Multi-Agent Systems

Summary

Multi-Agent Systems (MAS) in AI research focus on the interactions and decision-making processes of multiple autonomous agents in shared environments. Recent advancements in this field cover a wide range of topics, including equilibrium refinements in Multi-Agent Influence Diagrams (MAIDs), complex control in multiplayer games, transparent opponent learning, multi-agent negotiations, and decentralized reinforcement learning. These studies explore various aspects of MAS, such as graphical model representations, deep reinforcement learning techniques, self-play for policy improvement, and economic transaction-based decision-making. The research aims to develop sophisticated algorithms and frameworks that can handle complex state and action spaces, anticipate opponent behaviors, and achieve cooperative or competitive outcomes in diverse scenarios like game theory problems, traffic simulations, and decentralized problem-solving. These advancements contribute to the development of more robust, adaptive, and efficient multi-agent systems capable of addressing real-world challenges in areas such as autonomous driving, strategic planning, and distributed decision-making.

Research Papers

• Equilibrium Refinements for Multi-Agent Influence Diagrams Theory and Practice
• Mastering Complex Control in MOBA Games with Deep Reinforcement Learning
• Learning in two-player games between transparent opponents
• Towards Learning Multi-agent Negotiations via Self-Play
• Decentralized Reinforcement Learning Global Decision-Making via Local Economic Transactions