Forming Architecture for Organized Rules in Classifier Systems.
Kenichi Matsuura, Yukinori Kakazu · TRANSACTIONS OF THE JAPAN SOCIETY OF MECHANICAL ENGINEERS Series C · 1996
Recently, there have been many theoretical studies on classifier systems and their engineering applications. However, the learning architecture of classifier systems has brittleness about multi purpose environment. Namely, the learning is not convergent in a multi purpose environment. We consider a cause of those phenomena is that changing purpose in the environment causes to break previous learning results. The objective problem is how to improve the learning architecture of classifier systems in a multi purpose environment. For this purpose, this paper introduces the concept of organized rules which are formed virtually by combination of primitive rules. Using the concept of organized rules, we propose to extend the framework of classifier systems which can form and learn organized rules. Computer experiments show that the proposed method enable effective learning in a multi purpose environment through comparison with conventional classifier systems.