Unfully interconnected neural networks as associative memory

Unfully interconnected neural networks (UINNs) are proposed as associative memory. The basic idea is to form compact internal representations of patterns in order to increase the storage efficiency of the interconnections. Several effective methods for designing UINNs as associative memory, including monolayered and multilayered neural networks, are presented. A maximum-interconnection-preserving method which forms a rectangular grid structure of local interconnections is proposed. Dynamical modeling almost doubles the average storage per interconnection weight of the neural network compared with the Hopfield model. Multilayered neural networks are of relatively high storage capacity.<<ETX>>