CATEGORIZATION IN THE PSEUDO-INVERSE NEURAL-NETWORK

We investigate analytically the emergence of the categorization abilit y in the pseudo-inverse attractor neural network. More pointedly, we c onsider the problem of learning an extensive number of concepts alpha N by storing a finite number of examples s of each concept. We find th at there is a critical value s(c) = 1/alpha beyond which the categoriz ation error, as measured by the average fraction of unstable sites in the concepts, decreases monotonically with s.