Reinforcement Learning in Swarms that Learn

  • Authors:

  • James F. Peters;Christopher Henry;Sheela Ramanna

  • Affiliations:

  • University of Manitoba Department of Electrical and Computer Engineering 15 Gillson St., Winnipeg, Manitoba, Canada R3T 5V6;University of Manitoba Department of Electrical and Computer Engineering 15 Gillson St., Winnipeg, Manitoba, Canada R3T 5V6;Department of Applied Computer Science University of Winnipeg Winnipeg, Manitoba R3B 2E9 Canada

  • Venue:
  • IAT '05 Proceedings of the IEEE/WIC/ACM International Conference on Intelligent Agent Technology
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper introduces an approach to reinforcement learning by cooperating agents using a variation of the actor critic method. This is made possible by considering behavior patterns of swarms in the context of approximation spaces. Rough set theory introduced by Zdzis - law Pawlak in 1982 provides a ground for deriving pattern-based rewards within approximation spaces. The framework provided by an approximation space makes it possible to derive pattern-based reference rewards used to estimate action preferences. Approximation spaces are used to derive action-based reference rewards at the swarm intelligence level. Two different forms of the actor critic reinforcement learning method are considered as a part of a study of learning in realtime by a swarm. The contribution of this article is the presentation of a new actor critic method defined in the context of approximation spaces. An ecosystem designed to facilitate study of reinforcement learning by swarms is briefly described. In addition, the results of ecosystem experiments for two forms of the actor critic method are given.