THE ORGANIZATION OF METASTABLE STATES IN A NEURAL NETWORK WITH HIERARCHICHAL PATTERNS
S. Bacci, Germán Mato, Néstor Parga · International Journal of Neural Systems · 1989
We study the organization of the metastable states in a neural network where the memorized patterns are chosen according to a hierarchy. We find that the number of metastable states correlated with one of the stored patterns is dominated by states which exhibit the same hierarchy. It is argued that this property allows us to distinguish categories even when the stored patterns themselves are not retrieved well.