ROBUSTNESS AND PERTURBATION ANALYSIS OF A CLASS OF ARTIFICIAL NEURAL NETWORKS

We study robustness properties of a large class of artificial feedback neural networks for associative memories by addressing the following question: given a neural network with specified stable memories (speci fied asymptotically stable equilibria), under what conditions will a p erturbed model of the neural network possess stable memories that are close (in distance) to the stable memories of the unperturbed neural n etwork model? In arriving at our results, we establish robustness stab ility results for the perturbed neural network models considered and w e determine conditions that ensure the existence of asymptotically sta ble equilibria of the perturbed neural network model that are near the asymptotically stable equilibria of the original unperturbed neural n etwork. These results involve quantitative estimates of the distance b etween the corresponding equilibrium points of the unperturbed and per turbed neural network models.