Stability analysis for a class of neural networks

This paper considers the problem of characterizing the stability properties of the equilibria of an important class of recurrent neural networks. Sufficient conditions are given under which the neural network possesses a unique globally asymptotically stable equilibrium point for each external input. These conditions are less restrictive than those previously obtained and are easily checked, so that incorporating them in existing neural network design procedures should increase the flexibility and reduce the complexity of this synthesis process. Results are provided for both continuous-time and discrete-time networks.