Reproduce “Control Variates for Stochastic Gradient MCMC” 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
Jack Baker, Paul Fearnhead, Emily B. Fox, Christopher Nemeth
SGLD-CV reduces gradient noise by computing a full-data gradient at a center and estimating minibatch changes from it. Under strong log-concavity and smoothness, with an accurate center, its sampling work can be bounded independently of dataset size, excluding linear-cost setup. In cover-type logistic regression, it reaches similar held-out predictive performance sooner than SGLD as data grows, while SAGA performs better.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.