Reproduce “Constant-Stepsize Stochastic Approximation: Finite-Time Convergence, Gaussian Approximation, and Tail Bounds” 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
Zedong Wang, Yuyang Wang, Ijay Narang, Felix Wang, Yuzhou Wang, Siva Theja Maguluri
The authors analyze constant-stepsize stochastic approximation before idealized limits. They establish a unique stationary law with geometric Wasserstein convergence under contractive conditions, and bound the normalized stationary iterate’s Wasserstein-1 distance from a Gaussian by U√αlog(1/α) for sufficiently small α. For flat one-dimensional minima, simulations support a scale matched to flatness, while the Gibbs approximation remains conditional on conjectures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.