A multiple hyper-ellipsoidal subclass model for an evolutionary classifier

A pattern classification scheme in which the classifier is able to grow and evolve during the operation process is presented. The evolutionary propert y of the classifier is made possible by modeling the pattern vectors in mul tiple hyper-ellipsoidal subclass distributions. Learning of the classifier takes place at the subclass levels only. This property allows the classifie r to retain its previously learned patterns while accepting and learning ne w pattern classes. The classifier is suitable to operate in dynamical envir onments where continuous updating of the pattern class distributions is nee ded. (C) 2001 Pattern Recognition Society. Published by Elsevier Science Lt d. All rights reserved.