A HIERARCHICAL NEURAL-NETWORK MODEL SUITABLE FOR STORING CORRELATION PATTERNS

An improved hierarchical cascade model (PIDD model) has been obtained by using pseudoinverse method without degenerate states in the framewo rk of the hierarchical model improved by Deshpande, the PIDD model not only avoids errors on classification of patterns, but also suits to s tore correlative patterns. The dynamic behavior of the PIDD model has been analyzed, and phase structure and self-consistent equations have also been obtained by using the replica method.