective is not to solve a strategic problem, but to sample plausible boundedly rational behavior. In such settings, reasoning-enhanced models can become better solvers and worse simulators: they can over-optimize for strategically dominant actions, collapse compromise-oriented terminal behavior, and sometimes exhibit a diversity-without-fidelity pattern in which local variation survives without outcome-level fidelity. We study this solver-sampler mismatch in three multi-agent negotiation environments adapted from earlier simulation work: an ambiguous fragmented-authority trading-limits scenario, an ambiguous unified-opposition trading-limits scenario, and a new-domain grid-curtailment case in emergency electricity management. We compare three reflection conditions, no reflection, bounded reflection, and native reasoning, across two primary model families and then extend the same protocol to direct OpenAI runs with GPT-4.1 and GPT-5.2. Across all three experiments, bounded reflection produces substantially more diverse and compromise-oriented trajectories than either no reflection or native reasoning. In the direct OpenAI extension, GPT-5.2 native ends in authority decisions in 45 of 45 runs across the three experiments, while GPT-5.2 bounded recovers compromise outcomes in every environment. The contribution is not a claim that reasoning is generally harmful. It is a methodological warning: model capability and simulation fidelity are different objectives, and behavioral simulation should qualify models as samplers, not only as solvers.

Method: PERA introduces structured polynomial expansion directly into low-rank factor space, expanding each low-rank factor to synthesize high-order interaction terms before composition, transforming adaptation space into polynomial manifold.

Result: PERA consistently outperforms state-of-the-art methods across diverse benchmarks, with experiments showing high-order nonlinear components (particularly square terms) are crucial for enhanced expressive capacity and robust performance.

Conclusion: PERA offers enhanced expressive capacity and more effective feature utilization compared to existing linear adaptation approaches, providing a more powerful yet efficient fine-tuning method for LLMs.

Abstract: Low-rank adaptation (LoRA) is a widely used strategy for efficient fine-tuning of large language models (LLMs), but its strictly linear structure fundamentally limits expressive capacity. The bilinear formulation of weight updates captures only first-order dependencies between low-rank factors, restricting the modeling of nonlinear and higher-order parameter interactions. In this paper, we propose Polynomial Expansion Rank Adaptation (PERA), a novel method that introduces structured polynomial expansion directly into the low-rank factor space. By expanding each low-rank factor to synthesize high-order interaction terms before composition, PERA transforms the adaptation space into a polynomial manifold capable of modeling richer nonlinear coupling without increasing rank or inference cost. We provide theoretical analysis demonstrating that PERA offers enhanced expressive capacity and more effective feature utilization compare to existing linear adaptation approaches. Empirically, PERA consistently outperforms state-of-the-art methods across diverse benchmarks. Notably, our experiments show that incorporating high-order nonlinear components particularly square terms is crucial for enhancing expressive capacity and maintaining strong and robust performance under various rank settings. Our code is available at https://github.com/zhangwenhao6/PERA

Method: DBGL uses a patient-variable bipartite graph to capture irregular sampling patterns without artificial alignment and models variable relationships for temporal sampling irregularity. It also introduces node-specific temporal decay encoding to capture each variable’s decay rates based on sampling intervals.

Result: DBGL outperforms all baselines on four publicly available datasets, demonstrating superior performance in handling irregular medical time series.

Conclusion: DBGL effectively addresses the limitations of existing methods for irregular medical time series by simultaneously modeling temporal sampling irregularity and variable decay irregularity through a novel bipartite graph learning approach with decay encoding.

Abstract: Irregular Medical Time Series play a critical role in the clinical domain to better understand the patient’s condition. However, inherent irregularity arising from heterogeneous sampling rates, asynchronous observations, and variable gaps poses key challenges for reliable modeling. Existing methods often distort temporal sampling irregularity and missingness patterns while failing to capture variable decay irregularity, resulting in suboptimal representations. To address these limitations, we introduce DBGL, Decay-Aware Bipartite Graph Learning for Irregular Medical Time Series. DBGL first introduces a patient-variable bipartite graph that simultaneously captures irregular sampling patterns without artificial alignment and adaptively models variable relationships for temporal sampling irregularity modeling, enhancing representation learning. To model variable decay irregularity, DBGL designs a novel node-specific temporal decay encoding mechanism that captures each variable’s decay rates based on sampling interval, yielding a more accurate and faithful representation of irregular temporal dynamics. We evaluate the performance of DBGL on four publicly available datasets, and the results show that DBGL outperforms all baselines.

Method: Used a four-stage all-MIT distillation pipeline, followed by experiments with inference-time attention-head interventions and a frozen-base confidence-gated sidecar. Tested across three model families (Qwen3, Gemma 4, SmolLM2) and two domains using SFT/DPO LoRA, attention-head tempering, and training-free sidecar approaches.

Result: Initial reported gains (+33.9-point MCAS, +15.3-point HumanEval) were falsified - the HumanEval improvement was a truncation artifact that inverted to -8.0 points, and MCAS gains disappeared under proper scoring. No method successfully improved dispositions without damaging content or causing stylistic mimicry across five tested models.

Conclusion: The study presents a negative result showing that training behavioral dispositions into small language models is challenging, with methods either damaging content quality or causing mimicry. The paper contributes a falsification pipeline, taxonomy for probe failures, and reveals Gemma 4 E2B’s confidence-correctness decoupling.

Abstract: We set out to train behavioral dispositions (self-verification, uncertainty acknowledgment, feedback integration) into small language models (0.6B to 2.3B effective parameters) through a four-stage all-MIT distillation pipeline, with follow-on experiments on inference-time attention-head interventions and a frozen-base confidence-gated sidecar. An internal draft reported +33.9-point MCAS and +15.3-point HumanEval gains on a Qwen3-0.6B student; a second-pass sanity check falsified both numbers before publication. The HumanEval delta was a truncation artifact (n_predict=512) that inverted to -8.0 points at n_predict=1024; the MCAS gain disappeared under apples-to-apples scoring. That falsification triggered three subsequent arcs. Across (1) SFT/DPO LoRA on three model families and two domains, (2) inference-time attention-head tempering on o_proj, and (3) a training-free frozen-base sidecar reading the final-token hidden state h_last, we find no operator that moves judge-measured disposition without damaging content or collapsing into stylistic mimicry. The failure is consistent across five models (Qwen3-0.6B, Qwen3-1.7B, Qwen3.5-0.8B, Gemma 4 E2B, and SmolLM2-1.7B-Instruct). A within-distribution cross-validation pass (AUC=0.683) collapsed to chance on fresh prompts (AUC=0.516). We contribute a three-arc negative result with mechanism, a two-failure-mode taxonomy for linear h_last probes, and an honest falsification pipeline that converts the class of false positives we ourselves produced into publishable negatives. As an independent finding, Gemma 4 E2B exhibits near-complete confidence-correctness decoupling on the Chef domain (assertion asymmetry -0.009; the model asserts at 91% regardless of correctness).

Method: Extends APJN (averaged partial Jacobian norm) analysis to transformers with bidirectional attention and permutation-symmetric inputs by deriving recurrence relations for activation statistics and APJNs across layers.

Result: Theory predicts attention modifies APJN asymptotic behavior at large depth, matches measurements in deep vision transformers, and shows pre-LayerNorm has power-law APJN growth while tanh-like nonlinearities cause stretched-exponential growth (subcritical).

Conclusion: Criticality picture from residual networks extends to transformers; architectures with tanh-like nonlinearities (DyT, Derf) are more sensitive to initialization and require careful tuning for stable training.

Abstract: We study signal propagation at initialization in transformers through the averaged partial Jacobian norm (APJN), a measure of gradient amplification across layers. We extend APJN analysis to transformers with bidirectional attention and permutation-symmetric input token configurations by deriving recurrence relations for activation statistics and APJNs across layers. Our theory predicts how attention modifies the asymptotic behavior of the APJN at large depth and matches APJNs measured in deep vision transformers. The criticality picture known from residual networks carries over to transformers: the pre-LayerNorm architecture exhibits power-law APJN growth, whereas transformers with LayerNorm replaced by elementwise $\tanh$-like nonlinearities have stretched-exponential APJN growth, indicating that the latter are subcritical. Applied to Dynamic Tanh (DyT) and Dynamic erf (Derf) transformers, the theory explains why these architectures can be more sensitive to initialization and optimization choices and require careful tuning for stable training.

Method: Introduces Thermodynamic Liquid Manifold Network that projects 22 meteorological/geometric variables into Koopman-linearized Riemannian manifold, integrates Spectral Calibration unit and multiplicative Thermodynamic Alpha-Gate to synthesize real-time atmospheric opacity with theoretical clear-sky boundary models.

Result: Achieves RMSE of 18.31 Wh/m² and Pearson correlation of 0.988 over 5-year testing, maintains zero-magnitude nocturnal error across all 1826 testing days, and exhibits sub-30-minute phase response during high-frequency optical transients with only 63,458 parameters.

Conclusion: Establishes a robust, thermodynamically consistent standard for edge-deployable microgrid controllers by structurally enforcing celestial geometry compliance while maintaining lightweight design.

Abstract: The stable operation of autonomous off-grid photovoltaic systems requires solar forecasting algorithms that respect atmospheric thermodynamics. Contemporary deep learning models consistently exhibit critical anomalies, primarily severe temporal phase lags during cloud transients and physically impossible nocturnal power generation. To resolve this divergence between data-driven modeling and deterministic celestial mechanics, this research introduces the Thermodynamic Liquid Manifold Network. The methodology projects 22 meteorological and geometric variables into a Koopman-linearized Riemannian manifold to systematically map complex climatic dynamics. The architecture integrates a Spectral Calibration unit and a multiplicative Thermodynamic Alpha-Gate. This system synthesizes real-time atmospheric opacity with theoretical clear-sky boundary models, structurally enforcing strict celestial geometry compliance. This completely neutralizes phantom nocturnal generation while maintaining zero-lag synchronization during rapid weather shifts. Validated against a rigorous five-year testing horizon in a severe semi-arid climate, the framework achieves an RMSE of 18.31 Wh/m2 and a Pearson correlation of 0.988. The model strictly maintains a zero-magnitude nocturnal error across all 1826 testing days and exhibits a sub-30-minute phase response during high-frequency optical transients. Comprising exactly 63,458 trainable parameters, this ultra-lightweight design establishes a robust, thermodynamically consistent standard for edge-deployable microgrid controllers.

Method: Empirical evaluation on synthetic graph path-finding tasks and realistic reasoning benchmarks (Countdown, boolean satisfiability). Theoretical analysis of a simplified two-layer Transformer on a star graph task, proving MTP induces a two-stage reverse reasoning process.

Result: MTP consistently outperforms NTP on reasoning tasks. Theoretically, MTP induces a two-stage process: first attends to end node, then reconstructs path by tracing intermediate nodes backward. This arises from gradient decoupling property providing cleaner training signals.

Conclusion: Multi-token objectives inherently bias optimization toward robust and interpretable reasoning circuits, offering advantages over next-token prediction for planning and structured reasoning tasks.

Abstract: While next-token prediction (NTP) has been the standard objective for training language models, it often struggles to capture global structure in reasoning tasks. Multi-token prediction (MTP) has recently emerged as a promising alternative, yet its underlying mechanisms remain poorly understood. In this paper, we study how MTP facilitates reasoning, with a focus on planning. Empirically, we show that MTP consistently outperforms NTP on both synthetic graph path-finding tasks and more realistic reasoning benchmarks, such as Countdown and boolean satisfiability problems. Theoretically, we analyze a simplified two-layer Transformer on a star graph task. We prove that MTP induces a two-stage reverse reasoning process: the model first attends to the end node and then reconstructs the path by tracing intermediate nodes backward. This behavior arises from a gradient decoupling property of MTP, which provides a cleaner training signal compared to NTP. Ultimately, our results highlight how multi-token objectives inherently bias optimization toward robust and interpretable reasoning circuits.

Method: Using Artificial Life to test ML models’ ability to distinguish biotic from abiotic samples, showing they can be fooled into detecting life with high confidence even in non-living samples due to out-of-distribution issues.

Result: ML methods are easily fooled into detecting life with near 100% confidence even when analyzing samples incapable of life, due to their susceptibility to out-of-distribution samples that differ from terrestrial training data.

Conclusion: AI methods for extraterrestrial life detection are unreliable and bound to yield significant false positives because extraterrestrial samples are likely out-of the distribution of terrestrial training data, making ML models vulnerable to deception.

Abstract: Modern machine learning methods have been proposed to detect life in extraterrestrial samples, drawing on their ability to distinguish biotic from abiotic samples based on training models using natural and synthetic organic molecular mixtures. Here we show using Artificial Life that such methods are easily fooled into detecting life with near 100% confidence even if the analyzed sample is not capable of life. This is due to modern machine learning methods’ propensity to be easily fooled by out-of-distribution samples. Because extra-terrestrial samples are very likely out of the distribution provided by terrestrial biotic and abiotic samples, using AI methods for life detection is bound to yield significant false positives.

Method: Develops an adversarial attack framework for online bounded-buffer settings using an informed and selective attack strategy that targets time steps where the model shows high confidence and expected prediction error is maximal.

Result: The framework increases prediction error up to 2.42x while performing attacks in fewer than 10% of time steps, demonstrating substantial effectiveness with minimal intervention.

Conclusion: The proposed selective attack strategy provides a practical and effective approach for adversarial attacks in time-series forecasting, achieving significant error increases with very few targeted attacks.

Abstract: Time-series forecasting aims to predict future values by modeling temporal dependencies in historical observations. It is a critical component of many real-world systems, where accurate forecasts improve operational efficiency and help mitigate uncertainty and risk. More recently, machine learning (ML), and especially deep learning (DL)-based models, have gained widespread adoption for time-series forecasting, but they remain vulnerable to adversarial attacks. However, many state-of-the-art attack methods are not directly applicable in time-series settings, where storing complete historical data or performing attacks at every time step is often impractical. This paper proposes an adversarial attack framework for time-series forecasting under an online bounded-buffer setting, leveraging an informed and selective attack strategy. By selectively targeting time steps where the model exhibits high confidence and the expected prediction error is maximal, our framework produces fewer but substantially more effective attacks. Experiments show that our framework can increase the prediction error up to 2.42x, while performing attacks in fewer than 10% of time steps.

Method: Hybrid approach combining rapid frequentist screening for candidate selection with focused Bayesian inference for uncertainty quantification, enabling automated equation discovery at reduced computational cost.

Result: Outperforms SINDy in data efficiency for all systems and noise tolerance for 6/7 chaotic systems, with 100x computational cost reduction compared to bootstrap-based ARGOS; improves yield of valid latent equations in climate applications.

Conclusion: Provides a principled, automated, and computationally efficient framework for equation discovery from scarce, noisy observations, applicable from chaotic systems to latent dynamics in climate patterns.

Abstract: Our ability to predict, control, and ultimately understand complex systems rests on discovering the equations that govern their dynamics. Identifying these equations directly from noisy, limited observations has therefore become a central challenge in data-driven science, yet existing library-based sparse regression methods force a compromise between automation, statistical rigor, and computational efficiency. Here we develop Bayesian-ARGOS, a hybrid framework that reconciles these demands by combining rapid frequentist screening with focused Bayesian inference, enabling automated equation discovery with principled uncertainty quantification at a fraction of the computational cost of existing methods. Tested on seven chaotic systems under varying data scarcity and noise levels, Bayesian-ARGOS outperforms two state-of-the-art methods in most scenarios. It surpasses SINDy in data efficiency for all systems and noise tolerance for six out of the seven, with a two-order-of-magnitude reduction in computational cost compared to bootstrap-based ARGOS. The probabilistic formulation additionally enables a suite of standard statistical diagnostics, including influence analysis and multicollinearity detection that expose failure modes otherwise opaque. When integrated with representation learning (SINDy-SHRED) for high dimensional sea surface temperature reconstruction, Bayesian-ARGOS increases the yield of valid latent equations with significantly improved long horizon stability. Bayesian-ARGOS thus provides a principled, automated, and computationally efficient route from scarce and noisy observations to interpretable governing equations, offering a practical framework for equation discovery across scales, from benchmark chaotic systems to the latent dynamics underlying global climate patterns.

Method: Developed DIAX, a standardized JSON-based format for diabetes time-series data, with an open-source repository providing tools for dataset conversion, cross-format compatibility, visualization, and community contributions.

Result: DIAX is compatible with other standardization efforts and supports major diabetes datasets (DCLP3, DCLP5, IOBP2, PEDAP, T1Dexi, Loop), totaling over 10 million patient-hours of data.

Conclusion: DIAX serves as a translational resource that promotes interoperability, reproducibility, and extensibility for diabetes data analysis, particularly benefiting machine learning applications without imposing data-sharing constraints.

Abstract: Diabetes devices, including Continuous Glucose Monitoring (CGM), Smart Insulin Pens, and Automated Insulin Delivery systems, generate rich time-series data widely used in research and machine learning. However, inconsistent data formats across sources hinder sharing, integration, and analysis. We present DIAX (DIAbetes eXchange), a standardized JSON-based format for unifying diabetes time-series data, including CGM, insulin, and meal signals. DIAX promotes interoperability, reproducibility, and extensibility, particularly for machine learning applications. An open-source repository provides tools for dataset conversion, cross-format compatibility, visualization, and community contributions. DIAX is a translational resource, not a data host, ensuring flexibility without imposing data-sharing constraints. Currently, DIAX is compatible with other standardization efforts and supports major datasets (DCLP3, DCLP5, IOBP2, PEDAP, T1Dexi, Loop), totaling over 10 million patient-hours of data. https://github.com/Center-for-Diabetes-Technology/DIAX

Method: Introduces ResBM architecture with residual encoder-decoder bottleneck modules across pipeline boundaries that maintain low-rank identity paths, enabling end-to-end training as part of model parameters.

Result: Achieves state-of-the-art 128x activation compression without significant convergence rate loss, memory overhead, or compute overhead.

Conclusion: ResBM provides a practical solution for decentralized pipeline parallelism in low-bandwidth environments while maintaining training efficiency.

Abstract: Unlocking large-scale low-bandwidth decentralized training has the potential to utilize otherwise untapped compute resources. In centralized settings, large-scale multi-node training is primarily enabled by data and pipeline parallelism, two techniques that require ultra-high-bandwidth communication. While efficient methods now exist for decentralized data parallelism, pipeline parallelism remains the primary challenge. Recent efforts, such as Subspace Models (SM), have claimed up to 100x activation compression but rely on complex constrained optimization and diverge from true end-to-end training. In this paper, we propose a different approach, based on an architecture designed from the ground up to be native to low-bandwidth communication environments while still applicable to any standard transformer-based architecture. We call this architecture the Residual Bottleneck Model or ResBM, it introduces a residual encoder-decoder bottleneck module across pipeline boundaries that can be trained end-to-end as part of the model’s parameters while preserving an explicit low-rank identity path. We show that ResBMs achieve state-of-the-art 128x activation compression without significant loss in convergence rates and without significant memory or compute overhead.

Method: No method information available due to failed API request. The paper content could not be retrieved for analysis.

Result: No results available. The paper summary could not be fetched from arXiv due to rate limiting restrictions.

Conclusion: Cannot provide conclusion without paper content. The technical issue prevents analysis of this specific paper’s contributions.

Abstract: Failed to fetch summary for 2510.00310: Page request resulted in HTTP 429 (https://export.arxiv.org/api/query?search_query=&id_list=2510.00310&sortBy=relevance&sortOrder=descending&start=0&max_results=100)

Method: Proposes AILFM, an Active Imitation Learning-based scheduling framework that learns near-optimal thermal-aware scheduling policies from Oracle demonstrations with minimal runtime overhead. It accounts for core-level performance heterogeneity and kernel-specific behavior in LFMs.

Result: Extensive experiments show AILFM outperforms state-of-the-art baselines and generalizes well across diverse LFM workloads, maintaining thermal safety while maximizing performance.

Conclusion: AILFM provides an effective solution for thermal-aware scheduling of LFMs on 3D S-NUCA CPU systems, addressing the limitations of existing approaches through learned policies that handle system heterogeneity and workload diversity.

Abstract: Large Foundation Model (LFM) inference is both memory- and compute-intensive, traditionally relying on GPUs. However, the limited availability and high cost have motivated the adoption of high-performance general-purpose CPUs, especially emerging 3D-stacked Static Non-Uniform Cache Architecture (3D S-NUCA) systems. These architectures offer enhanced bandwidth and locality but suffer from severe thermal challenges and uneven cache latencies due to 3D Networks-on-Chip (NoC). Optimal management of thread migration and V/f scaling is non-trivial due to LFM kernel diversity and system heterogeneity. Existing thermal management approaches often rely on oversimplified analytical models and lack adaptability. We propose AILFM, an Active Imitation Learning (AIL)-based scheduling framework that learns near-optimal thermal-aware scheduling policies from Oracle demonstrations with minimal run-time overhead. AILFM accounts for both core-level performance heterogeneity and kernel-specific behavior in LFMs to maintain thermal safety while maximizing performance. Extensive experiments show that AILFM outperforms state-of-the-art baselines and generalizes well across diverse LFM workloads.

Method: Introduces the Linear Centroids Hypothesis (LCH) where features correspond to linear directions of centroids. Existing LRH tools (like sparse autoencoders) can be applied to centroids rather than latent activations. Also enables novel interpretability approaches like circuit identification.

Result: Applying LCH to DINO vision transformers yields sparser feature dictionaries that perform better on downstream tasks. LCH can identify circuits in GPT2-Large, demonstrating practical utility.

Conclusion: LCH provides a more robust framework for feature identification in deep networks, addressing LRH limitations while enabling both reuse of existing tools and novel interpretability approaches.

Abstract: Identifying and understanding the features that a deep network (DN) extracts from its inputs to produce its outputs is a focal point of interpretability research. The Linear Representation Hypothesis (LRH) identifies features in terms of the linear directions formed by the inputs in a DN’s latent space. However, the LRH is limited as it abstracts away from individual components (e.g., neurons and layers), is susceptible to identifying spurious features, and cannot be applied across sub-components (e.g., multiple layers). In this paper, we introduce the Linear Centroids Hypothesis (LCH) as a new framework for identifying the features of a DN. The LCH posits that features correspond to linear directions of centroids, which are vector summarizations of the functional behavior of a DN in a local region of its input space. Interpretability studies under the LCH can leverage existing LRH tools, such as sparse autoencoders, by applying them to the DN’s centroids rather than to its latent activations. We demonstrate that doing so yields sparser feature dictionaries for DINO vision transformers, which also perform better on downstream tasks. The LCH also inspires novel approaches to interpretability; for example, LCH can readily identify circuits in GPT2-Large. For code to study the LCH https://github.com/ThomasWalker1/LinearCentroidsHypothesis .

Method: Analyzed data from 55 epilepsy patients using multichannel iEEG recordings. Derived persistence diagrams from iEEG signals and vectorized them using TDA representations (Carlsson coordinates, persistence images, template functions). Conducted large-scale ablation study across multiple iEEG frequency bands, dimensionality reduction techniques, feature representations, and classifier architectures.

Result: Dimension-reduced topological representations achieved up to 80% balanced accuracy for three-class classification (preictal, ictal, interictal). Classical ML models performed comparably to deep learning models (up to 79.17% balanced accuracy). Pipelines preserving full multichannel feature structure exhibited severe overfitting due to high-dimensional feature space.

Conclusion: Carefully designed topological features can substantially reduce model complexity requirements for seizure detection. Structure-preserving dimensionality reduction is crucial when applying topology-based representations to multichannel neural data.

Abstract: Epileptic seizure detection from EEG signals remains challenging due to the high dimensionality and nonlinear, potentially stochastic, dynamics of neural activity. In this work, we investigate whether features derived from topological data analysis (TDA) can improve the classification of brain states in preictal, ictal and interictal iEEG recordings from epilepsy patients using multichannel data. We analyze data from 55 patients, significantly larger than many previous studies that rely on patient-specific models. Persistence diagrams derived from iEEG signals are vectorized using several TDA representations, including Carlsson coordinates, persistence images, and template functions. To understand how topological representations interact with modern machine learning pipelines, we conduct a large-scale ablation study across multiple iEEG frequency bands, dimensionality reduction techniques, feature representations, and classifier architectures. Our experiments show that dimension-reduced topological representations achieve up to 80% balanced accuracy for three-class classification. Interestingly, classical machine learning models perform comparably to deep learning models, achieving up to 79.17% balanced accuracy, suggesting that carefully designed topological features can substantially reduce model complexity requirements. In contrast, pipelines preserving the full multichannel feature structure exhibit severe overfitting due to the high-dimensional feature space. These findings highlight the importance of structure-preserving dimensionality reduction when applying topology-based representations to multichannel neural data.

Method: Introduces MH-RG DeepONet with parallel pathways: standard DeepONet state pathway for wave field, and conditioning pathway for physical descriptors. Uses pre-branch residual modulator, branch residual gate, and trunk residual gate with low-rank multi-head mechanism to capture multiple conditioned response patterns efficiently.

Result: Achieves consistently lower error than feature-augmented baselines on benchmarks including highly nonlinear conservative wave dynamics and dissipative trapped dynamics. Better preserves phase coherence and fidelity of physically relevant dynamical quantities.

Conclusion: The MH-RG framework effectively integrates compact physical descriptors as residual modulation factors, improving neural operator performance for nonlinear wave dynamics while maintaining parameter efficiency through multi-head gating.

Abstract: Coherent nonlinear wave dynamics are often strongly shaped by a compact set of physically meaningful descriptors of the initial state. Traditional neural operators typically treat the input-output mapping as a largely black-box high-dimensional regression problem, without explicitly exploiting this structured physical context. Common feature-integration strategies usually rely on direct concatenation or FiLM-style affine modulation in hidden latent spaces. Here we introduce a different paradigm, loosely inspired by the complementary roles of state evolution and physically meaningful observables in quantum mechanics: the wave field is learned through a standard DeepONet state pathway, while compact physical descriptors follow a parallel conditioning pathway and act as residual modulation factors on the state prediction. Based on this idea, we develop a Multi-Head Residual-Gated DeepONet (MH-RG), which combines a pre-branch residual modulator, a branch residual gate, and a trunk residual gate with a low-rank multi-head mechanism to capture multiple complementary conditioned response patterns without prohibitive parameter growth. We evaluate the framework on representative benchmarks including highly nonlinear conservative wave dynamics and dissipative trapped dynamics and further perform detailed mechanistic analyses of the learned multi-head gating behavior. Compared with feature-augmented baselines, MH-RG DeepONet achieves consistently lower error while better preserving phase coherence and the fidelity of physically relevant dynamical quantities.

Method: GCB maps graphs into a concept bottleneck subspace where each concept is a meaningful phrase. It applies the information bottleneck principle to refine the concept space, focusing on the most relevant concepts. Predictions are made based on concept activations, explicitly guiding the model toward correct decisions.

Result: GCB achieves intrinsic interpretability with accuracy comparable to black-box Graph Neural Networks. It shows better performance under distribution shifts and data perturbations, demonstrating improved robustness and generalizability through concept-guided prediction.

Conclusion: GCB provides a new paradigm for self-explainable graph learning through concept bottlenecks, offering meaningful phrase-level explanations while maintaining competitive performance and enhanced robustness.

Abstract: We introduce Graph Concept Bottleneck (GCB) as a new paradigm for self-explainable text-attributed graph learning. GCB maps graphs into a subspace, concept bottleneck, where each concept is a meaningful phrase, and predictions are made based on the activation of these concepts. Unlike existing interpretable graph learning methods that primarily rely on subgraphs as explanations, the concept bottleneck provides a new form of interpretation. To refine the concept space, we apply the information bottleneck principle to focus on the most relevant concepts. This not only yields more concise and faithful explanations but also explicitly guides the model to “think” toward the correct decision. We empirically show that GCB achieves intrinsic interpretability with accuracy on par with black-box Graph Neural Networks. Moreover, it delivers better performance under distribution shifts and data perturbations, showing improved robustness and generalizability, benefitting from concept-guided prediction.

Method: Propose an algorithm for linear mixture MDPs that adaptively uses offline data based on its informativeness, with theoretical guarantees on when offline data improves performance

Result: Established regret upper bounds characterizing when offline data is beneficial, with nearly matching lower bounds, and numerical experiments support theoretical findings

Conclusion: The algorithm safely leverages offline data when informative (due to sufficient coverage or small environment shift) and ignores it when uninformative, providing adaptive improvement over purely online learning

Abstract: We study offline-online reinforcement learning in linear mixture Markov decision processes (MDPs) under environment shift. In the offline phase, data are collected by an unknown behavior policy and may come from a mismatched environment, while in the online phase the learner interacts with the target environment. We propose an algorithm that adaptively leverages offline data. When the offline data are informative, either due to sufficient coverage or small environment shift, the algorithm provably improves over purely online learning. When the offline data are uninformative, it safely ignores them and matches the online-only performance. We establish regret upper bounds that explicitly characterize when offline data are beneficial, together with nearly matching lower bounds. Numerical experiments further corroborate our theoretical findings.

Method: Proposes a Bayesian active learning framework where any loss function can be used to derive a unique objective for optimal data acquisition. Shows that for losses taking the form of weighted Bregman divergences, a central component of the corresponding objective can be computed analytically, making the approach practical.

Result: In regression and classification experiments with various losses, the approach reduces test losses relative to existing active learning techniques.

Conclusion: The proposed loss-driven Bayesian active learning framework provides a flexible and rigorous approach to customize data acquisition for specific downstream decision problems and losses, with practical applicability for weighted Bregman divergence losses.

Abstract: The central goal of active learning is to gather data that maximises downstream predictive performance, but popular approaches have limited flexibility in customising this data acquisition to different downstream problems and losses. We propose a rigorous loss-driven approach to Bayesian active learning that allows data acquisition to directly target the loss associated with a given decision problem. In particular, we show how any loss can be used to derive a unique objective for optimal data acquisition. Critically, we then show that any loss taking the form of a weighted Bregman divergence permits analytic computation of a central component of its corresponding objective, making the approach applicable in practice. In regression and classification experiments with a range of different losses, we find our approach reduces test losses relative to existing techniques.

Method: Proposes a simple meta-BO algorithm that adaptively utilizes related-task information when determined useful, falling back to lookahead optimization otherwise within a unified framework.

Result: Demonstrates competitiveness with existing approaches on function optimization tasks while maintaining strong performance in low task-relatedness regimes where test tasks share limited structure with meta-training set.

Conclusion: The adaptive meta-BO approach provides robust performance across varying task-relatedness levels, addressing limitations of traditional meta-BO when task alignment is poor.

Abstract: Bayesian optimization (BO) has for sequential optimization of expensive black-box functions demonstrated practicality and effectiveness in many real-world settings. Meta-Bayesian optimization (meta-BO) focuses on improving the sample efficiency of BO by making use of information from related tasks. Although meta-BO is sample-efficient when task structure transfers, poor alignment between meta-training and test tasks can cause suboptimal queries to be suggested during online optimization. To this end, we propose a simple meta-BO algorithm that utilizes related-task information when determined useful, falling back to lookahead otherwise, within a unified framework. We demonstrate competitiveness of our method with existing approaches on function optimization tasks, while retaining strong performance in low task-relatedness regimes where test tasks share limited structure with the meta-training set.

Method: Uses PAC-learning framework for next-token generators, introduces combinatorial tools to analyze sample complexity scaling with generation length T, and compares End-to-End vs Chain-of-Thought supervision.

Result: For End-to-End learning, sample complexity can have essentially any growth rate between constant and linear in T. For Chain-of-Thought supervision, sample complexity is independent of T, showing intermediate reasoning steps eliminate dependence on generation length.

Conclusion: Chain-of-Thought supervision dramatically reduces sample complexity by eliminating dependence on generation length, providing theoretical justification for using intermediate reasoning steps in training language models.

Abstract: Modern large language models generate text autoregressively, producing tokens one at a time. To study the learnability of such systems, Joshi et al. (COLT 2025) introduced a PAC-learning framework for next-token generators, the primitive underlying autoregressive models. In this framework, an unknown next-token generator maps a sequence of tokens to the next token and is iteratively applied for $T$ steps, producing a chain of tokens whose final token constitutes the model’s output. The learning task is to learn the input-output mapping induced by this autoregressive process. Depending on the available supervision, training examples may reveal only the final output (End-to-End supervision) or the entire generated chain (Chain-of-Thought supervision). This raises two natural questions: how the sample complexity depends on the generation length $T$, and how much Chain-of-Thought supervision can reduce this dependence. In this work we give a nearly complete answer to both questions by uncovering a taxonomy of how the sample complexity scales with $T$. For End-to-End learning, we show that the landscape is remarkably rich: subject to mild conditions, essentially any growth rate $r(T)$ between constant and linear can arise as the sample complexity, and combined with the linear upper bound of Joshi et al., this yields an essentially complete characterization. In contrast, under Chain-of-Thought supervision we show that the sample complexity is independent of $T$, demonstrating that access to intermediate reasoning steps can eliminate the dependence on the generation length altogether. Our analysis introduces new combinatorial tools, and as corollaries we resolve several open questions posed by Joshi et al. regarding the dependence of learnability on the generation length and the role of Chain-of-Thought supervision.

Method: UKS uses model-consistent embeddings to induce discrete latent clusters, then estimates the number of unrevealed clusters within a candidate subset using a Smoothed Good-Turing estimator from its empirical frequency spectrum. It can be combined with existing selection methods via a regularized objective.

Result: Experiments on intent-classification and reasoning benchmarks with frontier LLMs show that augmenting strong baselines with UKS consistently improves ICL accuracy by up to 2-6% under the same selection budget, while also providing insights into task- and model-level latent cluster distributions.

Conclusion: UKS provides an effective, training-free approach to demonstration selection that prioritizes coverage of latent clusters, offering both performance improvements and analytical insights into the structure of tasks and models.

Abstract: In-context learning (ICL) performance depends critically on which demonstrations are placed in the prompt, yet most existing selectors prioritize heuristic notions of relevance or diversity and provide limited insight into the coverage of a demonstration set. We propose Unseen Coverage Selection (UKS), a training-free, subset-level coverage prior motivated by the principle that a good demonstration set should expose the model to latent cluster unrevealed by the currently selected subset. UCS operationalizes this idea by (1) inducing discrete latent clusters from model-consistent embeddings and (2) estimating the number of unrevealed clusters within a candidate subset via a Smoothed Good–Turing estimator from its empirical frequency spectrum. Unlike previous selection methods, UCS is coverage-based and training-free, and can be seamlessly combined with both query-dependent and query-independent selection baselines via a simple regularized objective. Experiments on multiple intent-classification and reasoning benchmarks with frontier Large Language Models show that augmenting strong baselines with UCS consistently improves ICL accuracy by up to 2-6% under the same selection budget, while also yielding insights into task- and model-level latent cluster distributions. Code is available at https://github.com/Raina-Xin/UCS.

Method: Multimodal framework with four-expert Mixture-of-Experts fusion module and trimodal cross-modal contrastive learning. Integrates sequence embeddings from ESM-2, structural embeddings from AlphaFold2-predicted geometries, and dynamics embeddings from Gaussian Network Model B-factors, mode shapes, and residue-residue cross-correlations.

Result: Achieves AUROC 0.897 on ProteinGym benchmark (217 DMS assays, 696k SAVs), outperforming all supervised baselines including Kermut (0.864) and ProteinNPT (0.844), and zero-shot model ESM3 (0.769). Dynamics provides largest marginal contribution over pairwise modality combinations.

Conclusion: TriFit demonstrates that integrating protein dynamics with sequence and structure information significantly improves variant effect prediction, with well-calibrated probabilistic outputs without post-hoc correction.

Abstract: Predicting the functional impact of single amino acid substitutions (SAVs) is central to understanding genetic disease and engineering therapeutic proteins. While protein language models and structure-based methods have achieved strong performance on this task, they systematically neglect protein dynamics; residue flexibility, correlated motions, and allosteric coupling are well-established determinants of mutational tolerance in structural biology, yet have not been incorporated into supervised variant effect predictors. We present TriFit, a multimodal framework that integrates sequence, structure, and protein dynamics through a four-expert Mixture-of-Experts (MoE) fusion module with trimodal cross-modal contrastive learning. Sequence embeddings are extracted via masked marginal scoring with ESM-2 (650M); structural embeddings from AlphaFold2-predicted C-alpha geometries; and dynamics embeddings from Gaussian Network Model (GNM) B-factors, mode shapes, and residue-residue cross-correlations. The MoE router adaptively weights modality combinations conditioned on the input, enabling protein-specific fusion without fixed modality assumptions. On the ProteinGym substitution benchmark (217 DMS assays, 696k SAVs), TriFit achieves AUROC 0.897 +/- 0.0002, outperforming all supervised baselines including Kermut (0.864) and ProteinNPT (0.844), and the best zero-shot model ESM3 (0.769). Ablation studies confirm that dynamics provides the largest marginal contribution over pairwise modality combinations, and TriFit achieves well-calibrated probabilistic outputs (ECE = 0.044) without post-hoc correction.

Method: VISTA uses a validation-informed Marginal Coverage score to identify expert anchors (earlier model states that retain specialized competence over distinct data regions). A coverage-weighted ensemble of these anchors is integrated online during training to regularize the loss landscape and preserve mastered knowledge.

Result: VISTA demonstrates improved robustness and generalization over standard training and prior self-distillation methods across multiple benchmarks, with a lightweight implementation reducing storage overhead by 90% without performance loss.

Conclusion: VISTA effectively addresses trajectory deviation in deep learning optimization by preserving specialized knowledge from earlier training stages through online self-distillation, leading to better generalization and robustness.

Abstract: Deep learning models may converge to suboptimal solutions despite strong validation accuracy, masking an optimization failure we term Trajectory Deviation. This is because as training proceeds, models can abandon high generalization states for specific data sub-populations, thus discarding previously learned latent features without triggering classical overfitting signals. To address this problem we introduce VISTA, an online self-distillation framework that enforces consistency along the optimization trajectory. Using a validation-informed Marginal Coverage score, VISTA identifies expert anchors, which are earlier model states that retain specialized competence over distinct data regions. A coverage-weighted ensemble of these anchors is integrated online during training, regularizing the loss landscape and preserving mastered knowledge. When evaluated across multiple benchmarks, VISTA demonstrates improved robustness and generalization over standard training and prior self-distillation methods, while a lightweight implementation reduces storage overhead by 90% without performance loss.

Method: DEFT adaptively generates high-level sequence features during tree construction by leveraging large language models to propose biologically-informed features tailored to local sequence distributions at each node, with iterative refinement through a reflection mechanism.

Result: Empirical demonstrations show that DEFT discovers human-interpretable and highly predictive sequence features across diverse genomic tasks, outperforming traditional interpretable methods.

Conclusion: DEFT successfully bridges the gap between interpretability and performance in DNA sequence analysis by combining the structured reasoning of decision trees with the feature generation capabilities of large language models.

Abstract: The analysis of DNA sequences has become critical in numerous fields, from evolutionary biology to understanding gene regulation and disease mechanisms. While deep neural networks can achieve remarkable predictive performance, they typically operate as black boxes. Contrasting these black boxes, axis-aligned decision trees offer a promising direction for interpretable DNA sequence analysis, yet they suffer from a fundamental limitation: considering individual raw features in isolation at each split limits their expressivity, which results in prohibitive tree depths that hinder both interpretability and generalization performance. We address this challenge by introducing DEFT, a novel framework that adaptively generates high-level sequence features during tree construction. DEFT leverages large language models to propose biologically-informed features tailored to the local sequence distributions at each node and to iteratively refine them with a reflection mechanism. Empirically, we demonstrate that DEFT discovers human-interpretable and highly predictive sequence features across a diverse range of genomic tasks.

Method: Formulates reward hacking as a robust policy optimization problem over all r-correlated proxy rewards. Derives a tractable max-min formulation where the agent maximizes performance under the worst-case proxy consistent with correlation constraints. Extends to linear reward functions with known features for improved policies and interpretable worst-case rewards.

Result: Experiments show the algorithms consistently outperform occupancy-regularized policy optimization (ORPO) in worst-case returns, offering improved robustness and stability across different levels of proxy-true reward correlation.

Conclusion: The approach provides both robustness and transparency in settings where reward design is inherently uncertain, offering strong guarantees against broader classes of correlated proxies.

Abstract: Designing robust reinforcement learning (RL) agents in the presence of imperfect reward signals remains a core challenge. In practice, agents are often trained with proxy rewards that only approximate the true objective, leaving them vulnerable to reward hacking, where high proxy returns arise from unintended or exploitative behaviors. Recent work formalizes this issue using r-correlation between proxy and true rewards, but existing methods like occupancy-regularized policy optimization (ORPO) optimize against a fixed proxy and do not provide strong guarantees against broader classes of correlated proxies. In this work, we formulate reward hacking as a robust policy optimization problem over the space of all r-correlated proxy rewards. We derive a tractable max-min formulation, where the agent maximizes performance under the worst-case proxy consistent with the correlation constraint. We further show that when the reward is a linear function of known features, our approach can be adapted to incorporate this prior knowledge, yielding both improved policies and interpretable worst-case rewards. Experiments across several environments show that our algorithms consistently outperform ORPO in worst-case returns, and offer improved robustness and stability across different levels of proxy-true reward correlation. These results show that our approach provides both robustness and transparency in settings where reward design is inherently uncertain. The code is available at https://github.com/ZixuanLiu4869/reward_hacking.

Method: SOLARIS uses speculative decoding-inspired approach to proactively predict which user-item pairs will appear in future requests, asynchronously precomputing their foundation model embeddings ahead of time, decoupling costly inference from latency-critical serving.

Result: Deployed across Meta’s advertising system serving billions of daily requests, SOLARIS achieved 0.67% revenue-driving top-line metrics gain, demonstrating effectiveness at scale.

Conclusion: SOLARIS enables real-time knowledge transfer from previously too-expensive foundation models by decoupling inference from serving, making large models practical for online recommendation systems.

Abstract: Recent advances in recommendation scaling laws have led to foundation models of unprecedented complexity. While these models offer superior performance, their computational demands make real-time serving impractical, often forcing practitioners to rely on knowledge distillation-compromising serving quality for efficiency. To address this challenge, we present SOLARIS (Speculative Offloading of Latent-bAsed Representation for Inference Scaling), a novel framework inspired by speculative decoding. SOLARIS proactively precomputes user-item interaction embeddings by predicting which user-item pairs are likely to appear in future requests, and asynchronously generating their foundation model representations ahead of time. This approach decouples the costly foundation model inference from the latency-critical serving path, enabling real-time knowledge transfer from models previously considered too expensive for online use. Deployed across Meta’s advertising system serving billions of daily requests, SOLARIS achieves 0.67% revenue-driving top-line metrics gain, demonstrating its effectiveness at scale.

Method: The method uses an E(3)-equivariant graph neural network with tensor-product message passing, spherical harmonic edge features, absorber-query attention pooling, custom equivariant layer normalization, adaptive gated residual connections, and a multi-scale Gaussian basis spectral readout with optional sigmoidal background term. Training employs a composite objective including pointwise spectral reconstruction plus first- and second-derivative matching.

Result: The model achieves a spectrum mean squared error of 1.0×10⁻³ on a test set of 5,941 FDMNES simulations of iron oxide surface facets, accurately reproducing main edge structure, relative peak intensities, pre-edge features, and post-edge oscillations. Ablation studies show each component improves performance.

Conclusion: XANE(3) establishes an accurate and efficient surrogate for XANES simulation, offering a promising route toward accelerated spectral prediction, ML-assisted spectroscopy, and data-driven materials discovery, though explicit tensorial channels may not be strictly required for low intensity error on this dataset.

Abstract: We present XANE(3), a physics-based E(3)-equivariant graph neural network for predicting X-ray absorption near-edge structure (XANES) spectra directly from atomic structures. The model combines tensor-product message passing with spherical harmonic edge features, absorber-query attention pooling, custom equivariant layer normalization, adaptive gated residual connections, and a spectral readout based on a multi-scale Gaussian basis with an optional sigmoidal background term. To improve line-shape fidelity, training is performed with a composite objective that includes pointwise spectral reconstruction together with first- and second-derivative matching terms. We evaluate the model on a dataset of 5,941 FDMNES simulations of iron oxide surface facets and obtain a spectrum mean squared error of $1.0 \times 10^{-3}$ on the test set. The model accurately reproduces the main edge structure, relative peak intensities, pre-edge features, and post-edge oscillations. Ablation studies show that the derivative-aware objective, custom equivariant normalization, absorber-conditioned attention pooling, adaptive gated residual mixing, and global background term each improve performance. Interestingly, a capacity-matched scalar-only variant achieves comparable pointwise reconstruction error but reduced derivative-level fidelity, indicating that explicit tensorial channels are not strictly required for low intensity error on this dataset, although they remain beneficial for capturing finer spectral structure. These results establish XANE(3) as an accurate and efficient surrogate for XANES simulation and offer a promising route toward accelerated spectral prediction, ML-assisted spectroscopy, and data-driven materials discovery.

Method: Analyze transformers trained on a finite set of discrete Markov chains, identifying four algorithmic phases and multi-layer subcircuits implementing context-adaptive computations through minimal models and symmetry-constrained training dynamics theory.

Result: Transformers develop distinct subcircuits implementing in-context learning with four phases: memorization vs generalization and 1-point vs 2-point statistics usage. Phase transitions occur at boundaries K₁* (kinetic competition) and K₂* (representational bottleneck).

Conclusion: Transformers develop specialized subcircuits for in-context learning, with specific conditions favoring different mechanisms, providing mechanistic understanding of how networks adapt to input statistics.

Abstract: Modern distributed networks, notably transformers, acquire a remarkable ability (termed `in-context learning’) to adapt their computation to input statistics, such that a fixed network can be applied to data from a broad range of systems. Here, we provide a complete mechanistic characterization of this behavior in transformers trained on a finite set $S$ of discrete Markov chains. The transformer displays four algorithmic phases, characterized by whether the network memorizes and generalizes, and whether it uses 1-point or 2-point statistics. We show that the four phases are implemented by multi-layer subcircuits that exemplify two qualitatively distinct mechanisms for implementing context-adaptive computations. Minimal models isolate the key features of both motifs. Memorization and generalization phases are delineated by two boundaries that depend on data diversity, $K = |S|$. The first ($K_1^\ast$) is set by a kinetic competition between subcircuits and the second ($K_2^\ast$) is set by a representational bottleneck. A symmetry-constrained theory of a transformer’s training dynamics explains the sharp transition from 1-point to 2-point generalization and identifies key features of the loss landscape that allow the network to generalize. Put together, we show that transformers develop distinct subcircuits to implement in-context learning and identify conditions that favor certain mechanisms over others.

Method: Proposes federated RLVR framework with LoRA-based local adaptation and public-data-based off-policy steps. Uses small shared public dataset to periodically exchange and reuse response-level training signals across organizations. Selectively replaces locally incorrect responses with globally correct ones during public-data steps.

Result: Method consistently improves over standard baselines across mathematical and medical reasoning benchmarks and models. Demonstrates effectiveness in federated reasoning post-training scenarios.

Conclusion: Simple and effective recipe for federated reasoning post-training: combine low-rank communication (LoRA) with limited public-data coordination. Enables efficient cross-client alignment without exposing private data.

Abstract: Reasoning post-training with reinforcement learning from verifiable rewards (RLVR) is typically studied in centralized settings, yet many realistic applications involve decentralized private data distributed across organizations. Federated training is a natural solution, but scaling RLVR in this regime is challenging: full-model synchronization is expensive, and performing many local steps can cause severe client drift under heterogeneous data. We propose a federated RLVR framework that combines LoRA-based local adaptation with public-data-based off-policy steps to improve both communication efficiency and cross-client coordination. In particular, a small shared public dataset is used to periodically exchange and reuse response-level training signals across organizations, providing a lightweight anchor toward a more globally aligned objective without exposing private data. Our method selectively replaces locally incorrect responses with globally correct ones during public-data steps, thereby keeping training closer to the local policy while still benefiting from cross-client coordination. Across mathematical and medical reasoning benchmarks and models, our method consistently improves over standard baselines. Our results highlight a simple and effective recipe for federated reasoning post-training: combining low-rank communication with limited public-data coordination.

Method: Uses a TC structure-aware masked autoencoder to learn transferable representations from multi-modal data, coupled with a discrete probabilistic gridding mechanism and pre-train/fine-tune paradigm for simultaneous deterministic and probabilistic forecasting.

Result: Outperforms leading NWP systems in pressure and wind forecasting up to 120 hours and in track forecasting up to 24 hours across five global ocean basins. Attribution analysis shows physically interpretable learning dynamics.

Conclusion: Establishes a scalable, probabilistic, and interpretable pathway for operational TC forecasting through transferable multi-modal representation learning.

Abstract: Tropical cyclones (TCs) rank among the most destructive natural hazards, yet their forecasting faces fundamental trade-offs: numerical weather prediction (NWP) models are computationally prohibitive and struggle to leverage historical data, while existing deep learning (DL)-based intelligent models are variable-specific and deterministic, which fail to generalize across different forecasting variables. Here we present CycloneMAE, a scalable multi-task forecasting model that learns transferable TC representations from multi-modal data using a TC structure-aware masked autoencoder. By coupling a discrete probabilistic gridding mechanism with a pre-train/fine-tune paradigm, CycloneMAE simultaneously delivers deterministic forecasts and probability distributions. Evaluated across five global ocean basins, CycloneMAE outperforms leading NWP systems in pressure and wind forecasting up to 120 hours and in track forecasting up to 24 hours. Attribution analysis via integrated gradients reveals physically interpretable learning dynamics: short-term forecasts rely predominantly on the internal core convective structure from satellite imagery, whereas longer-term forecasts progressively shift attention to external environmental factors. Our framework establishes a scalable, probabilistic, and interpretable pathway for operational TC forecasting.

Method: Two-component framework: 1) Feature-based transfer learning module using spectral-transform-based feature alignment to project source/target domains into shared latent subspace and reduce distribution discrepancies; 2) Clustering enhancement strategy combining K-Medoids clustering with PCA-based dimensionality reduction to improve cross-domain correlation estimation.

Result: Method improves unknown attack detection accuracy by up to 49% compared to five baselines, achieves larger F-score gains, shows stronger stability, and clustering enhancement boosts accuracy by up to 26% on representative tasks.

Conclusion: The proposed method effectively addresses data scarcity and domain shift, providing a practical solution for robust cross-domain intrusion detection in dynamic industrial environments.

Abstract: Industrial control systems operate in dynamic environments where traffic distributions vary across scenarios, labeled samples are limited, and unknown attacks frequently emerge, posing significant challenges to cross-domain intrusion detection. To address this issue, this paper proposes a clustering-enhanced domain adaptation method for industrial control traffic. The framework contains two key components. First, a feature-based transfer learning module projects source and target domains into a shared latent subspace through spectral-transform-based feature alignment and iteratively reduces distribution discrepancies, enabling accurate cross-domain detection. Second, a clustering enhancement strategy combines K-Medoids clustering with PCA-based dimensionality reduction to improve cross-domain correlation estimation and reduce performance degradation caused by manual parameter tuning. Experimental results show that the proposed method significantly improves unknown attack detection. Compared with five baseline models, it increases detection accuracy by up to 49%, achieves larger gains in F-score, and demonstrates stronger stability. Moreover, the clustering enhancement strategy further boosts detection accuracy by up to 26% on representative tasks. These results suggest that the proposed method effectively alleviates data scarcity and domain shift, providing a practical solution for robust cross-domain intrusion detection in dynamic industrial environments.

Method: Develops a substantially tighter lower bound for ORC that is computationally efficient. The method extends beyond 1-hop random walks to k-hop random walks (k > 1), enabling practical speedups of tens of times compared to exact ORC computation.

Result: Experiments on fundamental graph structures demonstrate the effectiveness of the proposed bound in terms of both approximation accuracy and computational efficiency, achieving significant speed improvements while maintaining accuracy.

Conclusion: The proposed method provides a practical solution to the computational bottleneck of ORC, enabling broader applications by offering a tight lower bound that balances accuracy and efficiency for graph curvature analysis.

Abstract: Ollivier-Ricci curvature (ORC), defined via the Wasserstein distance that captures rich geometric information, has received growing attention in both theory and applications. However, the high computational cost of Wasserstein distance evaluation has significantly limited the broader practical use of ORC. To alleviate this issue, previous work introduced a computationally efficient lower bound as a proxy for ORC based on 1-hop random walks, but this approach empirically exhibits large gaps from the exact ORC. In this paper, we establish a substantially tighter lower bound for ORC than the existing lower bound, while retaining much lower computational cost than exact ORC computation, with practical speedups of tens of times. Moreover, our bound is not restricted to 1-hop random walks, but also applies to k-hop random walks (k > 1). Experiments on several fundamental graph structures demonstrate the effectiveness of our bound in terms of both approximation accuracy and computational efficiency.

Method: Benchmark study evaluating three categories: (1) classical log parsers (Drain, Spell, AEL) with ML classifiers, (2) fine-tuned transformers (BERT, RoBERTa), and (3) prompt-based LLMs (GPT-3.5, GPT-4, LLaMA-3) in zero-shot/few-shot settings across HDFS, BGL, Thunderbird, and Spirit datasets.

Result: Fine-tuned transformers achieve highest F1-scores (0.96-0.99), while prompt-based LLMs demonstrate strong zero-shot capabilities (F1: 0.82-0.91) without requiring labeled training data. Analysis includes cost-accuracy trade-offs, latency characteristics, and failure modes.

Conclusion: Provides actionable guidelines for practitioners choosing log anomaly detection methods based on accuracy, latency, cost, and label availability constraints. Fine-tuned transformers perform best but require labeled data, while LLMs offer viable zero-shot alternatives.

Abstract: System log anomaly detection is critical for maintaining the reliability of large-scale software systems, yet traditional methods struggle with the heterogeneous and evolving nature of modern log data. Recent advances in Large Language Models (LLMs) offer promising new approaches to log understanding, but a systematic comparison of LLM-based methods against established techniques remains lacking. In this paper, we present a comprehensive benchmark study evaluating both LLM-based and traditional approaches for log anomaly detection across four widely-used public datasets: HDFS, BGL, Thunderbird, and Spirit. We evaluate three categories of methods: (1) classical log parsers (Drain, Spell, AEL) combined with machine learning classifiers, (2) fine-tuned transformer models (BERT, RoBERTa), and (3) prompt-based LLM approaches (GPT-3.5, GPT-4, LLaMA-3) in zero-shot and few-shot settings. Our experiments reveal that while fine-tuned transformers achieve the highest F1-scores (0.96-0.99), prompt-based LLMs demonstrate remarkablezero-shot capabilities (F1: 0.82-0.91) without requiring any labeled training data – a significant advantage for real-world deployment where labeled anomalies are scarce. We further analyze the cost-accuracy trade-offs, latency characteristics, and failure modes of each approach. Our findings provide actionable guidelines for practitioners choosing log anomaly detection methods based on their specific constraints regarding accuracy, latency, cost, and label availability. All code and experimental configurations are publicly available to facilitate reproducibility.

Method: MolMem uses a multi-turn agentic RL framework with dual-memory system: (1) Static Exemplar Memory retrieves relevant exemplars for cold-start grounding, and (2) Evolving Skill Memory distills successful trajectories into reusable strategies. The policy is trained with dense step-wise rewards, turning costly rollouts into long-term knowledge.

Result: MolMem achieves 90% success on single-property tasks (1.5× over best baseline) and 52% on multi-property tasks using only 500 oracle calls.

Conclusion: The memory-augmented RL framework enables sample-efficient molecular optimization by leveraging dual memory systems to ground decisions and distill reusable strategies from expensive oracle evaluations.

Abstract: In drug discovery, molecular optimization aims to iteratively refine a lead compound to improve molecular properties while preserving structural similarity to the original molecule. However, each oracle evaluation is expensive, making sample efficiency a key challenge for existing methods under a limited oracle budget. Trial-and-error approaches require many oracle calls, while methods that leverage external knowledge tend to reuse familiar templates and struggle on challenging objectives. A key missing piece is long-term memory that can ground decisions and provide reusable insights for future optimizations. To address this, we present MolMem (\textbf{Mol}ecular optimization with \textbf{Mem}ory), a multi-turn agentic reinforcement learning (RL) framework with a dual-memory system. Specifically, MolMem uses Static Exemplar Memory to retrieve relevant exemplars for cold-start grounding, and Evolving Skill Memory to distill successful trajectories into reusable strategies. Built on this memory-augmented formulation, we train the policy with dense step-wise rewards, turning costly rollouts into long-term knowledge that improves future optimization. Extensive experiments show that MolMem achieves 90% success on single-property tasks (1.5$\times$ over the best baseline) and 52% on multi-property tasks using only 500 oracle calls. Our code is available at https://github.com/REAL-Lab-NU/MolMem.

Method: Proposes Socrates Loss, a novel unified loss function that explicitly leverages uncertainty by incorporating an auxiliary unknown class whose predictions directly influence the loss function and a dynamic uncertainty penalty. This unified objective optimizes both classification and confidence calibration simultaneously without complex scheduled losses.

Result: Across four benchmark datasets and multiple architectures, Socrates Loss consistently improves training stability while achieving more favorable accuracy-calibration trade-off, often converging faster than existing methods. Theoretical guarantees show the method regularizes models to prevent miscalibration and overfitting.

Conclusion: Socrates Loss effectively mitigates the stability-performance trade-off in confidence calibration, providing a unified solution that improves both classification performance and calibration reliability without training instability.

Abstract: Deep neural networks, despite their high accuracy, often exhibit poor confidence calibration, limiting their reliability in high-stakes applications. Current ad-hoc confidence calibration methods attempt to fix this during training but face a fundamental trade-off: two-phase training methods achieve strong classification performance at the cost of training instability and poorer confidence calibration, while single-loss methods are stable but underperform in classification. This paper addresses and mitigates this stability-performance trade-off. We propose Socrates Loss, a novel, unified loss function that explicitly leverages uncertainty by incorporating an auxiliary unknown class, whose predictions directly influence the loss function and a dynamic uncertainty penalty. This unified objective allows the model to be optimized for both classification and confidence calibration simultaneously, without the instability of complex, scheduled losses. We provide theoretical guarantees that our method regularizes the model to prevent miscalibration and overfitting. Across four benchmark datasets and multiple architectures, our comprehensive experiments demonstrate that Socrates Loss consistently improves training stability while achieving more favorable accuracy-calibration trade-off, often converging faster than existing methods.

Method: Proposes Metropolis-Hastings with Lévy jumps (MHLJ) which introduces occasional long-range transitions to restore exploration while respecting local information constraints. This addresses entrapment in weighted random-walk learning where the random walk may become trapped in small network regions.

Result: Establishes convergence rate characterizing roles of data heterogeneity, network spectral gap, and jump probability. Experiments show MHLJ effectively eliminates entrapment and significantly speeds up decentralized learning compared to standard weighted random-walk approaches.

Conclusion: Metropolis-Hastings with Lévy jumps solves the entrapment problem in weighted random-walk decentralized learning, enabling efficient exploration while maintaining local information constraints and improving convergence speed.

Abstract: We study decentralized learning over networks where data are distributed across nodes without a central coordinator. Random walk learning is a token-based approach in which a single model is propagated across the network and updated at each visited node using local data, thereby incurring low communication and computational overheads. In weighted random-walk learning, the transition matrix is designed to achieve a desired sampling distribution, thereby speeding up convergence under data heterogeneity. We show that implementing weighted sampling via the Metropolis-Hastings algorithm can lead to a previously unexplored phenomenon we term entrapment. The random walk may become trapped in a small region of the network, resulting in highly correlated updates and severely degraded convergence. To address this issue, we propose Metropolis-Hastings with Levy jumps, which introduces occasional long-range transitions to restore exploration while respecting local information constraints. We establish a convergence rate that explicitly characterizes the roles of data heterogeneity, network spectral gap, and jump probability, and demonstrate through experiments that MHLJ effectively eliminates entrapment and significantly speeds up decentralized learning.

Method: RoleMAG distinguishes whether a neighbor should provide shared, complementary, or heterophilous signals, and routes them through separate propagation channels, enabling cross-modal completion while keeping heterophilous signals out of shared smoothing.

Result: Extensive experiments on three graph-centric MAG benchmarks show RoleMAG achieves best results on RedditS and Bili_Dance while remaining competitive on Toys, with ablation, robustness, and efficiency analyses supporting the effectiveness.

Conclusion: RoleMAG’s role-aware propagation design effectively addresses the issue of neighbor interference across modalities in multimodal attributed graphs, improving performance on graph-centric multimodal tasks.

Abstract: Multimodal attributed graphs (MAGs) combine multimodal node attributes with structured relations. However, existing methods usually perform shared message passing on a single graph and implicitly assume that the same neighbors are equally useful for all modalities. In practice, neighbors that benefit one modality may interfere with another, blurring modality-specific signals under shared propagation. To address this issue, we propose RoleMAG, a multimodal graph framework that learns how different neighbors should participate in propagation. Concretely, RoleMAG distinguishes whether a neighbor should provide shared, complementary, or heterophilous signals, and routes them through separate propagation channels. This enables cross-modal completion from complementary neighbors while keeping heterophilous ones out of shared smoothing. Extensive experiments on three graph-centric MAG benchmarks show that RoleMAG achieves the best results on RedditS and Bili_Dance, while remaining competitive on Toys. Ablation, robustness, and efficiency analyses further support the effectiveness of the proposed role-aware propagation design. Our code is available at https://anonymous.4open.science/r/RoleMAG-7EE0/

Method: Proposes SubFlow (Sub-mode Conditioned Flow Matching) which decomposes each class into fine-grained sub-modes via semantic clustering and conditions the flow on sub-mode indices. Each conditioned sub-distribution is approximately unimodal, allowing accurate targeting of individual modes without averaging distortion.

Result: Extensive experiments on ImageNet-256 show SubFlow yields substantial gains in generation diversity (Recall) while maintaining competitive image quality (FID). It integrates seamlessly into existing one-step models like MeanFlow and Shortcut Models without architectural modifications.

Conclusion: SubFlow effectively addresses diversity degradation in flow matching models by eliminating averaging distortion through sub-mode conditioning, providing a plug-and-play solution that improves mode coverage while maintaining quality.

Abstract: Flow matching has emerged as a powerful generative framework, with recent few-step methods achieving remarkable inference acceleration. However, we identify a critical yet overlooked limitation: these models suffer from severe diversity degradation, concentrating samples on dominant modes while neglecting rare but valid variations of the target distribution. We trace this degradation to averaging distortion: when trained with MSE objectives, class-conditional flows learn a frequency-weighted mean over intra-class sub-modes, causing the model to over-represent high-density modes while systematically neglecting low-density ones. To address this, we propose SubFlow, Sub-mode Conditioned Flow Matching, which eliminates averaging distortion by decomposing each class into fine-grained sub-modes via semantic clustering and conditioning the flow on sub-mode indices. Each conditioned sub-distribution is approximately unimodal, so the learned flow accurately targets individual modes with no averaging distortion, restoring full mode coverage in a single inference step. Crucially, SubFlow is entirely plug-and-play: it integrates seamlessly into existing one-step models such as MeanFlow and Shortcut Models without any architectural modifications. Extensive experiments on ImageNet-256 demonstrate that SubFlow yields substantial gains in generation diversity (Recall) while maintaining competitive image quality (FID), confirming its broad applicability across different one-step generation frameworks. Project page: https://yexionglin.github.io/subflow.

Method: Proposes Shortcut Guardrail framework with key insight: gradient-based attribution on biased models highlights shortcut tokens. Uses lightweight LoRA-based debiasing module trained with Masked Contrastive Learning (MaskCL) objective that encourages consistent representations with or without individual tokens.

Result: Across sentiment classification, toxicity detection, and natural language inference tasks under both naturally occurring and controlled shortcuts, Shortcut Guardrail improves overall accuracy and worst-group accuracy over unmitigated models under distribution shifts while preserving in-distribution performance.

Conclusion: Shortcut Guardrail provides an effective deployment-time solution for mitigating token-level shortcuts without requiring original training data or shortcut annotations, making it practical for real-world applications.

Abstract: Pretrained language models often rely on superficial features that appear predictive during training yet fail to generalize at test time, a phenomenon known as shortcut learning. Existing mitigation methods generally operate at training time and require heavy supervision such as access to the original training data or prior knowledge of shortcut type. We propose Shortcut Guardrail, a deployment-time framework that mitigates token-level shortcuts without access to the original training data or shortcut annotations. Our key insight is that gradient-based attribution on a biased model highlights shortcut tokens. Building on this finding, we train a lightweight LoRA-based debiasing module with a Masked Contrastive Learning (MaskCL) objective that encourages consistent representations with or without individual tokens. Across sentiment classification, toxicity detection, and natural language inference under both naturally occurring and controlled shortcuts, Shortcut Guardrail improves overall accuracy and worst-group accuracy over the unmitigated model under distribution shifts while preserving in-distribution performance.

Method: 1) TrustSet: Selects most informative data from labeled dataset with balanced class distribution to mitigate long-tail problem, using label information to prune redundant data. 2) RL-based sampling policy: Approximates selection of high-quality TrustSet candidates from unlabeled data. 3) BRAL-T framework: Combines TrustSet and RL for batch active learning.

Result: Achieves state-of-the-art results across 10 image classification benchmarks and 2 active fine-tuning tasks, demonstrating effectiveness and efficiency in various domains.

Conclusion: BRAL-T effectively addresses limitations of traditional BAL methods by leveraging labeled data feedback and model performance, while handling class imbalance through TrustSet and efficiently selecting unlabeled data via RL-based sampling.

Abstract: Batch active learning (BAL) is a crucial technique for reducing labeling costs and improving data efficiency in training large-scale deep learning models. Traditional BAL methods often rely on metrics like Mahalanobis Distance to balance uncertainty and diversity when selecting data for annotation. However, these methods predominantly focus on the distribution of unlabeled data and fail to leverage feedback from labeled data or the model’s performance. To address these limitations, we introduce TrustSet, a novel approach that selects the most informative data from the labeled dataset, ensuring a balanced class distribution to mitigate the long-tail problem. Unlike CoreSet, which focuses on maintaining the overall data distribution, TrustSet optimizes the model’s performance by pruning redundant data and using label information to refine the selection process. To extend the benefits of TrustSet to the unlabeled pool, we propose a reinforcement learning (RL)-based sampling policy that approximates the selection of high-quality TrustSet candidates from the unlabeled data. Combining TrustSet and RL, we introduce the Batch Reinforcement Active Learning with TrustSet (BRAL-T) framework. BRAL-T achieves state-of-the-art results across 10 image classification benchmarks and 2 active fine-tuning tasks, demonstrating its effectiveness and efficiency in various domains.

Method: Compared Multilayer Perceptron (MLP) using only weather features against Long Short-Term Memory (LSTM) using 24-step (2-hour) sliding consumption windows on 14 months of 5-minute smart meter data from two Melbourne households (one standard, one with rooftop solar), combined with daily weather observations.

Result: LSTM achieved R² = 0.883 (standard house) and R² = 0.865 (solar house), while weather-only MLPs performed poorly with R² = -0.055 and R² = 0.410 respectively, showing temporal autocorrelation dominates by 93.8 and 45.5 percentage point differences.

Conclusion: Temporal autocorrelation in consumption sequences is far more important than meteorological information for short-term 5-minute forecasting, with solar generation creating forecasting asymmetries where weather features implicitly capture solar patterns.

Abstract: Accurate short-term residential energy consumption forecasting at sub-hourly resolution is critical for smart grid management, demand response programmes, and renewable energy integration. While weather variables are widely acknowledged as key drivers of residential electricity demand, the relative merit of incorporating temporal autocorrelation - the sequential memory of past consumption; over static meteorological features alone remains underexplored at fine-grained (5-minute) temporal resolution for Australian households. This paper presents a rigorous empirical comparison of a Multilayer Perceptron (MLP) and a Long Short-Term Memory (LSTM) recurrent network applied to two real-world Melbourne households: House 3 (a standard grid-connected dwelling) and House 4 (a rooftop solar photovoltaic-integrated household). Both models are trained on 14 months of 5-minute interval smart meter data (March 2023-April 2024) merged with official Bureau of Meteorology (BOM) daily weather observations, yielding over 117,000 samples per household. The LSTM, operating on 24-step (2-hour) sliding consumption windows, achieves coefficients of determination of R^2 = 0.883 (House 3) and R^2 = 0.865 (House 4), compared to R^2 = -0.055 and R^2 = 0.410 for the corresponding weather-driven MLPs - differences of 93.8 and 45.5 percentage points. These results establish that temporal autocorrelation in the consumption sequence dominates meteorological information for short-term forecasting at 5-minute granularity. Additionally, we demonstrate an asymmetry introduced by solar generation: for the PV-integrated household, the MLP achieves R^2 = 0.410, revealing implicit solar forecasting from weather-time correlations. A persistence baseline analysis and seasonal stratification contextualise model performance. We propose a hybrid weather-augmented LSTM and federated learning extensions as directions for future work.

Method: Developed GCA framework with two components: (1) GCA-DS - curated multimodal dataset with ~200k QA pairs covering policies, frameworks, literature, event reporting, plus remote-sensing imagery with textual evidence; (2) Gulf Climate Agent - tool-augmented agent with modular pipeline for real-time/historical signal processing, geospatial analysis, and visualization generation.

Result: Benchmarking shows domain fine-tuning and tool integration substantially improve reliability over general-purpose LLM baselines for Gulf climate tasks.

Conclusion: The GCA framework addresses critical gaps in region-specific climate knowledge and tool interaction for LLMs, providing a foundation for actionable climate guidance in the Gulf region.

Abstract: Climate decision-making in the Gulf increasingly demands systems that can translate heterogeneous scientific and policy evidence into actionable guidance, yet general-purpose large language models (LLMs) remain weak both in region-specific climate knowledge and grounded interaction with geospatial and forecasting tools. We present the GCA framework, which unifies (i) GCA-DS, a curated Gulf-focused multimodal dataset, and (ii) Gulf Climate Agent (GCA), a tool-augmented agent for climate analysis. GCA-DS comprises ~200k question-answer pairs spanning governmental policies and adaptation plans, NGO and international frameworks, academic literature, and event-driven reporting on heatwaves, dust storms, and floods, complemented with remote-sensing inputs that couple imagery with textual evidence. Building on this foundation, the GCA agent orchestrates a modular tool pipeline grounded in real-time and historical signals and geospatial processing that produces derived indices and interpretable visualizations. Finally, we benchmark open and proprietary LLMs on Gulf climate tasks and show that domain fine-tuning and tool integration substantially improve reliability over general-purpose baselines.

Method: Proposes OptBias: Surrogate Learning with Optimization Bias via Synthetic Task Generation. Uses meta-learning to learn reusable optimization bias from synthetic tasks generated from Gaussian processes, then fine-tunes on small target task data.

Result: Outperforms state-of-the-art baselines across diverse continuous and discrete offline optimization benchmarks in small data regimes.

Conclusion: OptBias provides a robust and practical solution for offline optimization in realistic small data settings by effectively learning and transferring optimization bias.

Abstract: We consider the problem of offline black-box optimization, where the goal is to discover optimal designs (e.g., molecules or materials) from past experimental data. A key challenge in this setting is data scarcity: in many scientific applications, only small or poor-quality datasets are available, which severely limits the effectiveness of existing algorithms. Prior work has theoretically and empirically shown that performance of offline optimization algorithms depends on how well the surrogate model captures the optimization bias (i.e., ability to rank input designs correctly), which is challenging to accomplish with limited experimental data. This paper proposes Surrogate Learning with Optimization Bias via Synthetic Task Generation (OptBias), a meta-learning framework that directly tackles data scarcity. OptBias learns a reusable optimization bias by training on synthetic tasks generated from a Gaussian process, and then fine-tunes the surrogate model on the small data for the target task. Across diverse continuous and discrete offline optimization benchmarks, OptBias consistently outperforms state-of-the-art baselines in small data regimes. These results highlight OptBias as a robust and practical solution for offline optimization in realistic small data settings.

Method: Used DistilBERT, RoBERTa, and Llama 2 to classify gender of de-gendered LoRs from a U.S. medical residency program. Applied TF-IDF and SHAP for text interpretation to identify gender-proxy linguistic patterns. Experimented with removing implicit gender cues to create truly neutral LoRs.

Result: Models achieved up to 68% classification accuracy on anonymized LoRs. Linguistic patterns like “emotional” and “humanitarian” were strong gender proxies. Removing implicit cues reduced accuracy by up to 5.5% and F1 by 2.7%, but gender prediction remained above chance.

Conclusion: LoRs contain hard-to-remove gender-identifying cues that may activate bias. Technical interventions help but don’t eliminate gender leakage, necessitating upstream auditing of evaluative text alongside model-level fairness approaches.

Abstract: Letters of recommendation (LoRs) can carry patterns of implicitly gendered language that can inadvertently influence downstream decisions, e.g. in hiring and admissions. In this work, we investigate the extent to which Transformer-based encoder models as well as Large Language Models (LLMs) can infer the gender of applicants in academic LoRs submitted to an U.S. medical-residency program after explicit identifiers like names and pronouns are de-gendered. While using three models (DistilBERT, RoBERTa, and Llama 2) to classify the gender of anonymized and de-gendered LoRs, significant gender leakage was observed as evident from up to 68% classification accuracy. Text interpretation methods, like TF-IDF and SHAP, demonstrate that certain linguistic patterns are strong proxies for gender, e.g. “emotional’’ and “humanitarian’’ are commonly associated with LoRs from female applicants. As an experiment in creating truly gender-neutral LoRs, these implicit gender cues were remove resulting in a drop of up to 5.5% accuracy and 2.7% macro $F_1$ score on re-training the classifiers. However, applicant gender prediction still remains better than chance. In this case study, our findings highlight that 1) LoRs contain gender-identifying cues that are hard to remove and may activate bias in decision-making and 2) while our technical framework may be a concrete step toward fairer academic and professional evaluations, future work is needed to interrogate the role that gender plays in LoR review. Taken together, our findings motivate upstream auditing of evaluative text in real-world academic letters of recommendation as a necessary complement to model-level fairness interventions.

Method: PEV integrates three key components: (1) adaptive checkpointing to retain critical historical updates for fast reconstruction, (2) layer adaptive differentially private calibration to selectively remove client influence while minimizing accuracy loss, and (3) fingerprint-based verification enabling decentralized confirmation of unlearning.

Result: Experiments on image, handwritten character, and medical datasets show PEV achieves 2-3× faster unlearning than retraining, provides formal indistinguishability guarantees with reduced performance degradation, and supports scalable verification.

Conclusion: PEV is the first framework to simultaneously deliver efficiency, privacy, and verifiability for federated unlearning, moving FL closer to practical and regulation-compliant deployment.

Abstract: Federated learning (FL) enables collaborative model training without sharing raw data, offering a promising path toward privacy preserving artificial intelligence. However, FL models may still memorize sensitive information from participants, conflicting with the right to be forgotten (RTBF). To meet these requirements, federated unlearning has emerged as a mechanism to remove the contribution of departing clients. Existing solutions only partially address this challenge: FedEraser improves efficiency but lacks privacy protection, FedRecovery ensures differential privacy (DP) but degrades accuracy, and VeriFi enables verifiability but introduces overhead without efficiency or privacy guarantees. We present PrivEraserVerify (PEV), a unified framework that integrates efficiency, privacy, and verifiability into federated unlearning. PEV employs (i) adaptive checkpointing to retain critical historical updates for fast reconstruction, (ii) layer adaptive differentially private calibration to selectively remove client influence while minimizing accuracy loss, and (iii) fingerprint based verification, enabling participants to confirm unlearning in a decentralized and noninvasive manner. Experiments on image, handwritten character, and medical datasets show that PEV achieves up to 2 to 3 times faster unlearning than retraining, provides formal indistinguishability guarantees with reduced performance degradation, and supports scalable verification. To the best of our knowledge, PEV is the first framework to simultaneously deliver efficiency, privacy, and verifiability for federated unlearning, moving FL closer to practical and regulation compliant deployment.

Method: SCPT constructs similarity-constrained triplets (scaffold, better, worse) via scaffold alignment and chemistry-driven filters for validity, synthesizability, and meaningful property gains. These preferences align a pretrained molecular LLM as a conditional editor for scaffold-preserving property improvements.

Result: SCPT improves optimization success and property gains while maintaining higher scaffold similarity than baselines. It outperforms non-LLM methods in scaffold-constrained and multi-objective optimization, and shows effective generalization from single/two-property to three-property tasks.

Conclusion: SCPT enables controlled molecular optimization with predictable similarity-gain trade-offs, offering systematic adaptation to diverse optimization regimes while preserving molecular scaffolds through chemistry-grounded preference alignment.

Abstract: Molecular property optimization is central to drug discovery, yet many deep learning methods rely on black-box scoring and offer limited control over scaffold preservation, often producing unstable or biologically implausible edits. While large language models (LLMs) are promising molecular generators, optimization remains constrained by the lack of chemistry-grounded preference supervision and principled data curation. We introduce \textbf{Scaffold-Conditioned Preference Triplets (SCPT)}, a pipeline that constructs similarity-constrained triplets $\langle\text{scaffold}, \text{better}, \text{worse}\rangle$ via scaffold alignment and chemistry-driven filters for validity, synthesizability, and meaningful property gains. Using these preferences, we align a pretrained molecular LLM as a conditional editor, enabling property-improving edits that retain the scaffold. Across single- and multi-objective benchmarks, SCPT improves optimization success and property gains while maintaining higher scaffold similarity than competitive baselines. Compared with representative non-LLM molecular optimization methods, SCPT-trained LLMs are better suited to scaffold-constrained and multi-objective optimization. In addition, models trained on single-property and two-property supervision generalize effectively to three-property tasks, indicating promising extrapolative generalization under limited higher-order supervision. SCPT also provides controllable data-construction knobs that yield a predictable similarity-gain frontier, enabling systematic adaptation to diverse optimization regimes.

Method: The paper presents a complete end-to-end framework implementing industrial-style long-sequence training with sliding windows. Key contributions include: (1) runtime-aware ablation study quantifying accuracy-compute trade-offs across windowing regimes and strides, and (2) a novel k-shift embedding layer enabling million-scale vocabularies on commodity GPUs with minimal accuracy loss.

Result: The framework trains reliably on modest university clusters and delivers competitive retrieval quality (up to +6.04% MRR and +6.34% Recall@10 on Retailrocket) with approximately 4× training-time overheads compared to standard approaches.

Conclusion: By providing a robust pipeline, reporting training time costs, and introducing an embedding mechanism for low-resource settings, the authors transform long-sequence training from a closed industrial technique into a practical, open, and extensible methodology for the research community.

Abstract: Long interaction histories are central to modern recommender systems, yet training with long sequences is often dismissed as impractical under realistic memory and latency budgets. This work demonstrates that it is not only practical but also effective-at academic scale. We release a complete, end-to-end framework that implements industrial-style long-sequence training with sliding windows, including all data processing, training, and evaluation scripts. Beyond reproducing prior gains, we contribute two capabilities missing from earlier reports: (i) a runtime-aware ablation study that quantifies the accuracy-compute frontier across windowing regimes and strides, and (ii) a novel k-shift embedding layer that enables million-scale vocabularies on commodity GPUs with negligible accuracy loss. Our implementation trains reliably on modest university clusters while delivering competitive retrieval quality (e.g., up to +6.04% MRR and +6.34% Recall@10 on Retailrocket) with $\sim 4 \times $ training-time overheads. By packaging a robust pipeline, reporting training time costs, and introducing an embedding mechanism tailored for low-resource settings, we transform long-sequence training from a closed, industrial technique into a practical, open, and extensible methodology for the community.

Method: Pre-trained a 120B parameter hybrid Mamba-Attention Mixture-of-Experts model on 25T tokens using NVFP4 precision, incorporating LatentMoE architecture for parameter efficiency and MTP layers for speculative decoding acceleration, followed by SFT and RL post-training.

Result: Achieves comparable accuracy on benchmarks with 1M context length, while delivering 2.2x higher inference throughput than GPT-OSS-120B and 7.5x higher than Qwen3.5-122B. Model checkpoints and datasets are open-sourced.

Conclusion: Nemotron 3 Super demonstrates that architectural innovations like hybrid Mamba-Attention and LatentMoE can enable efficient large-scale models with high throughput while maintaining competitive performance.

Abstract: We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemotron 3 Super is the first model in the Nemotron 3 family to 1) be pre-trained in NVFP4, 2) leverage LatentMoE, a new Mixture-of-Experts architecture that optimizes for both accuracy per FLOP and accuracy per parameter, and 3) include MTP layers for inference acceleration through native speculative decoding. We pre-trained Nemotron 3 Super on 25 trillion tokens followed by post-training using supervised fine tuning (SFT) and reinforcement learning (RL). The final model supports up to 1M context length and achieves comparable accuracy on common benchmarks, while also achieving up to 2.2x and 7.5x higher inference throughput compared to GPT-OSS-120B and Qwen3.5-122B, respectively. Nemotron 3 Super datasets, along with the base, post-trained, and quantized checkpoints, are open-sourced on HuggingFace.

Method: Proposes a post-hoc self-supervised method that trains a decoder to forecast the second half of observed trajectories from the first half. Uses the L2 norm of the gradient of this forecasting loss with respect to the decoder’s final layer as a score to identify distribution shifts.

Result: Demonstrates substantial improvements on distribution shift detection for trajectory prediction on Shifts and Argoverse datasets. Also shows the method can early detect collisions of a deep Q-Network motion planner in the Highway simulator.

Conclusion: The proposed self-supervised approach effectively detects distribution shifts without interfering with original prediction performance, enhancing safety in automated driving systems.

Abstract: Trajectory prediction models often fail in real-world automated driving due to distributional shifts between training and test conditions. Such distributional shifts, whether behavioural or environmental, pose a critical risk by causing the model to make incorrect forecasts in unfamiliar situations. We propose a self-supervised method that trains a decoder in a post-hoc fashion on the self-supervised task of forecasting the second half of observed trajectories from the first half. The L2 norm of the gradient of this forecasting loss with respect to the decoder’s final layer defines a score to identify distribution shifts. Our approach, first, does not affect the trajectory prediction model, ensuring no interference with original prediction performance and second, demonstrates substantial improvements on distribution shift detection for trajectory prediction on the Shifts and Argoverse datasets. Moreover, we show that this method can also be used to early detect collisions of a deep Q-Network motion planner in the Highway simulator. Source code is available at https://github.com/Michedev/forecasting-the-past.

Method: Used controlled multi-hop relational reasoning tasks based on family stories where difficulty scales with relationship hops. Monitored prediction evolution across layers via early readouts (logit lens) and task-relevant information integration via causal patching across tokens. Compared pretrained models with models finetuned on the task.

Result: Pretrained models showed limited evidence for adaptive depth use - some larger models used fewer layers for easier tasks, and models generally used more layers to integrate information as chain length increased. Finetuned models showed clearer and more consistent adaptive depth use, with stronger effects in less constrained finetuning regimes that didn’t preserve general language modeling abilities.

Conclusion: Transformers can learn to use their depth adaptively when finetuned on specific reasoning tasks, but this capability is limited in general-purpose pretrained models. The degree of adaptive depth use depends on the finetuning regime, with more task-specific training enabling better adaptation to task difficulty.

Abstract: We investigate whether transformers use their depth adaptively across tasks of increasing difficulty. Using a controlled multi-hop relational reasoning task based on family stories, where difficulty is determined by the number of relationship hops that must be composed, we monitor (i) how predictions evolve across layers via early readouts (the logit lens) and (ii) how task-relevant information is integrated across tokens via causal patching. For pretrained models, we find some limited evidence for adaptive depth use: some larger models need fewer layers to arrive at plausible answers for easier tasks, and models generally use more layers to integrate information across tokens as chain length increases. For models finetuned on the task, we find clearer and more consistent evidence of adaptive depth use, with the effect being stronger for less constrained finetuning regimes that do not preserve general language modeling abilities.

Method: Systematically study noise impact across three pretrained model families (GPT-2, Qwen2, Llama-2) and three diverse NLP tasks. Introduce controlled perturbations for three common real-world noise types: label noise, grammatical noise, and typographical noise. Analyze layer-wise representation changes and attention patterns to understand noise propagation.

Result: Label noise causes largest performance degradation, while grammatical and typographical noise can occasionally yield mild regularization benefits. Noise effects are localized primarily to task-specific layers, while attention structures remain comparatively stable.

Conclusion: Different noise types affect LLM fine-tuning differently, with label noise being most harmful. Understanding internal learning dynamics under noise conditions provides insights for developing more robust fine-tuning approaches.

Abstract: Fine-tuning is the dominant paradigm for adapting pretrained large language models (LLMs) to downstream NLP tasks. In practice, fine-tuning datasets may contain various forms of noise arising from annotation errors, preprocessing artifacts, or automated data collection. While prior work has focused on designing robust learning algorithms to mitigate performance degradation under noisy conditions, comparatively little is known about how different types of noise affect the internal learning dynamics of LLMs during fine-tuning. In this work, we systematically study the impact of noise on model behavior across three pretrained model families (GPT-2, Qwen2 and Llama-2) and three diverse NLP tasks. We introduce controlled perturbations corresponding to three common real-world noise types: label noise, grammatical noise, and typographical noise. Beyond task-level performance, we analyze layer-wise representation changes and attention patterns to understand how noise propagates through the network. Our results show that corrupting labels (i.e. label noise) consistently causes the largest performance degradation, whereas grammatical noise and typographical noise can occasionally yield mild regularization benefits. We further find that noise effects are localized primarily to task-specific layers, while attention structures remain comparatively stable.

Method: Proposes an adaptive allocation algorithm that learns which questions are hardest for the LLM while simultaneously collecting human responses. Each human label serves dual roles: improving estimates for that question and revealing how well the LLM predicts human responses on it. The algorithm directs more budget to questions where LLM is least reliable, without requiring prior knowledge of question-level LLM accuracy.

Result: On real survey data with 68 questions and over 2000 respondents, the algorithm reduces waste from 10-12% (uniform allocation) to 2-6%. The advantage grows as questions become more heterogeneous in LLM prediction quality. Achieves same estimation quality as uniform sampling with fewer human samples, requires no pilot study, and has formal performance guarantees.

Conclusion: The framework provides an efficient solution for allocating scarce human oversight across tasks where LLM reliability is unknown, with applications beyond surveys to any scenario requiring human verification of LLM outputs.

Abstract: Large language models (LLMs) can generate survey responses at low cost, but their reliability varies substantially across questions and is unknown before data collection. Deploying LLMs in surveys still requires costly human responses for verification and correction. How should a limited human-labeling budget be allocated across questions in real time? We propose an adaptive allocation algorithm that learns which questions are hardest for the LLM while simultaneously collecting human responses. Each human label serves a dual role: it improves the estimate for that question and reveals how well the LLM predicts human responses on it. The algorithm directs more budget to questions where the LLM is least reliable, without requiring any prior knowledge of question-level LLM accuracy. We prove that the allocation gap relative to the best possible allocation vanishes as the budget grows, and validate the approach on both synthetic data and a real survey dataset with 68 questions and over 2000 respondents. On real survey data, the standard practice of allocating human labels uniformly across questions wastes 10–12% of the budget relative to the optimal; our algorithm reduces this waste to 2–6%, and the advantage grows as questions become more heterogeneous in LLM prediction quality. The algorithm achieves the same estimation quality as traditional uniform sampling with fewer human samples, requires no pilot study, and is backed by formal performance guarantees validated on real survey data. More broadly, the framework applies whenever scarce human oversight must be allocated across tasks where LLM reliability is unknown.

Method: Trained 11 instruction-tuned LLMs (0.5B-14B parameters) with on-policy RL across 3 different environments, conducted controlled ablations to trace effects to environment-specific features like role framing and implicit gameability cues.

Result: Model size acts as a safety buffer in some environments but enables greater harmful exploitation in others; safety benchmarks generally don’t predict RL-induced misalignment except for Sycophancy scores when exploits rely on inferring user preferences; on-policy RL preserves safety buffers inherent in model generation distributions.

Conclusion: The relationship between model size and safety is complex and environment-dependent; current safety benchmarks are inadequate for predicting RL-induced misalignment; on-policy RL preserves inherent safety mechanisms that off-policy methods bypass.

Abstract: Specification gaming under Reinforcement Learning (RL) is known to cause LLMs to develop sycophantic, manipulative, or deceptive behavior, yet the conditions under which this occurs remain unclear. We train 11 instruction-tuned LLMs (0.5B–14B) with on-policy RL across 3 environments and find that model size acts as a safety buffer in some environments but enables greater harmful exploitation in others. Controlled ablations trace this reversal to environment-specific features such as role framing and implicit gameability cues. We further show that most safety benchmarks do not predict RL-induced misalignment, except in the case of Sycophancy scores when the exploit relies on inferring the user’s preference. Finally, we find that on-policy RL preserves a safety buffer inherent in the model’s own generation distribution, one that is bypassed during off-policy settings.

Method: Combines local variational hidden computation (EVE), a homeostatic latent regulator, structurally aware checkpoint retention, and a calibrated uncertainty-aware controller that operates on retained models. Treats uncertainty as an operational signal that regulates training, supports checkpoint retention, and guides inference-time intervention.

Result: The variational backbone outperforms a matched deterministic reference on language modeling while exhibiting richer uncertainty profiles. The calibrated controller remains active, uses multiple actions under full agentic evaluation, and yields positive quality-cost trade-offs.

Conclusion: Internal uncertainty in variational language models can serve not only as a descriptive property but also as a practical control interface for regulation, checkpoint retention, and minimal agentic routing, supporting closed-loop internal control.

Abstract: We study whether a variational language model can support a minimal and measurable form of agentic control grounded in its own internal evidence. Our model combines local variational hidden computation (EVE), a homeostatic latent regulator, structurally aware checkpoint retention and a calibrated uncertainty-aware controller operating on top of the retained model. Rather than treating uncertainty as a passive diagnostic measured after prediction, we treat it as an operational signal that can regulate training, support checkpoint retention and guide inference-time intervention. The resulting framework is deliberately focused. It studies a closed-loop form of internal control in which structural and predictive signals become actionable. Empirically, the variational backbone improves over a matched deterministic reference on the language-modeling task while also exhibiting a richer and more usable uncertainty profile. On top of this backbone, the calibrated controller remains active, uses multiple actions under a full agentic evaluation and yields a positive quality-cost trade-off. These results support a precise claim: internal uncertainty can serve not only as a descriptive property of a variational language model, but also as a practical control interface for regulation, checkpoint retention and minimal agentic routing.

Method: Instantiates generalized-Fano framework using squared Hellinger distance to compare hard vs reference models, combining lower bounds on reference hinge terms with model distinguishability bounds.

Result: Derived explicit Bayesian CVaR lower bounds for canonical examples including Gaussian bandits, making dependence on key problem parameters transparent.

Conclusion: The generalized-Fano Bayesian CVaR framework can serve as a practical lower-bound tool for interactive learning and risk-sensitive decision making applications.

Abstract: Recent work established a generalized-Fano framework for lower bounding prior-predictive (Bayesian) CVaR in interactive statistical decision making. In this paper, we show how to instantiate that framework in concrete interactive problems and derive explicit Bayesian CVaR lower bounds from its abstract corollaries. Our approach compares a hard model with a reference model using squared Hellinger distance, and combines a lower bound on a reference hinge term with a bound on the distinguishability of the two models. We apply this approach to canonical examples, including Gaussian bandits, and obtain explicit bounds that make the dependence on key problem parameters transparent. These results show how the generalized-Fano Bayesian CVaR framework can be used as a practical lower-bound tool for interactive learning and risk-sensitive decision making.

Method: Uses Singular Value Decomposition (SVD)-guided orthogonal subspace projection to constrain each new LoRA update during training to lie in the orthogonal complement of subspaces used by earlier unlearning tasks, enabling task isolation without dynamic routing at deployment.

Result: After thirty sequential unlearning tasks on CIFAR-100 with ResNet-20 and MNIST, the method maintains baseline performance (~58.1%) while preserving strong unlearning efficacy, compared to state-of-the-art static fusion which reduces retained accuracy from 60.39% to 12.70%.

Conclusion: SVD-guided orthogonal subspace projection effectively addresses parameter collision in continual machine unlearning with LoRA, providing stable performance across long sequences of unlearning tasks while maintaining task isolation and unlearning efficacy.

Abstract: Continual machine unlearning aims to remove the influence of data that should no longer be retained, while preserving the usefulness of the model on everything else. This setting becomes especially difficult when deletion requests arrive sequentially, because the model must repeatedly adapt without erasing previously retained knowledge. Low-Rank Adaptation (LoRA) offers an efficient way to implement such updates, but naively combining many sequential LoRA modules leads to parameter collision, causing \textit{strong interference} between tasks. We propose a static alternative based on Singular Value Decomposition (SVD)-guided orthogonal subspace projection. Our method constrains each new LoRA update during training so that it lies in the orthogonal complement of the subspaces used by earlier unlearning tasks. This preserves task isolation without requiring dynamic routing at deployment. Experiments on CIFAR-100 with ResNet-20 and on MNIST show stable behavior across long sequences of unlearning tasks. After thirty sequential unlearning tasks, state-of-the-art static fusion reduces retained accuracy from 60.39% to 12.70%, whereas the proposed in-training constrained optimization maintains baseline performance ($\sim$58.1%) while preserving strong unlearning efficacy.

Method: Proposes EEG-MoCE framework with learnable-curvature hyperbolic spaces for each modality (experts). Uses curvature-aware fusion strategy to dynamically weight experts based on hierarchical information richness. Each modality gets its own hyperbolic space with adaptive curvature learning.

Result: Achieves state-of-the-art performance on benchmark datasets for emotion recognition, sleep staging, and cognitive assessment tasks.

Conclusion: Hyperbolic mixture-of-curvature experts framework effectively models hierarchical structures in multimodal EEG data, outperforming previous methods on various mental state assessment tasks.

Abstract: Electroencephalography (EEG)-based multimodal learning integrates brain signals with complementary modalities to improve mental state assessment, providing great clinical potential. The effectiveness of such paradigms largely depends on the representation learning on heterogeneous modalities. For EEG-based paradigms, one promising approach is to leverage their hierarchical structures, as recent studies have shown that both EEG and associated modalities (e.g., facial expressions) exhibit hierarchical structures reflecting complex cognitive processes. However, Euclidean embeddings struggle to represent these hierarchical structures due to their flat geometry, while hyperbolic spaces, with their exponential growth property, are naturally suited for them. In this work, we propose EEG-MoCE, a novel hyperbolic mixture-of-curvature experts framework designed for multimodal neurotechnology. EEG-MoCE assigns each modality to an expert in a learnable-curvature hyperbolic space, enabling adaptive modeling of its intrinsic geometry. A curvature-aware fusion strategy then dynamically weights experts, emphasizing modalities with richer hierarchical information. Extensive experiments on benchmark datasets demonstrate that EEG-MoCE achieves state-of-the-art performance, including emotion recognition, sleep staging, and cognitive assessment.

Method: KumoRFM-2 processes relational data natively without flattening, using early task information injection to select relevant columns. It’s pre-trained on synthetic and real-world data across four axes: row/column dimensions at table level, and foreign key/cross-sample dimensions at database level. The model supports both in-context learning and fine-tuning.

Result: Outperforms supervised and foundational approaches by up to 8% on 41 challenging benchmarks, maintains strong performance under cold start and noisy data conditions, and scales to billion-scale relational datasets. First few-shot foundation model to surpass supervised approaches on common benchmark tasks.

Conclusion: KumoRFM-2 represents a significant advancement in relational data foundation models, demonstrating superior performance over supervised methods while offering scalability and robustness to challenging data conditions.

Abstract: We introduce KumoRFM-2, the next iteration of a pre-trained foundation model for relational data. KumoRFM-2 supports in-context learning as well as fine-tuning and is applicable to a wide range of predictive tasks. In contrast to tabular foundation models, KumoRFM-2 natively operates on relational data, processing one or more connected tables simultaneously without manual table flattening or target variable generation, all while preserving temporal consistency. KumoRFM-2 leverages a large corpus of synthetic and real-world data to pre-train across four axes: the row and column dimensions at the individual table level, and the foreign key and cross-sample dimensions at the database level. In contrast to its predecessor, KumoRFM-2 injects task information as early as possible, enabling sharper selection of task-relevant columns and improved robustness to noisy data. Through extensive experiments on 41 challenging benchmarks and analysis around expressivity and sensitivity, we demonstrate that KumoRFM-2 outperforms supervised and foundational approaches by up to 8%, while maintaining strong performance under extreme settings of cold start and noisy data. To our knowledge, this is the first time a few-shot foundation model has been shown to surpass supervised approaches on common benchmark tasks, with performance further improving upon fine-tuning. Finally, while KumoRFM-1 was limited to small-scale in-memory datasets, KumoRFM-2 scales to billion-scale relational datasets.

Method: SOAR performs single stop-gradient rollouts from real samples, re-noises the resulting off-trajectory states, and supervises the model to steer back toward original clean targets. This provides dense per-timestep supervision without reward models or credit assignment issues.

Result: On SD3.5-Medium, SOAR improves GenEval from 0.70 to 0.78 and OCR from 0.64 to 0.67 over SFT, while raising all model-based preference scores. It surpasses Flow-GRPO on aesthetic and text-image alignment tasks despite having no reward model access.

Conclusion: SOAR effectively bridges the SFT-RL gap in diffusion model post-training, providing a stronger first stage that can replace SFT while remaining compatible with subsequent RL alignment, addressing exposure bias without reward model dependencies.

Abstract: The post-training pipeline for diffusion models currently has two stages: supervised fine-tuning (SFT) on curated data and reinforcement learning (RL) with reward models. A fundamental gap separates them. SFT optimizes the denoiser only on ground-truth states sampled from the forward noising process; once inference deviates from these ideal states, subsequent denoising relies on out-of-distribution generalization rather than learned correction, exhibiting the same exposure bias that afflicts autoregressive models, but accumulated along the denoising trajectory instead of the token sequence. RL can in principle address this mismatch, yet its terminal reward signal is sparse, suffers from credit-assignment difficulty, and risks reward hacking. We propose SOAR (Self-Correction for Optimal Alignment and Refinement), a bias-correction post-training method that fills this gap. Starting from a real sample, SOAR performs a single stop-gradient rollout with the current model, re-noises the resulting off-trajectory state, and supervises the model to steer back toward the original clean target. The method is on-policy, reward-free, and provides dense per-timestep supervision with no credit-assignment problem. On SD3.5-Medium, SOAR improves GenEval from 0.70 to 0.78 and OCR from 0.64 to 0.67 over SFT, while simultaneously raising all model-based preference scores. In controlled reward-specific experiments, SOAR surpasses Flow-GRPO in final metric value on both aesthetic and text-image alignment tasks, despite having no access to a reward model. Since SOAR’s base loss subsumes the standard SFT objective, it can directly replace SFT as a stronger first post-training stage after pretraining, while remaining fully compatible with subsequent RL alignment.

Method: Proposes Calibration-Aware Policy Optimization (CAPO) with: 1) logistic AUC surrogate loss for theoretically consistent uncertainty-aware advantage estimation, and 2) noise masking mechanism for stable joint optimization of calibration and accuracy.

Result: CAPO-1.5B improves calibration by up to 15% while maintaining comparable/better accuracy than GRPO, boosts downstream inference-time scaling accuracy by up to 5%, and achieves Pareto-optimal precision-coverage trade-off when allowed to abstain under low confidence.

Conclusion: CAPO effectively addresses the calibration-accuracy trade-off in LLM reasoning, offering practical value for hallucination mitigation through better uncertainty estimation.

Abstract: Group Relative Policy Optimization (GRPO) enhances LLM reasoning but often induces overconfidence, where incorrect responses yield lower perplexity than correct ones, degrading relative calibration as described by the Area Under the Curve (AUC). Existing approaches either yield limited improvements in calibration or sacrifice gains in reasoning accuracy. We first prove that this degradation in GRPO-style algorithms stems from their uncertainty-agnostic advantage estimation, which inevitably misaligns optimization gradients with calibration. This leads to improved accuracy at the expense of degraded calibration. We then propose Calibration-Aware Policy Optimization (CAPO). It adopts a logistic AUC surrogate loss that is theoretically consistent and admits regret bound, enabling uncertainty-aware advantage estimation. By further incorporating a noise masking mechanism, CAPO achieves stable learning dynamics that jointly optimize calibration and accuracy. Experiments on multiple mathematical reasoning benchmarks show that CAPO-1.5B significantly improves calibration by up to 15% while achieving accuracy comparable to or better than GRPO, and further boosts accuracy on downstream inference-time scaling tasks by up to 5%. Moreover, when allowed to abstain under low-confidence conditions, CAPO achieves a Pareto-optimal precision-coverage trade-off, highlighting its practical value for hallucination mitigation.

Method: Proposes TimeSAF with hierarchical asynchronous fusion: 1) Independent cross-modal semantic fusion trunk using learnable queries to aggregate global semantics from temporal and prompt backbones bottom-up, 2) Stage-wise semantic refinement decoder that asynchronously injects high-level signals back into temporal backbone, decoupling unimodal feature learning from cross-modal interaction.

Result: Extensive experiments on standard long-term forecasting benchmarks show TimeSAF significantly outperforms state-of-the-art baselines and exhibits strong generalization in few-shot and zero-shot transfer settings.

Conclusion: Hierarchical asynchronous fusion effectively addresses semantic perceptual dissonance in multimodal time-series forecasting, providing stable semantic guidance without interfering with low-level temporal dynamics.

Abstract: Despite the recent success of large language models (LLMs) in time-series forecasting, most existing methods still adopt a Deep Synchronous Fusion strategy, where dense interactions between textual and temporal features are enforced at every layer of the network. This design overlooks the inherent granularity mismatch between modalities and leads to what we term semantic perceptual dissonance: high-level abstract semantics provided by the LLM become inappropriately entangled with the low-level, fine-grained numerical dynamics of time series, making it difficult for semantic priors to effectively guide forecasting. To address this issue, we propose TimeSAF, a new framework based on hierarchical asynchronous fusion. Unlike synchronous approaches, TimeSAF explicitly decouples unimodal feature learning from cross-modal interaction. It introduces an independent cross-modal semantic fusion trunk, which uses learnable queries to aggregate global semantics from the temporal and prompt backbones in a bottom-up manner, and a stage-wise semantic refinement decoder that asynchronously injects these high-level signals back into the temporal backbone. This mechanism provides stable and efficient semantic guidance while avoiding interference with low-level temporal dynamics. Extensive experiments on standard long-term forecasting benchmarks show that TimeSAF significantly outperforms state-of-the-art baselines, and further exhibits strong generalization in both few-shot and zero-shot transfer settings.

Method: Combines supervised learning with consistency regularization, confidence-aware pseudo-labeling, and selective temporal invariance to conservatively exploit unlabeled traffic while suppressing unreliable samples. Leverages temporal structure of network flows for improved robustness.

Result: Extensive evaluations on CIC-IDS2017, CSE-CIC-IDS2018, and UNSW-NB15 datasets under limited-label conditions show consistent outperformance of state-of-the-art supervised and semi-supervised NIDS in detection performance, label efficiency, and resilience to adversarial/non-stationary traffic.

Conclusion: The proposed framework effectively addresses key challenges in real-world cloud intrusion detection by robustly handling adversarial contamination and temporal drift in unlabeled traffic through temporal-aware semi-supervised learning.

Abstract: Cloud networks increasingly rely on machine learning based Network Intrusion Detection Systems to defend against evolving cyber threats. However, real-world deployments are challenged by limited labeled data, non-stationary traffic, and adaptive adversaries. While semi-supervised learning can alleviate label scarcity, most existing approaches implicitly assume benign and stationary unlabeled traffic, leading to degraded performance in adversarial cloud environments. This paper proposes a robust semi-supervised temporal learning framework for cloud intrusion detection that explicitly addresses adversarial contamination and temporal drift in unlabeled network traffic. Operating on flow-level data, this framework combines supervised learning with consistency regularization, confidence-aware pseudo-labeling, and selective temporal invariance to conservatively exploit unlabeled traffic while suppressing unreliable samples. By leveraging the temporal structure of network flows, the proposed method improves robustness and generalization across heterogeneous cloud environments. Extensive evaluations on publicly available datasets (CIC-IDS2017, CSE-CIC-IDS2018, and UNSW-NB15) under limited-label conditions demonstrate that the proposed framework consistently outperforms state-of-the-art supervised and semi-supervised network intrusion detection systems in detection performance, label efficiency, and resilience to adversarial and non-stationary traffic.

Method: Constructed a multi-scale candlestick charts dataset with standardized evaluation framework combining confusion-matrix-based diagnostics with information coefficient (IC) time series metrics. Included XGBoost as a feature-based temporal baseline. Benchmarked representative VLMs on their ability to leverage multi-scale stock price data.

Result: Most VLMs perform well only under persistent uptrend or downtrend conditions but exhibit weak predictive capability in more common market scenarios. Identified significant prediction biases and limited sensitivity to explicitly specified forecast horizons in prompts, indicating inherent limitations in precise temporal reasoning.

Conclusion: Current VLMs have fundamental limitations in understanding multi-scale visual market dynamics and temporal reasoning for stock price forecasting, despite being applied to this domain. The proposed dataset and framework reveal these weaknesses and provide a better evaluation standard.

Abstract: Vision-language models(VLMs) are increasingly applied to visual stock price forecasting, yet existing benchmarks inadequately evaluate their understanding of stock price in candlestick charts. First, prior studies fail to isolate VLMs’ comprehension of visual inputs genuinely improves predictive performance and whether VLMs truly comprehend candlestick patterns. Further, most existing datasets and evaluation setups are designed around single-period or tabular inputs. However, human analysts strongly rely on multi-scale candlestick charts, where longer-term horizons capture trend direction and shorter-term horizons provide cues for inflection points, making it difficult to systematically assess VLMs’ ability to integrate short-term and long-term visual market dynamics. To bridge this gap, we construct a multi-scale candlestick charts dataset and a standardized evaluation framework to assess VLMs’ ability to utilize multi-scale visual market signals. Evaluation combines confusion-matrix-based diagnostics with information coefficient(IC) time series metrics and includes XGBoost as a feature-based temporal baseline. Using this dataset, we benchmark representative VLMs and analyze their ability to leverage multi-scale stock price data. Experimental results show that most VLMs perform well only under persistent uptrend or downtrend conditions, while exhibiting weak predictive capability in more common market scenarios. We also identify significant prediction biases and limited sensitivity to explicitly specified forecast horizons in prompts, indicating inherent limitations in precise temporal reasoning.

Method: Created Triton dataset (590k instances) via Structural-Semantic Hard Negative Mining and Dual-Agent Consensus pipeline, then progressive curriculum training producing three models: Triton-SFT-32B (basic imitation), Triton-ORPO-32B (robust discrimination), and Triton-GRPO-32B (long-horizon consistency).

Result: Triton-GRPO-32B achieves 58.7% Step Success Rate on Mind2Web, surpassing GPT-4.5 (42.4%) and Claude-4.5 (41.4%) by over 16%.

Conclusion: Specialized data curriculum outweighs raw parameter scale for web navigation, demonstrating that robust training approaches can outperform much larger general-purpose models.

Abstract: Text-based web agents offer computational efficiency for autonomous web navigation, yet developing robust agents remains challenging due to the noisy and heterogeneous nature of real-world HTML. Standard Supervised Fine-Tuning (SFT) approaches fail in two critical dimensions: they lack discrimination capabilities to reject plausible but incorrect elements in densely populated pages, and exhibit limited generalization to unseen website layouts. To address these challenges, we introduce the Triton dataset (590k instances) and a progressive training curriculum. Triton is constructed via Structural-Semantic Hard Negative Mining, which explicitly mines topologically similar distractors, and a Dual-Agent Consensus pipeline that synthesizes diverse cross-domain tasks with strict verification. Building upon this foundation, our progressive curriculum produces three models: Triton-SFT-32B for basic imitation, Triton-ORPO-32B for robust discrimination via Odds Ratio Preference Optimization, and Triton-GRPO-32B for long-horizon consistency through Group Relative Policy Optimization. Empirical evaluation on Mind2Web demonstrates that Triton-GRPO-32B achieves state-of-the-art performance among open-source models with 58.7% Step Success Rate, surpassing GPT-4.5 (42.4%) and Claude-4.5 (41.4%) by over 16%, validating that specialized data curriculum outweighs raw parameter scale for web navigation.

Method: Proposes Bi-Directional Low-Rank Adaptation (BID-LoRA) with three dedicated adapter pathways (retain, new, and unlearn) applied to attention layers, combined with escape unlearning that pushes forget-class embeddings to positions maximally distant from retained knowledge, updating only 5% of parameters.

Result: BID-LoRA outperforms CLU baselines across multiple adaptation cycles on CIFAR-100. Also evaluated on CASIA-Face100, demonstrating practical applicability to real-world identity management systems where users must be enrolled and withdrawn users removed.

Conclusion: The paper successfully formalizes Continual Learning Unlearning (CLU) as a unified paradigm and demonstrates BID-LoRA’s effectiveness in achieving precise deletion of unwanted knowledge, efficient integration of new knowledge, and minimizing knowledge leakage across cycles.

Abstract: Recent advances in deep learning underscore the need for systems that can not only acquire new knowledge through Continual Learning (CL) but also remove outdated, sensitive, or private information through Machine Unlearning (MU). However, while CL methods are well-developed, MU techniques remain in early stages, creating a critical gap for unified frameworks that depend on both capabilities. We find that naively combining existing CL and MU approaches results in knowledge leakage a gradual degradation of foundational knowledge across repeated adaptation cycles. To address this, we formalize Continual Learning Unlearning (CLU) as a unified paradigm with three key goals: (i) precise deletion of unwanted knowledge, (ii) efficient integration of new knowledge while preserving prior information, and (iii) minimizing knowledge leakage across cycles. We propose Bi-Directional Low-Rank Adaptation (BID-LoRA), a novel framework featuring three dedicated adapter pathways-retain, new, and unlearn applied to attention layers, combined with escape unlearning that pushes forget-class embeddings to positions maximally distant from retained knowledge, updating only 5% of parameters. Experiments on CIFAR-100 show that BID-LoRA outperforms CLU baselines across multiple adaptation cycles. We further evaluate on CASIA-Face100, a curated face recognition subset, demonstrating practical applicability to real-world identity management systems where new users must be enrolled and withdrawn users removed.

Method: Formalize task-adapted CS-MRI optimization by maximizing mutual information between undersampled k-space measurements and clinical tasks; use amortized optimization and construct tractable variational bounds for mutual information to jointly optimize sampling, reconstruction, and task-inference models

Result: Achieves highly competitive performance on standard metrics like Dice compared to deterministic counterparts, provides better distribution matching to ground-truth posterior distribution as measured by generalized energy distance (GED)

Conclusion: Proposed framework addresses uncertainty problem in medical diagnosis, enables flexible sampling ratio control with single end-to-end model, and handles both joint task/reconstruction scenarios and privacy-protected task-only scenarios

Abstract: Task-adapted compressed sensing magnetic resonance imaging (CS-MRI) is emerging to address the specific demands of downstream clinical tasks with significantly fewer k-space measurements than required by Nyquist sampling. However, existing task-adapted CS-MRI methods suffer from the uncertainty problem for medical diagnosis and cannot achieve adaptive sampling in end-to-end optimization with reconstruction or clinical tasks. To address these limitations, we propose the first task-adapted CS-MRI from the information-theoretic perspective to simultaneously achieve probabilistic inference for uncertainty prediction and adapt to arbitrary sampling ratios and versatile clinical applications. Specifically, we formalize the task-adapted CS-MRI optimization problem by maximizing the mutual information between undersampled k-space measurements and clinical tasks to enable probabilistic inference for addressing the uncertainty problem. We leverage amortized optimization and construct tractable variational bounds for mutual information to jointly optimize sampling, reconstruction, and task-inference models, which enables flexible sampling ratio control using a single end-to-end trained model. Furthermore, the proposed framework addresses two kinds of distinct clinical scenarios within a unified approach, i.e., i) joint task and reconstruction, where reconstruction serves as an auxiliary process to enhance task performance; and ii) task implementation with suppressed reconstruction, applicable for privacy protection. Extensive experiments on large-scale MRI datasets demonstrate that the proposed framework achieves highly competitive performance on standard metrics like Dice compared to deterministic counterpart but provides better distribution matching to the ground-truth posterior distribution as measured by the generalized energy distance (GED).

Method: Identifies semantic bottlenecks in LLMs (intermediate layers where representations are governed by semantic content rather than language identity), then proposes LASA which anchors safety alignment directly in these semantic bottlenecks.

Result: LASA substantially improves safety across all languages: average attack success rate drops from 24.7% to 2.8% on LLaMA-3.1-8B-Instruct and remains around 3-4% across Qwen2.5 and Qwen3 Instruct models (7B-32B).

Conclusion: Safety alignment should be anchored in language-agnostic semantic space rather than surface text, offering a representation-level perspective on LLM safety that addresses cross-lingual vulnerabilities.

Abstract: Large language models (LLMs) often demonstrate strong safety performance in high-resource languages, yet exhibit severe vulnerabilities when queried in low-resource languages. We attribute this gap to a mismatch between language-agnostic semantic understanding ability and language-dominant safety alignment biased toward high-resource languages. Consistent with this hypothesis, we empirically identify the semantic bottleneck in LLMs, an intermediate layer in which the geometry of model representations is governed primarily by shared semantic content rather than language identity. Building on this observation, we propose Language-Agnostic Semantic Alignment (LASA), which anchors safety alignment directly in semantic bottlenecks. Experiments show that LASA substantially improves safety across all languages: average attack success rate (ASR) drops from 24.7% to 2.8% on LLaMA-3.1-8B-Instruct and remains around 3-4% across Qwen2.5 and Qwen3 Instruct models (7B-32B). Together, our analysis and method offer a representation-level perspective on LLM safety, suggesting that safety alignment requires anchoring safety understanding not in surface text, but in the model’s language-agnostic semantic space.

Method: Systematic investigation of OPD dynamics through weak-to-strong reverse distillation experiments, token-level mechanism analysis, and proposing practical recovery strategies like off-policy cold start and teacher-aligned prompt selection.

Result: Identified two governing conditions for OPD success: compatible thinking patterns and teacher offering new capabilities. Found that successful OPD involves progressive alignment on high-probability tokens at student-visited states. Proposed recovery strategies for failing OPD.

Conclusion: OPD’s apparent benefits come with costs, raising scalability questions for long-horizon distillation. Understanding OPD dynamics enables better distillation strategies and reveals limitations of current approaches.

Abstract: On-policy distillation (OPD) has become a core technique in the post-training of large language models, yet its training dynamics remain poorly understood. This paper provides a systematic investigation of OPD dynamics and mechanisms. We first identify that two conditions govern whether OPD succeeds or fails: (i) the student and teacher should share compatible thinking patterns; and (ii) even with consistent thinking patterns and higher scores, the teacher must offer genuinely new capabilities beyond what the student has seen during training. We validate these findings through weak-to-strong reverse distillation, showing that same-family 1.5B and 7B teachers are distributionally indistinguishable from the student’s perspective. Probing into the token-level mechanism, we show that successful OPD is characterized by progressive alignment on high-probability tokens at student-visited states, a small shared token set that concentrates most of the probability mass (97%-99%). We further propose two practical strategies to recover failing OPD: off-policy cold start and teacher-aligned prompt selection. Finally, we show that OPD’s apparent free lunch of dense token-level reward comes at a cost, raising the question of whether OPD can scale to long-horizon distillation.