Reproduce “Rennala NSGD: Asynchronous Stochastic Optimization Beyond Euclidean Geometry” 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
Igor Sokolov, Alexander Tyurin, Peter Richtárik
The authors analyze geometry-matched noise bounds for optimization and extend Rennala’s semi-asynchronous protocol beyond Euclidean updates. They measure noise in the update’s dual norm, establishing an oracle-evaluation guarantee under smoothness and light-tailed noise. For dense Rademacher noise with an ℓ1 update, the statistical coefficient grows as ln(ed), rather than d; this is a theoretical comparison, not a speedup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.