A game-theoretic analysis on adaptive categorization in ART networks

Analysis of a game-theoretic formulation of adaptive categorization in ART-type networks is presented. Classical ART-types networks, however, have only fixed single size clusters formation in categorization, which is controlled by the scalar vigilance parameter /spl rho/. This categorization methodology usually cannot give satisfactory results as the data pattern space is not covered thoroughly by fixed boundary clusters. Analysis on the adapted /spl rho/ based on the unique Nash equilibrium of the adaptive categorization game /spl Gamma//sub AC/ is investigated for parameter selection. /spl rho/-adaptation also helps to solve the difficult problem of choosing a suitable vigilance parameter for data categorization. Simulations of the /spl rho/ adaptation rule on patterns from mixture of distributions are presented.