Associative Memory Based on Hysteretic Neural Network

摘要

A new hysteretic neural network was proposed by using hysteretic activation function to take the place of the traditional activation function. Two hysteretic activation functions were given for unipolarity and bipolarity patterns. The hysteretic property could enhance the memory ability of the neuron and the neural network. The history state could affect the current output response of neuron and neural network. The states of the hysteretic neurons could not be changed by the slight change of the input. Therefore, the risk of the wrong reversion state was reduced, and the associative successful rate could be enhanced obviously. Experimental results show that the method could enhance the associative success rate validly.