Improving learning of genetic rule-based classifier systems

Abstract
A genetic classifier system is reviewed and used for learning rules for classification. Two new strategies are described that enable all the letters of the alphabet to be learned. A ''remembering'' strategy locks in good rules to overcome forgetting that otherwise occurs during learning. A ''specializing'' strategy fine tunes the search process for rules. Experiments and an encoding scheme are described. Results show, for the first time, that a genetic classifier-type system can learn to classify all the letters of the alphabet. Further, computer experiments show that the new strategies result in faster and more robust classification involving images of varying position, size, and shape.

This publication has 10 references indexed in Scilit: