Reproduce “Stochastic Runge-Kutta Accelerates Langevin Monte Carlo and Beyond” 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
Xuechen Li, Denny Wu, Lester Mackey, Murat A. Erdogdu
The paper introduces SRK-LD, a Runge–Kutta sampler using intermediate gradients to reduce discretization error. W2 is the minimum root mean squared distance for paired samples from two distributions. Under strong convexity and four-times differentiable potentials, it achieves ε accuracy in W2 in ~O(dε^-2/3) iterations, versus ~O(dε^-1) for a baseline, with two extra gradient evaluations per iteration; experiments do not establish universal speedups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.