A Classifier System for Learning Spatial Representations Based on a Morphological Wave Propagation Algorithm

Contents

A system is proposed which learns spatial representatious of planar feature point sets under supervised learning. A key (Wewtgessing3 aspect of the learning system is the transformarion of the each "static" point set instance into a "dynamic" set of measm'es of sp~tisd relationships spread out over time. A morphologically based wave propagation algorithm [1, 2, 3] performs this transformation of Sl~tla' structm- e into temporal structure. The learning system [4] ls based upon classifiers using bucket brigade and genetic algorithms [5] to respectively modify strengths and create new classifier rules. Such learning systems are designed to exploit temporal reg-larities in learning environments and, thus, fit well with the wave Ixopagntion preprocessing. An initial test environment is proposed that attempts to re-label arbitrary feature labels into structurally memfingful labels.