Reproduce “Wasserstein Gradient Flows and Forward-Only Diffusion Are Not Enough for Multimodal Sampling” 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
Daniel McBride, Pratik Khandagale, Cristina Garcia-Cardona, Yen Ting Lin
The authors analyze slow sampling from separated modes by showing that canonical KL-induced Wasserstein gradient flow and overdamped forward diffusion share the same evolving marginal density under stated assumptions, though not sample paths. In a two-mode Gaussian-mixture study, spectral-gap and mean-first-passage estimates agree from separation 2d ≈ 0.87; predicted times exceed 10^7 in an illustrated case, not universally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.