Multi-Agent System Optimization
Overview
Multi-agent system optimization studies how to organize, train, and coordinate multiple agents so they solve tasks more effectively than they would alone. Common approaches optimize workflows and communication topologies, route tasks to suitable agents, refine prompts, or use reinforcement learning, evolutionary search, and game-theoretic methods to improve behavior and coordination. Work spans language-model agent teams, multi-agent reinforcement learning, robotics, scheduling, and distributed resource allocation, with evaluation on reasoning and coding benchmarks as well as simulations and deployed systems. A central challenge is balancing accuracy and robust coordination against token, compute, communication, and latency costs, while ensuring gains generalize beyond particular models, tasks, and evaluation setups.
Latest papers
161 papers16 papers in the last four weeks, up 45% on the four weeks before. 0.1% of all new papers.