Reproduce “A proximal gradient algorithm for composite log-concave 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
Linghai Liu, Sinho Chewi
The composite sampler draws from distributions with a smooth potential and a possibly nonsmooth convex term, using a restricted Gaussian oracle for the latter and an independent Metropolis–Hastings correction for a gradient-informed proposal. Under stated assumptions, it achieves ε total-variation error with Õ(κ√d log⁴(1/ε)) oracle evaluations. In a Gaussian-box experiment, it reaches sliced Wasserstein-2 distance 0.05 at all tested dimensions, 4 to 512; this evidence is limited to that target.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.