DOI: 10.17587/prin.9.185-190

Comparative Analysis of the Evolution Method with use of 'Isolates' and the Annealing Simulation Method in Learning of Artificial Neural Networks

Ya. A. Turovsky, yaroslav_turovsk@mail.ru, Voronezh State University, Voronezh, 394018, Russian Federation, A. A. Adamenko, adamenko.artem@gmail.com, Voronezh State University of Engineering Technologies, Voronezh, 394036, Russian Federation

Corresponding author: Turovsky Yaroslav A., Associate Professor, Voronezh State University, Voronezh, 394018, Russian Federation, E-mail: yaroslav_turovsk@mail.ru

A series of computational experiments on the training of artificial neural networks was carried out using the evolutionary method of teaching ANN with the use of "isolates" and an annealing simulation method. The results of a comparative analysis of these experiments showed the possibility of using the developed software package ANNBuilder to simulate the training of a neurochip (NP) with the aim of restoring damaged neural tissue. The first group of these experiments used a simple evolutionary algorithm for teaching ANN with the use of the "isolation" mechanism, which consists in crossing the parent ANN and the emergence of a daughter ANN inside one "isolate" — the spatially restricted region of weight coefficients (WC) of the ANN. The second group used the simulated annealing method, which is based on the physical process of heating the metal to a certain temperature, which forces the atoms of the crystal lattice to leave their positions. It is concluded that when using the evolutionary method of teaching ANN using the "isolation" mechanism in comparison with the second group of experiments, in most cases the best value of the ANN error function is achieved, which justifies the use of this method in simulating the training of low frequencies.

Keywords: neurochip, evolutionary algorithm, annealing simulation method, isolation, isolate, artificial neural networks

pp. 185–190

For citation:
Turovsky Ya. A., Adamenko A. A. Comparative Analysis of the Evolution Method with use of "Isolates" and the Annealing Simulation Method in Learning of Artificial Neural Networks, Programmnaya Ingeneria, 2018, vol. 9, no. 4, pp. 185—190.