ZCS redux

Bull, Larry; Hurst, Jacob

Abstract

Learning classifier systems traditionally use genetic algorithms to facilitate rule discovery, where rule fitness is payoff based. Current research has shifted to the use of accuracy-based fitness. This paper re-examines the use of a particular payoff-based learning classifier system - ZCS. By using simple difference equation models of ZCS, we show that this system is capable of optimal performance subject to appropriate parameter settings. This is demonstrated for both single- and multistep tasks. Optimal performance of ZCS in well-known, multistep maze tasks is then presented to support the findings from the models.

• Journal Article Type: Article
• Publication Date: Jan 1, 2002
• Deposit Date: Jan 22, 2010
• Publicly Available Date: Jul 29, 2016
• Journal: Evolutionary Computation
• Print ISSN: 1063-6560
• Electronic ISSN: 1530-9304
• Publisher: Massachusetts Institute of Technology Press (MIT Press)
• Peer Reviewed: Peer Reviewed
• Volume: 10
• Issue: 2
• Pages: 185-205
• DOI: https://doi.org/10.1162/106365602320169848
• Keywords: ZCS
• Public URL: https://uwe-repository.worktribe.com/output/1079595
• Publisher URL: http://dx.doi.org/10.1162/106365602320169848
• Additional Information: Additional Information : Learning Classifier Systems (LCS) are an evolutionary computing-based form of reinforcement learner. They were originally presented by John Holland shortly after his genetic algorithm in 1976. In 1995 a new form of LCS was presented and Holland's approach was much criticised. This paper presents formal and experimental results showing how such criticism is incorrect. Specifically, the use of fitness sharing avoids the propogation of inaccurate generalizations in the input space. The result has maintained research with Holland's system which have been showed better able to learn in noisy domains than the newer forms, for example. The work was undertaken as part of a PhD studentship funded by BT Labs.
• Contract Date: Jul 29, 2016