On the multilayered Hopfield neural networks

A multilayered structure of Hopfield neural network is proposed in this paper for the purpose of reducing computational requirement during associative learning. The novel structure which may be viewed as a natural extension of a feedforward multilayered neural network from a static structure to a dynamic system consists of two visible layers and some hidden layers with only interlayer connections between the layers. The mathematical model, state convergence, stability of an equilibrium point, and learning phase for this dynamic neural structure are considered. The advantages of such an architecture are that it lends itself to a simple design procedure and the reductions of the computations.<>