Hybrid neural network for correction of errors «on fly» in the residue number system

The paper «Hybrid neural network for correction of errors «on fly» in the residue number system» is devoted to detection and correction of mistakes in modular neurocomputer in real time. Problems of correction of the erroneous data demand knowledge of size of all number and consequently considerably reduce performance of modular neurocomputer. Classical algorithms of correction in residual classes are not deprived the given lacks and consequently cannot be used in modern high-precision and high-efficiency application. The offered hybrid neural network has the important advantage before analogues - high information capacity. Besides neural networks of similar purpose, but realized on the basis of Hamming and Hopfield networks cannot be retrained at gradual degradation modular neurocomputer in real time. The unification of final ring neural networks considered in article and counter propagation neural network allows to correct mistakes «on fly».