Cascade associative memory storing hierarchically correlated patterns withvarious correlations

In conventional models for storing hierarchically correlated patterns, corr elations between ancestors (first-level patterns) and their descendants (se cond-level ones) are assumed to be uniform, so that the descendants are dis tributed around their ancestors with equal distances. However, this assumpt ion might be unnatural. We believe that objects are encoded into patterns b y preserving the similarity between them. In this case, descendants are dis tributed around their ancestors with various distances, so that the assumpt ion is invalid and the conventional models become inapplicable. To overcome this, we propose a model CASM3 for storing hierarchically correlated patte rns with various correlations. In CASM3, critical load levels vary with the descendants, and become higher with increasing correlations. Increase in l oad level successively destroys the memories of the descendants in descendi ng order of their correlations. The size of the basins of attraction depend s on the range of the correlations, and becomes larger as the correlation r ange is shifted toward lower levels. (C) 2000 Elsevier Science Ltd. All rig hts reserved.