A NOVEL MODEL OF AUTOASSOCIATIVE MEMORY AND ITS SELF-ORGANIZATION

Two types of autoassociative networks are well known, one based on the correlation matrix and the other based on the orthogonal projection matrix, both of which are calculated from the pattern vectors to be stored in the network. Although the latter type of networks have a desirable associative property compared to the former ones, they require, in conventional models, nonlocal calculation of the pattern vectors (i.e., pseudoinverse of a matrix) or some supervised, error-correcting learning procedure. A new model of autoassociative networks is proposed, in which the orthogonal projection operator is implemented in a special relation between the strengths of the connections between neural units is proposed. It is also shown that the connection weights can be achieved by an unsupervised, local learning (Hebbian learning), requiring no pseudoinverse calculation nor the error correction type of learning.