Efficient Parallel Learning in Classifier Systems

Abstract

Classifier systems are simple production systems working on binarymessages of fixed length. Genetic algorithms are employed in classifiersystems in order to discover new classifiers. We use methods of thecomputational complexity theory in order to analyse the inherentdifficulty of learning in classifier systems. Hence our results donot depend on special (possibly genetic) learning algorithms. Thepaper formalises this rule discovery or learning problem for classifiersystems which has been proved to be hard in general. It will be provedthat restrictions on two distinct learning problems lead to problemsin NC, i.e. problems which are efficiently solvable in parallel.