An action-oriented perspective of learning in classifier systems

• Published in: Journal of experimental & theoretical artificial intelligence Vol. 8; no. 1; pp. 43 - 62
• Main Author: WEIss, GERHARD
• Format: Journal Article
• Language: English
• Published: Taylor & Francis Group 01.01.1996
• Subjects: Classifier Systems Bucket Brigade Profit Sharing Plan Action-oriented Learning
• ISSN: 0952-813X; 1362-3079
• DOI: 10.1080/09528139650042538

Classifier systems constitute a general model of low-level rule-based systems that are capable of environmental interaction and learning. A central characteristic and drawback of the traditional approaches to learning in such systems is that they exclusively work on the rule level, without taking into consideration that the individual rules possess a very complex activity behaviour. This article investigates an alternative, action-oriented perspective of learning in classifier systems which does not suffer from this drawback. According to this perspective learning is realized on the finer action level instead of the coarser rule level. Comparative theoretical and experimental results are presented that show the advantages of the action-oriented over the traditional perspective.