Reproduce “Near-Optimal $\sqrt{K}$-Regret for Online Learning in Uninformed Markov Games” 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

Junyan Liu, Haipeng Luo, Lillian J. Ratliff, Zihan Zhang

The authors develop a method for uninformed Markov games where learners see losses and next states but not opponents’ actions. It combines global weights over future-loss predictions with state-specific policy mixtures to align feedback with executed losses. It guarantees empirical Nash-value regret \widetilde O(H^{3/2}S\sqrt{AK}) against arbitrary adaptive opponents with probability at least 1−δ. Its candidate class is exponentially large.

Summary by alphaXiv.

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