Reproduce “Shifted Composition IV: Toward Ballistic Acceleration for 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
Jason M. Altschuler, Sinho Chewi, Matthew S. Zhang
The randomized-midpoint underdamped Langevin method (RM–ULMC) samples high-dimensional target distributions by using momentum and an analysis that controls numerical error in KL divergence without requiring Wasserstein contractivity. Under a Poincaré inequality, smooth convex potential, and low friction, its iteration bound scales as κ^{5/6}, with stated dimension and initialization factors. This is sublinear, not square-root κ acceleration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.