Reproduce “Noise Shapes the Surplus: Strategic Play Against No-Regret Learners” 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
Yiheng Su, Emmanouil-Vasileios Vlatakis-Gkaragkounis
The authors analyze how an optimizer can earn strategic surplus against a no-regret learner whose FTRL updates use noisy payoff scores and a decaying learning rate. They distinguish selection, which exploits a profitable limiting mean response, from movement, which exploits changes in the learner’s response. At η_t proportional to t^-1/2, selection guarantees Ω(√T) expected surplus under centered finite-variance noise, provided its conditions hold.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.