Reproduce “Game Theoretic Rating in N-player general-sum games with Equilibria” 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
Luke Marris, Marc Lanctot, Ian Gemp, Shayegan Omidshafiei, Stephen McAleer, Jerome Connor, Karl Tuyls, Thore Graepel
This paper develops a game-theoretic payoff rating system for N-player, general-sum normal-form games, leveraging maximum entropy Coarse-Correlated Equilibria (MECCE). The approach accurately captures complex strategic relationships, including non-transitive cycles, assigning equal ratings to cyclically related strategies and providing robust evaluation for real-world sports data and multi-agent reinforcement learning policies.
Key insights
- Game-theoretic payoff ratings based on Maximum Entropy Coarse-Correlated Equilibria (MECCE) effectively capture coordination and strategic cycles in N-player general-sum games.
- The "grouping property" of MECCE ratings assigns similar values to strategies involved in a strategic cycle, even if their raw win/loss records differ, reflecting their true strategic equivalence.
- Using the Maximum Entropy criterion ensures full-support solutions, providing well-defined payoff ratings for all strategies, including those that might appear dominated.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.