Reproduce “Sampling Decisions: Exact Path-Space Control for Physics-Informed Generative 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

Michael Chertkov, Hamidreza Behjoo, Sungsoo Ahn

Sampling Decisions constructs Gibbs-target samples by correcting sequential proposals with path-dependent importance weights and a look-ahead Doob h-transform. Its finite-path estimator approximates this population correction using weighted paths. On one exactly enumerable ferromagnetic 5×5 Ising instance, Local-Boltzmann guidance retains 0.591 of the path budget in effective sample size, versus 0.008 for uniform and BP proposals; this is not a runtime or scaling result.

Summary by alphaXiv.

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