On the optimization of the constructive method of training neural networks

The article suggests a constructive method for training neural networks in which neurons added just before the current epoch of training assume the main training load on the new class to ensure the stability of the network in relation to learning on new data classes. The results of computational experiments are presented.