Reproduce “A Stochastic Analysis Perspective on Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Stanislas Strasman
The authors explain score-based generation as reversal of a forward diffusion’s path law, rather than tracking fixed-time distributions. Under regularity assumptions, the reverse drift combines the negated forward drift with the noise covariance times the score. In a two-cluster example, this reversal matches the target more closely than frozen Langevin, with exact scores and terminal initialization, and positive-time stopping.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.