Self-Organization of Communication in Distributed Learning Classifier Systems

Abstract

In this paper, an application of learning classifier systems is presented.An artificial multi-agent environment has been designed. Mate findingproblem, a learning task inspired by nature, is considered whichneeds cooperation by two distinct agents to achieve the goal. Themain feature of our system is existence of two parallel learningsubsystems which have to agree on a common communication protocolto succeed in accomplishing the task. Apart from standard learningalgorithms, a unification mechanism has been introduced to encouragecoordinated behavior among the agents belonging to the same class.Experimental results are presented which demonstrate the effectivenessof this mechanism and the learning capabilities of classifier systems.