Flexible pattern matching using a Hopfield-Amari neural network

This paper proposes an approach to flexible recognition of patterns based on the asymptotic stationary property of Hopfield-Amari autoassociative memory networks (a synchronously updating randomly generated recurrent network). In this neural network, the weights of the connective matrix are constructed in terms of the energy function, and the pattern-matching problem is reduced to the procedure of retrieving a target memory pattern that is stored in the neural network. Two experiments, the recognition of handwritten Chinese characters and the recognition of plane figures, are performed to verify the approach. The experimental results show that the method is robust and reliable.