Research map / Multi-agent systems and games / Games and equilibria

Learning Large-Scale Competitive Team Behaviors with Mean-Field Interactions

NeurIPS 2026 · Multi-agent systems and games › Games and equilibria · approach: Mean-field game learning

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a mean-field MARL approach for learning competitive team behaviors among large numbers of agents, using population-distribution interactions instead of pairwise ones.

Problem

Multi-agent reinforcement learning struggles with large numbers of agents because joint interactions grow combinatorially; existing mean-field formulations focus on fully cooperative or single-team settings rather than competitive team behaviors.

Approach

Extends mean-field approximations to competitive multi-team settings, where each agent interacts with the population distributions of teams rather than with individual agents, yielding tractable large-population policies.

Experimental setup

Builds on: mean-field games mean-field MARL

Closest papers by meaning

• Learning in Stackelberg Mean Field Games: A Non-Asymptotic Analysis (NeurIPS 2025 · similarity 0.89)
• Bench-MFG: A Benchmark Suite for Learning in Stationary Mean Field Games (NeurIPS 2026 · similarity 0.89)
• Stochastic Semi-Gradient Descent for Learning Mean Field Games with Population-Aware Function Approximation (ICLR 2025 · similarity 0.89)
• Mean-Field Sampling for Cooperative Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
• Team-Fictitious Play for Reaching Team-Nash Equilibrium in Multi-team Games (NeurIPS 2024 · similarity 0.88)
• Finite-Sample Convergence Bounds for Trust Region Policy Optimization in Mean Field Games (ICML 2025 · similarity 0.88)