Track:
Contents
Downloads:
Abstract:
The Dynamic Classifier System extends the traditional classifier system by replacing its fixed-width ternary representation with Lisp expressions. Genetic programming applied to the classifiers allows the system to discover building blocks in a flexible, fitness directed manner. In this paper, I describe the prior art of problem decomposition using genetic programming and classifier systems. I then show how the proposed system builds oil work in these two areas, extending them in a way that provides for flexible representation and fitness directed discovery of useful building blocks.