Two Stage Habituation Based Neural

This article describes a novel neural network structure designed for the dynamic classiica-tion of spatio-temporal signals. The network is a two-stage structure consisting of a biologically motivated temporal encoding stage followed by a static neural classiier stage. The temporal encoding stage is based upon a simple biological learning mechanism known as habituation. This habituation based neural structure is capable of approximating arbitrarily well any continuous , causal, time-invariant, mapping from one discrete time sequence to another. Such a structure is applicable to SONAR and speech signal classiication problems, among others. Experiments on classiication of high dimensional feature vectors obtained from Banzhaf sonograms, demonstrate that the proposed network performs better than time delay neural networks while using a substantially simpler structure.