A fast supervised learning scheme for recurrent neural networks with application to associative memory design

With the concept of integral manifold that exists in the Hopfield model with supervised learning, a learning scheme which guarantees stable learning with the prescribed learning rate is proposed. With the proposed learning rule and selection of initial conditions, one can achieve a fast and smooth search for the synaptic interconnection that accommodates the desired patterns as stable equilibria of the neural network. Connective stability using the M-matrix condition is used to ensure stable learning. This learning methodology is applied to train a Hopfield model for storing multiple vector patterns.<
>