Multi-agent systems lie at the heart of future intelligent learning systems. At IIS Corp., we are investigating various issues that arise in multi-agent learning, such as: 

• How can several agents working in parallel on similar tasks cooperate to speed up their learning.
• What coordination mechanisms are necessary when several agents are working in the same domain and are sharing opportunities for learning.
• How to adapt the cooperation and coordination strategies.

Publications

• Hamid R. Berenji, David A. Vengerov. (2000) Learning, Cooperation, and Coordination in Multi-Agent Systems, Technical report, IIS-00-10. Describes three multi-agent domains we have considered so far and presents experimental results on multi-agent cooperation and coordination. Also describes the single agent architecture and learning  algorithms used in multi-agent simulations.
• Hamid R. Berenji, David A. Vengerov. (2000) Advantages of Cooperation Between Reinforcement Learning Agents in Difficult Stochastic Problems. In proceedings of the 9th IEEE International Conference onFuzzy Systems (FUZZ-IEEE '00). Explains the cooperative advantage reported in the previous paper and determines the necessary conditions for its presence in more general problem settings. A revised and updated copy of this paper to be submitted to the Transactions on Fuzzy Systems journal 
• Hamid R. Berenji, David A. Vengerov. (1999) Cooperation and Coordination Between Fuzzy Reinforcement Learning Agents in Continuous-State Partially Observable Markov Decision Processes. In proceedings of the 8th IEEE International Conference on Fuzzy Systems (FUZZ-IEEE'99). Describes a complex simulation domain and demonstrates the phenomenon that K cooperative agents learning for N time steps can learn a better policy than a single agent learning for K*N time steps.

For more information, please contact Dr. Hamid Berenji at IIS Corp. Email: berenji@iiscorp.comPhone: (408)730-8345