Reproduce “Learned Reference-based Diffusion Sampling for multi-modal distributions” 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
Maxence Noble, Louis Grenioux, Marylou Gabrié, Alain Oliviero Durmus
The authors present LRDS, a diffusion sampler for multimodal distributions with known mode locations but unknown normalizing constants. It learns a reference from MCMC samples initialized at those modes and uses the reference to train a sampler. On a 64-dimensional mixture, GMM-LRDS estimates the stronger-mode weight with 4.1% ± 0.6% error, but performance declines at higher dimensions and with more modes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.