Diffusion Model Notes

The latent state has the same dimension as the input.
The noise added in each step is a Gaussian that is only dependent on the current state and the previous state. $$q(x_t|x_{t-1}) = \mathcal{N}(x_t; \sqrt{1-\beta_t}x_{t-1}, \beta_t I)$$ where $$\beta_t$$ is set by the user. It controls how fast we add noise to the latent state.
There are different kins of noise scheduler.