Abstract: Hopfield neural networks are widely applied in picture identification and engineering field, but there are parameter perturbations and time delay in the neural network when it is implemented with the hardware. T hese factors will badly influence the dynamic performance of the neural network and even cause instability of the network. This paper discusses the robustness of ne ural networks with time delay by building a model. And the fairly general and easily verifiable criterion is presented.

Key words: Hopfield neural networks, robustness, fault model, time delayed networks