An associate memory model based on hopfield neural network with redundant neurons

Abstract In the associate memory model based on the Hopfield neural network, the memorized pattern must satisfy a certain kind of orthogonal condition. However, in general, it is very difficult to satisfy this condition, which has been a great drawback in the application of the associative memory model based on the Hopfield neural network. This paper aims at the remedy of the foregoing point in the associative memory model based on the Hopfield neural networks. Also, a new associative system is proposed based on the new memorized pattern which is generated by redundant neurons to satisfy the orthogonal condition. The proposed method is similar to the traditional method in that the Hopfield neural network is employed. However, it has the following differences: (i) new memorized patterns satisfying the orthogonal condition are generated using redundant neurons; and (ii) the initial state and the timeevolution rule for the redundant neuron are introduced based on the a priori information. Using computer simulation, the case of 3, 4 and 5 memorized patterns are examined and compared, and it is shown that the recall ability is improved greatly in the proposed system, compared to the traditional systems.