Representational difficulties with classifier systems
Dale Schuurmans, Jonathan Schaeffer · 1989
Classifier systems are currently in vogue as a way of using genetic algorithms to demonstrate machine learning. However, there are a number of difficulties with the formalization that can influence how knowledge is represented and the rate at which the system can learn. Some of the problems are inherent in classifier systems, and one must learn to cope with them, while others are pitfalls waiting to catch the unsuspecting implementor. This paper identifies some of these difficulties, suggesting directions for the further evolution of classifier systems. 1. Introduction In the last five years, genetic algorithms (GA) have become an expanding area of research in computational models of learning. These models are motivated by concepts from evolutionary biology and population genetics. A recent result of this work has been the development of classifier systems (CS), a simple representational and computational paradigm which uses GAs [Gol89, HHN86]. Already a number of CS implementations h...