Optimal Coarse Correlated Equilibria in Mean Field Games: Linear Programming and No-Regret Learning

2026-06-18 · Source: Machine Learning · Field: Science & Research — Mathematics & Computational Sciences, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

This paper introduces optimal coarse correlated equilibria (OCCE) for continuous-time mean field games, a novel concept where a moderator selects an equilibrium to optimize a specific performance criterion, potentially distinct from individual player objectives. A coarse correlated equilibrium is defined as a randomized recommendation scheme from which no player benefits by deviating. The authors develop a linear programming (LP) formulation for this problem, proving the existence of optimal LP coarse correlated equilibria and establishing its relationship to the original probabilistic framework. Building on this characterization, they design a no-regret primal-dual algorithm, utilizing a Lagrangian formulation of the external-regret constraint, for learning these equilibria. The work includes explicit convergence rates for the algorithm and supporting numerical examples.

Key takeaway

For research scientists developing multi-agent systems or economic models, this work offers a robust framework for achieving system-wide objectives in mean field games. You should consider integrating optimal coarse correlated equilibria, utilizing the proposed linear programming formulation to define desired outcomes and the no-regret primal-dual algorithm for practical implementation and learning. This approach provides a mechanism to guide decentralized agents towards a global optimum, even when individual incentives differ.

Key insights

Optimal coarse correlated equilibria in mean field games can be found via linear programming and learned using a no-regret primal-dual algorithm.

Principles

• Moderators can optimize CCEs.
• LP characterizes optimal CCEs.
• No-regret learning finds equilibria.

Method

The method involves formulating the problem as a linear program, proving existence, and then designing a no-regret primal-dual algorithm based on a Lagrangian formulation of external-regret constraints to learn the equilibria.

In practice

• Apply LP to find optimal CCEs.
• Implement no-regret learning for MFG.

Topics

• Mean Field Games
• Coarse Correlated Equilibria
• Linear Programming
• No-Regret Learning
• Primal-Dual Algorithms
• Multi-Agent Systems

Best for: AI Scientist, Research Scientist

Related on AIssential

• Maximum Entropy Inverse Reinforcement Learning for Mean-Field Games with Average Reward — Recovering unknown reward policies in average-reward MFGs is feasible using maximum causal entropy and a novel contraction method.
• Regret Minimization with Adaptive Opponents in Repeated Games — Repeated Policy Regret (RP-Regret) enables finding higher-utility equilibria in repeated games by modeling adaptive opponents and dynamic strategies.
• A New Lower Bound for the Random Offerer Mechanism in Bilateral Trade using AI-Guided Evolutionary Search — AI-guided search found a new lower bound for the Random-Offerer mechanism's worst-case efficiency.
• Coordination Graphs for Constrained Multi-Agent Reinforcement Learning — The CG-CMARL framework efficiently scales multi-agent reinforcement learning by combining coordination graphs and Lagrangian duality for constraint handling.
• Enhancing Decision-Making with Large Language Models through Multi-Agent Fictitious Play — Multi-Agent Fictitious Play (MAFP) resolves interdependent decision-making by modeling stakeholder stances as agents in an equilibrium-seeking process.
• Convex Markov Games and Beyond: New Proof of Existence, Characterization and Learning Algorithms for Nash Equilibria — Nash equilibria in General Utility Markov Games coincide with fixed points of projected pseudo-gradient dynamics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.