<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Publishing DTD v1.3 20210610//EN" "JATS-journalpublishing1-3.dtd">
<article article-type="research-article" dtd-version="1.3" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xml:lang="ru"><front><journal-meta><journal-id journal-id-type="publisher-id">dt</journal-id><journal-title-group><journal-title xml:lang="ru">Цифровая трансформация</journal-title><trans-title-group xml:lang="en"><trans-title>Digital Transformation</trans-title></trans-title-group></journal-title-group><issn pub-type="ppub">2522-9613</issn><issn pub-type="epub">2524-2822</issn><publisher><publisher-name>Educational Establishment “Belarusian State University of Informatics and Radioelectronics”</publisher-name></publisher></journal-meta><article-meta><article-id custom-type="elpub" pub-id-type="custom">dt-478</article-id><article-categories><subj-group subj-group-type="heading"><subject>Research Article</subject></subj-group><subj-group subj-group-type="section-heading" xml:lang="ru"><subject>ТЕХНИЧЕСКИЕ НАУКИ</subject></subj-group><subj-group subj-group-type="section-heading" xml:lang="en"><subject>TECHNICAL SCIENCES</subject></subj-group></article-categories><title-group><article-title>Влияние гиперпараметров нейронной сети на её численную обусловленность</article-title><trans-title-group xml:lang="en"><trans-title>Influence of the Neural Network Hyperparameters on its Numerical Conditioning</trans-title></trans-title-group></title-group><contrib-group><contrib contrib-type="author" corresp="yes"><contrib-id contrib-id-type="orcid">https://orcid.org/0000-0003-0266-7135</contrib-id><name-alternatives><name name-style="eastern" xml:lang="ru"><surname>Шолтанюк</surname><given-names>С. В.</given-names></name><name name-style="western" xml:lang="en"><surname>Sholtanyuk</surname><given-names>S. V.</given-names></name></name-alternatives><bio xml:lang="ru"><p>Ассистент кафедры компьютерных технологий и систем ФПМИ</p><p>пр. Независимости, д. 4, 220030 , г. Минск</p></bio><bio xml:lang="en"><p>Assistant of the Department of Computer Applications and Systems, FAMCS </p><p>4 Independence Ave., 220030 Minsk</p></bio><email xlink:type="simple">SSholtanyuk@bsu.by</email><xref ref-type="aff" rid="aff-1"/></contrib></contrib-group><aff-alternatives id="aff-1"><aff xml:lang="ru"><institution>Белорусский государственный университет</institution></aff><aff xml:lang="en"><institution>Belarusian State University</institution></aff></aff-alternatives><pub-date pub-type="collection"><year>2020</year></pub-date><pub-date pub-type="epub"><day>16</day><month>04</month><year>2020</year></pub-date><volume>0</volume><issue>1</issue><fpage>43</fpage><lpage>50</lpage><permissions><copyright-statement>Copyright &amp;#x00A9; Шолтанюк С.В., 2020</copyright-statement><copyright-year>2020</copyright-year><copyright-holder xml:lang="ru">Шолтанюк С.В.</copyright-holder><copyright-holder xml:lang="en">Sholtanyuk S.V.</copyright-holder><license xml:lang="ru" license-type="creative-commons-attribution" xlink:href="https://creativecommons.org/licenses/by/4.0/" xlink:type="simple"><license-p>Данная работа распространяется под лицензией Creative Commons Attribution 4.0.</license-p></license><license xml:lang="en" license-type="creative-commons-attribution" xlink:href="https://creativecommons.org/licenses/by/4.0/" xlink:type="simple"><license-p>This work is licensed under a Creative Commons Attribution 4.0 License.</license-p></license></permissions><self-uri xlink:href="https://dt.bsuir.by/jour/article/view/478">https://dt.bsuir.by/jour/article/view/478</self-uri><abstract><p>В данной работе рассмотрена задача оценивания численной обусловленности многослойного персептрона, прогнозирующего временные ряды методом скользящего окна. Рассмотрена работа прогностического персептрона при различных наборах гиперпараметров, в частности, при различном количестве нейронов на разных слоях нейронной сети, а также при использовании тех или иных функций активации. Выявлены основные факторы, влияющие на обусловленность нейронной сети, а также особенности её работы при различных функциях активации. Предложены формулы для оценки чисел обусловленности отдельных компонентов прогностического персептрона и самой нейронной сети в целом. Проведён сравнительный анализ результатов обучения прогностического персептрона при различных гиперпараметрах на примере смоделированных временных рядов. Сформулированы условия, обеспечивающие лучшую устойчивость и обусловленность нейронной сети.</p></abstract><trans-abstract xml:lang="en"><p>In this paper, the task of assessment of numerical conditioning of multilayer perceptron, forecasting time series with sliding window method, has been considered. Performance of the forecasting perceptron with various hyperparameters sets, with different amount of neurons and various activation functions in particular, has been considered. Main factors, influencing on the neural net conditioning, have been revealed, as well as performance features, when using various activation functions. Formulas for assessment of condition numbers of individual components of the forecasting perceptron and of the neural network itself have been proposed. Comparative analysis of results of training the forecasting perceptron with various hyperparameters on modeled time series has been performed. Conditions, providing the best stability and conditioning for the neural network, have been formulated.</p></trans-abstract><kwd-group xml:lang="ru"><kwd>прогнозирование временных рядов</kwd><kwd>нейронные сети</kwd><kwd>персептрон</kwd><kwd>численная обусловленность</kwd><kwd>функция активации</kwd></kwd-group><kwd-group xml:lang="en"><kwd>time series forecasting</kwd><kwd>neural networks</kwd><kwd>perceptron</kwd><kwd>numerical conditioning</kwd><kwd>activation function</kwd></kwd-group></article-meta></front><back><ref-list><title>References</title><ref id="cit1"><label>1</label><citation-alternatives><mixed-citation xml:lang="ru">Sengupta, B. How Robust are Deep Neural Networks [Electronic resource] / B. Sengupta, K.J. Friston // arXiv.org e-Print archive – Mode of access: https://arxiv.org/abs/1804.11313. – Date of access: 02.02.2020. – (Preprint / arXiv:1804.11313).</mixed-citation><mixed-citation xml:lang="en">B. Sengupta, K.J. Friston. How Robust are Deep Neural Networks? arXiv preprint arXiv:1804.11313, 2018.</mixed-citation></citation-alternatives></ref><ref id="cit2"><label>2</label><citation-alternatives><mixed-citation xml:lang="ru">Godfellow, I.J. Explaining and Harnessing Adversarial Examples [Electronic resource] / I.J. Goodfellow, J. Shlens, C. Szegedy // International Conference on Learning Representations: proceedings of 3rd International Conference, San Diego, 7-9 May 2015 // arXiv.org e-Print archive – Mode of access: https://arxiv.org/abs/1412.6572. – Date of access: 02.02.2020. – (Preprint / arXiv:1412.6572v3).</mixed-citation><mixed-citation xml:lang="en">I.J. Goodfellow, J. Shlens, C. Szegedy. Explaining and Harnessing Adversarial Examples. International Conference on Learning Representations, arXiv:1412.6572, 2015.</mixed-citation></citation-alternatives></ref><ref id="cit3"><label>3</label><citation-alternatives><mixed-citation xml:lang="ru">Maas, A.L. Rectifier Nonlinearities Improve Neural Network Acoustic Models // A.L. Maas, A.Y. Hannun, A.Y. Ng // International Conference on Machine Learning: proceedings of 30th International Conference, Atlanta, 16-21 June 2013 // Stanford Artificial Intelligence Laboratory – Mode of access: https://ai.stanford.edu/~amaas/papers/relu_hybrid_icml2013_final.pdf. – Date of access: 02.02.2020.</mixed-citation><mixed-citation xml:lang="en">A.L. Maas, A.Y. Hannun, A.Y. Ng. Rectifer Nonlinearities Improve Neural Network Acoustic Models. International Conference on Machine Learning, 2013.</mixed-citation></citation-alternatives></ref><ref id="cit4"><label>4</label><citation-alternatives><mixed-citation xml:lang="ru">Trefethen, L. N. Numerical Linear Algebra / L.N. Trefethen, D. Bau. – Philadelphia : Society for Industrial and Applied Mathematics, 1997. – 390 p.</mixed-citation><mixed-citation xml:lang="en">L. N. Trefethen, D. Bau. Numerical Linear Algebra, Philadelphia, Society for Industrial and Applied Mathematics, 1997, 390 p.</mixed-citation></citation-alternatives></ref><ref id="cit5"><label>5</label><citation-alternatives><mixed-citation xml:lang="ru">Duchi, J. Adaptive Subgradient Methods for Online Learning and Stochastic Optimization / J. Duchi, E. Hazan, Y. Singer // Journal of Machine Learning Research – 2011. – Vol. 12 – P. 2121–2159.</mixed-citation><mixed-citation xml:lang="en">J. Duchi, E. Hazan, Y. Singer. Adaptive Subgradient Methods for Online Learning and Stochastic Optimization. Journal of Machine Learning Research, 2011, vol. 12, pp. 2121-2159.</mixed-citation></citation-alternatives></ref><ref id="cit6"><label>6</label><citation-alternatives><mixed-citation xml:lang="ru">Шолтанюк, С. В. Сравнительный анализ нейросетевой и регрессионных моделей прогнозирования временных рядов / С. В. Шолтанюк // Цифровая трансформация. – 2019. – № 2 (7). – С. 60–68.</mixed-citation><mixed-citation xml:lang="en">Sholtanyuk S.V. Comparative Analysis of Neural Networking and Regression Models for Time Series Forecasting. Cifrovaja transformacija [Digital transformation], 2019, 2 (7), pp. 60–68 (in Russian).</mixed-citation></citation-alternatives></ref></ref-list><fn-group><fn fn-type="conflict"><p>The authors declare that there are no conflicts of interest present.</p></fn></fn-group></back></article>
