Optimization of Hopfield Neural Network for Improved Pattern Recall and Storage Using Lyapunov Energy Function and Hamming Distance

Abstract

In this paper, we propose a multiconnection-based Hopfield neural network (MC-HNN) based on the hamming distance and Lyapunov energy function to address the limited storage and inadequate recalling capability problems of Hopfield Neural Network (HNN). This study uses the Lyapunov energy function and Hamming Distance to recall correct stored patterns corresponding to noisy test patterns during the convergence phase. The proposed method also extends the traditional HNN storage capacity by storing the individual patterns in the form of etalon arrays through the unique connections among neurons. Hence, the storage capacity now depends on the number of connections and is independent of the total number of neurons in the network. The proposed method achieved the average recall success rate of 100% for bit map images with a noise level of 0, 2, 4, 6 bits, which is a better recall success rate than traditional and genetic algorithm-based HNN methods, respectively. The proposed method also shows quite encouraging results on hand-written images compared with some latest state of art methods.