Rule-Based Evolutionary Online Learning Systems: A Principled Approach to LCS Analysis and Design

Dr. Martin V. Butz (auth.)9783540253792, 3-540-25379-3

This book offers a comprehensive introduction to learning classifier systems (LCS) – or more generally, rule-based evolutionary online learning systems. LCSs learn interactively – much like a neural network – but with an increased adaptivity and flexibility. This book provides the necessary background knowledge on problem types, genetic algorithms, and reinforcement learning as well as a principled, modular analysis approach to understand, analyze, and design LCSs. The analysis is exemplarily carried through on the XCS classifier system – the currently most prominent system in LCS research. Several enhancements are introduced to XCS and evaluated. An application suite is provided including classification, reinforcement learning and data-mining problems. Reconsidering John Holland’s original vision, the book finally discusses the current potentials of LCSs for successful applications in cognitive science and related areas.

Table of contents :
Introduction….Pages 1-7
Prerequisites….Pages 9-30
Simple Learning Classifier Systems….Pages 31-50
The XCS Classifier System….Pages 51-64
How XCS Works: Ensuring Effective Evolutionary Pressures….Pages 65-90
When XCS Works: Towards Computational Complexity….Pages 91-122
Effective XCS Search: Building Block Processing….Pages 123-146
XCS in Binary Classification Problems….Pages 147-156
XCS in Multi-Valued Problems….Pages 157-179
XCS in Reinforcement Learning Problems….Pages 181-195
Facetwise LCS Design….Pages 197-206
Towards Cognitive Learning Classifier Systems….Pages 207-217
Summary and Conclusions….Pages 219-225