Exploring XCS in multiagent environments

This paper investigates the adaptability of XCS in four different multiagent environments. The environments are realized in a simplified soccer game, and they include (1) singleagent environment, (2) multiagent environment with an opponent, (3) multiagent environment with a teammate, (4) multiagent environment with both an opponent and a teammate. Although XCS generally seems inferior to strengthbased XCS in such stochastic environments, experimental results in a specific stochastic environment show that XCS is superior to strength-based XCS. Furthermore, XCS with profit sharing is more effective than one using the bucket brigade in multiagent environments. Categories and Subject Descriptors I.2.6 [Learning]: Knowledge acquisition General Terms Algorithms, experimentation Keywords Learning Classifier System, Multiagent, Profit sharing, Bucket brigade

• Type: Conference
• Year: 2005
• Where: GECCO
• Authors: Hiroyasu Inoue, Keiki Takadama, Katsunori Shimohara
• Keywords: Bucket Brigade, GECCO 2005, Multiagent Environments, Stochastic Environment