Reproduce “Convergence Guarantees for Federated SARSA with Local Training and Heterogeneous Agents” 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
Paul Mangold, Eloïse Berthier, Eric Moulines
The authors analyze FedSARSA, in which agents perform local SARSA updates under a shared policy before periodically averaging parameters. They prove convergence to a self-consistent fixed point defined by heterogeneous agents, not necessarily optimal for each or equivalent to an averaged environment. In a moderate-heterogeneity, high-precision regime, leading per-agent sample complexity scales as 1/(N epsilon squared), though other error terms can dominate.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.