STRUCTURED HIERARCHICAL NEURAL NETWORK
Viktor S. Dotsenko, Brunello Tirozzi · International Journal of Modern Physics B · 1989
The spatially structured hierarchical neural network which can store ultrametrically correlated patterns is considered. The patterns belonging to a common ancestor state in the two-level hierarchical tree are supposed to be strongly correlated. The models with both uncorrelated and strongly correlated classes of patterns are considered. In the framework of replica symmetric theory the free energy is calculated and in the low-temperature limit the maximum capacity is found.