A corporate classifier system

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.