Reproduce “Exact Theory and Optimal Stepsizes for Stochastic Gradient Descent in Smooth Convex Minimization” 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

Benjamin Grimmer, Qining Shen

The paper gives an exact worst-case bound for the expected final objective gap of plain stochastic gradient descent (SGD) with fixed short steps. The bound is attained when normalized step sizes lie in [0,1.5] and dimension is at least N+1. It also proves the unique minimax-optimal schedule decreases under a stated variance threshold and dimension condition.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.