Software implementation of the neural network training algorithm

Abstract

Software implementation of the neural network training algorithm

Kabyshev O.A., Maslakov M.P., Kabyshev A.M.

Incoming article date: 10.02.2021

The article deals with the issues related to the design of control devices for technological equipment based on artificial neural networks. The algorithm of the neural network training process based on the Widrow-Hoff method is developed, designed for computer-aided design of artificial neural networks operating in the environment of a microprocessor system or a personal computer. The authors perform a software implementation of the algorithm for training a three-layer neural network using the C# programming language on the platform Microsoft.NET. The program uses a neural network training algorithm with a "teacher". The article shows the interface of interaction of the developed program with the "teacher" (designer of neural networks). An example is given of the development of an artificial neural network that performs the function of a binary number adder. The results of the research can be further used in the development of a computer-aided design system for artificial neural networks, and can also be used in the development of software for microprocessor control systems for technological equipment.

Keywords: artificial neural network, training, algorithm, programming language, neuron bias, adder, input signals, output signals, synaptic connections, neurons