Reproduce “Gaussian Limits for SGD Without Stationary Moments” 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
Xiaoli Li, Wei Biao Wu
The authors analyze constant-step least-squares SGD with persistent designs. In a Gaussian-refresh example, retaining each covariate with probability rho yields infinite stationary error moments of every positive order for 0<a<2/3, despite a Gaussian small-step limit. For independent Gaussian errors, they derive a first-order total-variation correction under stronger design conditions, not a result for arbitrary dependent SGD.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.