XCS Classifier System Reliably Evolves Accurate, Complete, and Minimal Representations for Boolean Functions

Abstract

Wilson's recent XCS classifier system forms complete mappings of thepayoff environment in the reinforcement learning tradition thanksto its accuracy based fitness. According to Wilson's GeneralizationHypothesis, XCS has a tendency towards generalization. With the XCSOptimality Hypothesis, I suggest that XCS systems can evolve optimalpopulations (representations); populations which accurately map allinput/action pairs to payoff predictions using the smallest possibleset of non-overlapping classifiers. The ability of XCS to evolveoptimal populations for boolean multiplexer problems is demonstratedusing condensation, a technique in which evolutionary search is suspendedby setting the crossover and mutation rates to zero. Condensationis automatically triggered by self-monitoring of performance statistics,and the entire learning process is terminated by autotermination.Combined, these techniques allow a classifier system to evolve optimalrepresentations of boolean functions without any form of supervision.A more complex but more robust and efficient technique for obtainingoptimal populations called subset extraction is also presented andcompared to condensation.