Evolved cooperation and emergent communication structures in learning classifier based organic computing systems
Alexander Scheidler, Martin Middendorf · 2009
In this paper we look at systems consisting of many autonomous components or agents which have only limited amount of resources (e.g. memory) but are able to communicate with each other. The aim of these systems is to solve classification problems (usually to classify binary strings). We incorporate a pittsburgh style learning classifier system into the agents and extend its possible actions by actions for passing the classification requests to other agents. We show that the system is able to overcome the limited resources of its parts by evolving cooperation between them. We take a deeper look at the structure of the generated rule sets and investigate the occurring communication patterns.