Mechanisms of Fine Tuning of Neuroevolutionary Synthesis of Artificial Neural Networks

Abstract

During performing complex synthesis (structural and parametric) of artificial neural networks (ANN) of complex topologies (for example, recurrent neural networks: RNN and deep neural networks: DNN), neuroevolution is a promising approach. After all, methods of neuroevolution synthesis allow to obtain the appropriate topology and parameters of the neuromodel with less involvement of an external expert, by gradually increasing, increasing the complexity and changing the network parameters. However, neuroevolution synthesis has a significant execution time. There are also sometimes problems with local extremes. That’s why the urgent task is to develop modifications that can be introduced point-by-point during the synthesis process in order to overcome disadvantages that were listed and improve results of using such methods.