Reproduce “MCMC-Correction of Score-Based Diffusion Models for Model Composition” 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
Anders Sjöberg, Jakob Lindqvist, Magnus Önnheim, Mats Jirstrand, Lennart Svensson
The authors extend corrected MCMC sampling to score-based diffusion models by estimating energy differences through path integrals of the model’s score field and using them in an accept/reject step. On CIFAR-100, corrected score HMC reports FID 13.3, versus 28.6 for unadjusted HMC and 12.4 for corrected energy HMC; nonconservative scores mean this approximation does not guarantee detailed balance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.