A synthesis procedure for a generalized neural network

An FTTB (from top to bottom) design strategy for a neural network is introduced. From the scheme the authors present a design method and construct a generalized neural network which has some good qualitative behavior, e.g. any number of isolated prototypes can be set to asymptotically stable fixpoints, and any track converges to one of the equilibria. The time/space complexity of the model is O(mn ), as opposed to O(nn) of Hopfield-like models. A simulation experiment shows its good recognition ability