Reproduce “Improved sampling via learned diffusions” 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

Lorenz Richter, Julius Berner

The authors unify learned diffusion samplers in a path-space framework and replace reverse KL with a log-variance objective that compares stochastic trajectories without differentiating through them. On a five-dimensional double-well target, Path Integral Sampler log-normalizer estimation error falls from 3.567 to 0.214, while normalized effective sample size rises from 0.0004 to 0.6744. Results remain task-dependent, with exceptions on other tests.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.