Do We Really Need to Estimate Rule Utilities in Classifier Systems?

Classiier systems have traditionally used explicit measures of utility (strength, predicted payoo, accuracy, etc.) to quantify the performance and tness of classiier rules. Much of the eeort in designing and implementing these systems has focused on getting these utilities \right". One alternative worth exploring is the idea of using endogenous t-ness; that is, reinforcing successful performance with \resources" that rules need in order to reproduce. Under this regime, the best rules are those that accumulate the most resources over their lifetime and, consequently, have the most oospring. This paper describes a classiier system designed along these lines. Rules have no associated utility measure, just a resource reservoir. When enough resources have been accumulated, the rule reproduces and the reservoir is reduced. Preliminary tests of this system on the multiplexor problem show that it performs as well as utility-based classiier systems such as XCS.