Work place: Faculty of Applied Mathematics, National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute", Kyiv, Ukraine
E-mail: terejkowski@ukr.net
Website:
Research Interests: Neural Networks, Pattern Recognition, Information Security, Network Security, Speech Recognition, Analysis of Algorithms, Models of Computation
Biography
Igor A. Tereykovskiy graduated from National Aviation University, Kiev, Ukraine. Currently, he is a Doctor of Science, professor at Faculty of Applied Mathematics, National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute", Ukraine. He has currently published more than 100 publications.
His research interests are information security, neural network systems for cyber-attacks recognition, recognition of voice signals.
By Zhengbing Hu Igor A. Tereykovskiy Lyudmila O. Tereykovska Volodymyr V. Pogorelov
DOI: https://doi.org/10.5815/ijisa.2017.10.07, Pub. Date: 8 Oct. 2017
The paper is dedicated to the problem of efficiency increasing in case of applying multilayer perceptron in context of parameters estimation for technical systems. It is shown that the increase of efficiency is possible by adaptation of structure of the multilayer perceptron to the problem specification set. It is revealed that the structure adaptation lies in the determination the following parameters:
1. The number of hidden neuron layers;
2. The number of neurons within each layer.
In terms of the paper, we introduce mathematical apparatus that allows conducting the structure adaptation for minimization of the relative error of the neuro-network model generalization. A numerical experiment to demonstrate efficiency of the mathematical apparatus was developed and described in terms of the article. Further research in this sphere lies in the development of a method for calculation of optimum relationship between the number of the hidden neuron layers and the number of hidden neurons within each layer.
Subscribe to receive issue release notifications and newsletters from MECS Press journals