Reproduce “Measuring Sample Quality with 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
Jackson Gorham, Andrew B. Duncan, Sebastian J. Vollmer, Lester Mackey
The authors introduce a sample-quality discrepancy from an Itô diffusion whose stationary distribution is the target. It uses the target’s log-density gradient without its unknown normalization constant; under fast-coupling and regularity conditions, its relation to Wasserstein distance controls error. In a Student’s t regression, it selects step size 10^-4, matching a surrogate reference metric, while ESS selects 10^-2; this does not certify arbitrary targets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.