Reproduce “Error accumulation in reverse diffusion with an imperfect score: beyond path-space analyses” 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
George Giapitzakis, Ankur Moitra, Andrej Risteski, Dhruv Rohatgi
The authors analyze error accumulation in learned reverse diffusion by tracking final-sample statistics rather than entire paths. Under a Poincaré condition and a uniform score-error bound, their total variation guarantee has logarithmic dependence on the Poincaré constant, plus initialization error and a squared-error horizon term. A multiscale-Gaussian construction shows this dependence cannot generally be removed; the result concerns the reverse SDE.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.