Improved Architecture and the Synthesis Algorithm for Bithreshold Neural Network Classifier

The model of the 3-layer feed-forward neural network is introduced whose first hidden layer consists of bithreshold neurons and the other layers—of single-threshold ones. The proposed network is capable to recognize compact finite set of patterns using a union of hyperrectangular decision regions in n-dimensional space. We design a multiclass classifier on the base of such network, propose the synthesis algorithm for it and estimate the networks size as well as the time of computations. The simulation results demonstrate that the application of the additional hidden layer improves the accuracy of classification.