2013 年 133 巻 9 号 p. 1752-1761
A learning classifier system (LCS) is a model of an intelligent agent interacting with an environment. Many complex yet powerful LCS models exist today. However, they are designed with a single agent approach in mind. LCS applications in multi-agent environment have been problematic. Their complexity limits the agents' cooperation and coordination abilities. This study proposes a simple LCS model for a multi-agent system that allows agents to cooperate and coordinate their actions. New learning methods inspired by organizational learning theories are introduced, giving the agents a capability to recognize useful knowledge. It not only prevents the knowledge from being “forgotten” due to evolutionary process, but also transfers it into less experienced agents. Results show that, with these implementations, the agents manage to coordinate actions better than typical LCS model.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/