Hysteretic Neural Network and Its Applications in Associative Memory

A hysteretic neural network is proposed based on the associative memory principle of Hopfield neural network. The hysteretic character make the neurons in the hysteretic neural network have better holding property to the original states, which decreases the possibility of changing the states mistakenly, and enhances the accuracy and the successful rate of associative memory. Furthermore, a learning algorithm for multi-values patterns associative memory is proposed based Hebb rules. The weight matrix is designed dynamically according to the sample patterns and input pattern. Using the learning algorithm, the hysteretic neural network can realize any multi-values patterns associative memory. The simulation results prove the validity of the algorithm.