Maximal Domains of Attraction in a Hopfield Neural Network with Learning

In this paper we describe an approach to maximizing the domains of attraction for equilibria in a Hopfield neural network with learning. The concept of learning in a Hopfield net is introduced and a method is given to construct a hetero-associative memory using a Hopfield net that "learns" the correct weights required to store arbitrarily specified input/output pairs. By proper choice of the feedback gains in the weight update equation it is possible to maximize the domain of attraction for the stored equilibrium points, resulting in a robust associative memory.