Coordinating Learning Agents via Utility Assignment
In this paper, a coordination technique is described for fully cooperative learning based multiagent systems, based on the Collective Intelligence work by Wolpert et al. Our work focuses on a practical implementation of these approaches within a FIPA compliant agent system, using the FIPA-OS agent development toolkit. The functionality of this system is illustrated with a simple buyer/seller agent application, where it is shown that the buyer agents are capable of self-organising behaviour in order to maximise their contribution to the global utility of the system.
KeywordsUtility Function Multiagent System Collective Intelligence Global Utility Leaf System
Unable to display preview. Download preview PDF.
- 1.D. Wolpert & K. Tumer: An Introduction to Collective Intelligence, In Handbook of Agent Technology. AAAI Press/MIT.Google Scholar
- 2.David Wolpert and Kagan Tumer Optimal Payoff Functions for Members of Collectives In Advances in Complex Systems, 2001 (in press).Google Scholar
- 3.R. S. Sutton & A. G. Barto: Reinforcement Learning: An introduction, MIT Press, Cambridge, MA, 1998. Press, 1999.Google Scholar
- 4.S.J. Poslad, P. Buckle & R. Hadingham: The FIPA-OS Agent Platform: Open Source for Open Standards,Proceedings of PAAM 2000, Manchester, UK, (April 2000).Google Scholar
- 5.FIPA (Foundation of Intelligent Physical Agents) homepage: http://www.fipa.org
- 6.S. Sen & G. Weiss: Learning in Multiagent Systems, In Multiagent Systems: A Modern Approach to Distributed Artificial Intelligence The MIT Press (1999).Google Scholar
- 7.G. Zlotkin & J.S. Rosenschein: Negotiation and Task Sharing Among Autonomous Agents in Cooperative Domains, Proceedings of the International Joint Conference on Artificial Intelligence 1989.Google Scholar
- 8.Michael L. Littman. Markov games as a framework for multi-agent reinforcement learning In 11th International Conference on Machine Learning, pages 157–163, 1994.Google Scholar