A corporate classifier system

Tomlinson, Andy; Bull, Larry

Abstract

Based on the proposals of Wilson and Goldberg we introduce a macro-level evolutionary operator which creates structural links between rules in the ZCS model and thus forms "corporations" of rules within the classifier system population. Rule co-dependencies influence both the behaviour of the discovery components of the system and the production system, where a corporation can take control for a number of timesteps. The system is compared to ZCS and also ZCSM in a number of maze environments which include Woods 1 and Woods 7. The corporate classifier system is shown to be the most suitable design to tackle a range of these types of problems.

• Presentation Conference Type: Conference Paper (Published)
• Conference Name: Parallel Problem Solving from Nature—PPSN V
• Start Date: Sep 27, 1998
• End Date: Sep 30, 1998
• Publication Date: Jan 1, 1998
• Publisher: Springer Verlag
• Peer Reviewed: Not Peer Reviewed
• Pages: 550-559
• Series Title: Lecture Notes in Computer Science
• Series Number: 1498
• Series ISSN: 0302-9743
• Book Title: Parallel Problem Solving from Nature—PPSN V
• ISBN: 9783540650782
• DOI: https://doi.org/10.1007/BFb0056897
• Keywords: corporate classifier system
• Public URL: https://uwe-repository.worktribe.com/output/1101720
• Publisher URL: http://dx.doi.org/10.1007/BFb0056897