A game-theoretic adaptive categorization mechanism for ART-type networks

Abstract

A game-theoretic formulation of adaptive categorization mechanism for ART-type networks is proposed in this paper. We have derived the game-theoretic model ΓAC for competitive processes of categorization of ART-type networks and an update rule for vigilance parameters using the concept of learning automata. Numbers of clusters generated by ART adaptive categorization are similar regardless of the initial vigilance parameters ρ assigned to the ART networks as demonstrated in the experiments provided. The proposed ART adaptive categorization mechanism can thus avoid the problem of choosing suitable vigilance parameter a priori for pattern categorization.