Efficient Learning of Contextual Mappings by Context-Dependent Neural Nets

The paper addresses the problem of using contextual information by neural nets solving problems of contextual nature. The model of a context-dependent neuron is unique in the fact that allows weights to change according to some contextual variables even after the learning process has been completed. The structures of context-dependent neural nets are outlined, the Vapnik-Chervonenkis dimension of the single context-dependent neuron and multilayer net shortly discussed, decision boundaries are analyzed and compared with the traditional nets. The main goal of the article is to present highly effective contextual training algorithms for the considered models of neural nets.