Reproduce “Denoising Is Stein: Moment Access and a Resource Hierarchy for Learning Hidden Structure in 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
Rui Wang
The authors present a two-stage learner using fresh noisy views and smoothed denoising curvature to find a hidden direction in covariance-matched data. It sharpens the estimate with moment-based updates. For a symmetric mixture whose first distinguishing moment is fourth, it achieves fixed score gain with O(d^2 log d) clean points. A width-8 network shows a d^2-like threshold over d=64–192, but its asymptotic exponent and mechanism remain unresolved.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.