Reproduce “Model-Based Multi-Agent RL in Zero-Sum Markov Games with Near-Optimal Sample Complexity” 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

Kaiqing Zhang, Sham M. Kakade, Tamer Başar, Lin F. Yang

The authors study model-based learning in discounted two-player zero-sum Markov games, estimating transitions from a generative model before planning equilibrium policies. With reward withheld until after sampling, their transition-sample bound is Õ(|S||A||B|(1−γ)⁻³ε⁻²), matching the minimax lower bound up to logarithmic factors. The policy guarantee requires a smooth planning oracle; with known rewards, the action-count dependence is not shown to be minimax-optimal.

Summary by alphaXiv.

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