Nikolay Anatolievich Vershkov

Dr. Nikolay Anatolievich Vershkov

orcid

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 January 2020
Verified
Conference Paper
Closed access
NVNikolay Anatolievich Vershkov
Nikolay Anatolievich Vershkov
VKViktor A. Kuchukov
Viktor A. Kuchukov
NKNataliya N. Kuchukova
Nataliya N. Kuchukova
MBMikhail Babenko
Mikhail Babenko
The article deals with the modeling of artificial neural networks as a system of information transmission. The analysis of the existing theoretical approaches to the optimization of the structure and training of neural networks is carried out. The proposed model is based on the commonality of decoding processes in information transmission systems and clustering processes in neural networks. In the process of building a neuron model, we consider the well-known problem of determining the bandwidth capacity of the communication channel with noise in the geometric terms and its adaptation to the problem of assigning the input signal to a certain cluster. A neuron is considered to be a universal network element capable of performing orthogonal transformations, filtering, and other transformations of the input sequence. The layer of neurons is considered as an information converter with a certain kernel for solving the problems of orthogonal transformation, matched filtering, and nonlinear transformation for combining the spectra of the input influence of the network and its response. Based on the analysis of the proposed model, it is concluded that it is possible to reduce the number of neurons in the hidden layer and reduce the number of features for training the classifier.

28 November 2022 in Programming and Computer Software
Verified
Journal Article
Closed access
PDF file
NVNikolay Anatolievich Vershkov
Nikolay Anatolievich Vershkov
Mikhail BabenkoMikhail Babenko
Mikhail Babenko
Andrei TchernykhAndrei Tchernykh
Andrei Tchernykh
VKViktor Kuchukov
Viktor Kuchukov
NKNikolay Kucherov
Nikolay Kucherov
ADA. Yu. Drozdov
A. Yu. Drozdov
The article presents the artificial neural networks performance optimization using wavelet transform. The existing approaches of wavelet transform implementation in neural networks imply either transformation before neural network or using “wavenet” architecture, which requires new neural network training approaches. The proposed approach is based on the representation of the neuron as a nonrecursive adaptive filter and wavelet filter application to obtain the low-frequency part of the image. It reduces the image size and filtering interference, which is usually high-frequency. Our wavelet transform model is based on the classical representation of a forward propagation neural network or convolutional layers. It allows designing neural networks with the wavelet transform based on existing libraries and does not require changes in the neural network training algorithm. It was tested on three MNIST-like datasets. As a result of testing, it was found that the speed gain is approximately 50 ± 5% with a slight loss of recognition quality of no more than 4%. For practitioner programmers, the proposed algorithm was tested on real images to distinguish animals and showed similar results as the MNIST-like tests.

1 January 2021 in Proceedings of the Institute for System Programming of the RAS
Verified
Journal Article
Open access
PDF file
NVNikolay Anatolievich Vershkov
Nikolay Anatolievich Vershkov
Mikhail BabenkoMikhail Babenko
Mikhail Babenko
VKViktor Kuchukov
Viktor Kuchukov
NKNatalia Nikolaevna Kuchukova
Natalia Nikolaevna Kuchukova
The article deals with the problem of recognition of handwritten digits using feedforward neural networks (perceptrons) using a correlation indicator. The proposed method is based on the mathematical model of the neural network as an oscillatory system similar to the information transmission system. The article uses theoretical developments of the authors to search for the global extremum of the error function in artificial neural networks. The handwritten digit image is considered as a one-dimensional input discrete signal representing a combination of "perfect digit writing" and noise, which describes the deviation of the input implementation from "perfect writing". The ideal observer criterion (Kotelnikov criterion), which is widely used in information transmission systems and describes the probability of correct recognition of the input signal, is used to form the loss function. In the article is carried out a comparative analysis of the convergence of learning and experimentally obtained sequences on the basis of the correlation indicator and widely used in the tasks of classification of the function CrossEntropyLoss with the use of the optimizer and without it. Based on the experiments carried out, it is concluded that the proposed correlation indicator has an advantage of 2-3 times.

28 December 2021 in Programming and Computer Software
Verified
Journal Article
Closed access
PDF file
NVNikolay Anatolievich Vershkov
Nikolay Anatolievich Vershkov
Mikhail BabenkoMikhail Babenko
Mikhail Babenko
Andrei TchernykhAndrei Tchernykh
Andrei Tchernykh
BPB. Pulido-Gaytan
B. Pulido-Gaytan
JCJorge M. Cortés-Mendoza
Jorge M. Cortés-Mendoza
NKN. Kuchukova
N. Kuchukova
The paper discusses optimization issues of training Artificial Neural Networks (ANNs) using a nonlinear trigonometric polynomial function. The proposed method presents the mathematical model of an ANN as an information transmission system where effective techniques to restore signals are widely used. To optimize ANN training, we use energy characteristics assuming ANNs as data transmission systems. We propose a nonlinear layer in the form of a trigonometric polynomial that approximates the “syncular” function based on the generalized approximation theorem and the wave model. To confirm the theoretical results, the efficiency of the proposed approach is compared with standard ANN implementations with sigmoid and Rectified Linear Unit (ReLU) activation functions. The experimental evaluation shows the same accuracy of standard ANNs with a time reduction of the training phase of supervised learning for the proposed model.

22 December 2020 in Programming and Computer Software
Verified
Journal Article
Closed access
PDF file
NVNikolay Anatolievich Vershkov
Nikolay Anatolievich Vershkov
Mikhail BabenkoMikhail Babenko
Mikhail Babenko
VKViktor Kuchukov
Viktor Kuchukov
NKN. Kuchukova
N. Kuchukova
The article discusses the search for a global extremum in the training of artificial neural networks using a correlation indicator. A method based on a mathematical model of an artificial neural network presented as an information transmission system is proposed. Drawing attention to the fact that in information transmission systems widely used methods that allow effective analysis and recovery of useful signal against the background of various interferences: Gaussian, concentrated, pulsed, etc., it is possible to make an assumption about the effectiveness of the mathematical model of artificial neural network, presented as a system of information transmission. The article analyzes the convergence of training and experimentally obtained sequences based on a correlation indicator for fully-connected neural network. The possibility of estimating the convergence of the training and experimentally obtained sequences based on the joint correlation function as a measure of their energy similarity (difference) is confirmed. To evaluate the proposed method, a comparative analysis is made with the currently used indicators. The potential sources of errors in the least-squares method and the possibilities of the proposed indicator to overcome them are investigated. Simulation of the learning process of an artificial neural network has shown that the use of the joint correlation function together with the Adadelta optimizer allows us to get again in learning speed 2-3 times compared to CrossEntropyLoss.