vestifmИзвестия Национальной академии наук Беларуси. Серия физико-математических наукProceedings of the National Academy of Sciences of Belarus. Physics and Mathematics Series1561-24302524-2415The Republican Unitary Enterprise Publishing House "Belaruskaya Navuka"10.29235/1561-2430-2026-62-2-149-163vestifm-905Research ArticleИНФОРМАТИКАINFORMATICSАлгоритм обучения многослойной нейронной сети прямого распространенияAlgorithm of the multilayer artificial feed-forward neural network learningМухаВ. С.MukhaV. S.

Муха Владимир Степанович – доктор технических наук, профессор, профессор кафедры информационных технологий автоматизированных систем

ул. П. Бровки, 6, 220013, Минск

Vladimir S. Mukha – Dr. Sc. (Engineering), Professor, Professor of the Department of Information Technologies of Automated Systems

6, P. Brovka Str., 220013, Minsk

mukhavladimir@gmail.comБелорусский государственный университет информатики и радиоэлектроникиБеларусьBelarusian State University of Informatics and RadioelectronicsBelarus202610072026622149163Copyright © Муха В.С., 20262026Муха В.С.Mukha V.S.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://vestifm.belnauka.by/jour/article/view/905

Разработан алгоритм обучения многослойной нейронной сети прямого распространения (feedforward neural network), который реализует идею так называемого метода обратного распространения ошибок (error back-propagation). Особенностью алгоритма является его многомерно-матричная форма, обеспечивающая ему теоретическую и алгоритмическую общность. Выполнена программная реализация алгоритма как функции языка программирования Matlab. Несмотря на многомерно-матричную форму алгоритма, эта функция полностью определяется обычными матрицами. Корректность алгоритма подтверждена компьютерным моделированием на примере различных задач аппроксимации, включая задачу распознавания образов.

The algorithm of the multilayer feed-forward neural network learning is developed. The algorithm realizes an error back-propagation idea. The particularity of the algorithm is its multidimensional-matrix form, which provides its theoretical and algorithmic generality. The program realization of the algorithm is performed as the function of Matlab programming language. In spite of the multidimensional-matrix form of the algorithm, this function is defined fully by the usual matrices. The validity of the algorithm is confirmed by the computer simulation of different approximation problems including the problem of pattern recognition.

нейронная сеть прямого распространенияобратное распространение ошибкиглубокое обучениеfeed-forward neural networkerror back-propagationdeep learningReferencesGolovinov A. O., Klimova E. N. Advantages of neural networks over traditional algorithms. Eksperimental’nye i teoreticheskie issledovaniya v sovremennoi nauke: sbornik statei po materialam V Mezhdunarodnoi nauchno-prakticheskoi konferentsii [Experimental and theoretical research in modern science: a collection of articles based on the materials of the V International scientific and practical conference]. Novosibirsk, Association of Researchers “Siberian Academic Book”, 2017, pp. 11–15 (in Russian).Golovinov A. O., Klimova E. N. Advantages of neural networks over traditional algorithms. Eksperimental’nye i teoreticheskie issledovaniya v sovremennoi nauke: sbornik statei po materialam V Mezhdunarodnoi nauchno-prakticheskoi konferentsii [Experimental and theoretical research in modern science: a collection of articles based on the materials of the V International scientific and practical conference]. Novosibirsk, Association of Researchers “Siberian Academic Book”, 2017, pp. 11–15 (in Russian).Mitrea C. A., Lee C. K. M., Wu Z. A Comparison between Neural Networks and Traditional Forecasting Methods: A Case Study. International Journal of Engineering Business Management, 2009, vol. 1, no. 2, pp. 19–24. https://doi.org/10.5772/6777Mitrea C. A., Lee C. K. M., Wu Z. A Comparison between Neural Networks and Traditional Forecasting Methods: A Case Study. International Journal of Engineering Business Management, 2009, vol. 1, no. 2, pp. 19–24. https://doi.org/10.5772/6777Blackard J. A., Dean D. J. Comparative accuracies of artificial neural networks and discriminant analysis in predicting forest cover types from cartographic variables. Computers and Electronics in Agriculture, 1999, vol. 24, no. 3, pp. 131–151. https://doi.org/10.1016/s0168-1699(99)00046-0Blackard J. A., Dean D. J. Comparative accuracies of artificial neural networks and discriminant analysis in predicting forest cover types from cartographic variables. Computers and Electronics in Agriculture, 1999, vol. 24, no. 3, pp. 131–151. https://doi.org/10.1016/s0168-1699(99)00046-0Eze C. M., Ugwuowo I. F., Asogwa O. A comparative analysis of vector autoregressive model and neural networks. EPH – International Journal of Mathematics and Statistics, 2018, vol. 4, no. 2, pp. 1–13. https://doi.org/10.53555/eijms.v4i2.21Eze C. M., Ugwuowo I. F., Asogwa O. A comparative analysis of vector autoregressive model and neural networks. EPH – International Journal of Mathematics and Statistics, 2018, vol. 4, no. 2, pp. 1–13. https://doi.org/10.53555/eijms.v4i2.21Charef F., Ayachi F. A Comparison between Neural Networks and GARCH Models in Exchange Rate Forecasting. International Journal of Academic Research in Accounting, Finance and Management Sciences, 2016, vol. 6, no. 1, pp. 94–99. https://doi.org/10.6007/ijarafms/v6-i1/1996Charef F., Ayachi F. A Comparison between Neural Networks and GARCH Models in Exchange Rate Forecasting. International Journal of Academic Research in Accounting, Finance and Management Sciences, 2016, vol. 6, no. 1, pp. 94–99. https://doi.org/10.6007/ijarafms/v6-i1/1996Rumelhart D. E., Hinton G. E., Williams R. J. Learning internal representations by error propagation. Parallel Distributed Processing: Explorations in the Microstructures of Cognition, vol. 1. Cambridge, MA, MIT Press, 1986, pp. 318– 362. https://doi.org/10.7551/mitpress/4943.003.0128Rumelhart D. E., Hinton G. E., Williams R. J. Learning internal representations by error propagation. Parallel Distributed Processing: Explorations in the Microstructures of Cognition, vol. 1. Cambridge, MA, MIT Press, 1986, pp. 318– 362. https://doi.org/10.7551/mitpress/4943.003.0128Bishop C. M. Pattern Recognition and Machine Learning. Springer Science + Business Media, LLC, 2006. 738 p.Bishop C. M. Pattern Recognition and Machine Learning. Springer Science + Business Media, LLC, 2006. 738 p.Mukha V. S. Analysis of Multidimensional Data. Minsk, Tekhnoprint Publ., 2004. 368 p. (in Russian).Mukha V. S. Analysis of Multidimensional Data. Minsk, Tekhnoprint Publ., 2004. 368 p. (in Russian).Golovko V. A. Neural Network: Learning, Organization and Application. Moscow, IPRJP Publ., 2001. 256 p. (in Russian).Golovko V. A. Neural Network: Learning, Organization and Application. Moscow, IPRJP Publ., 2001. 256 p. (in Russian).Schmidhuber J. Deep learning in neural networks: An overview. Neural Networks, 2015, vol. 61, pp. 85–117. https://doi.org/10.1016/j.neunet.2014.09.003Schmidhuber J. Deep learning in neural networks: An overview. Neural Networks, 2015, vol. 61, pp. 85–117. https://doi.org/10.1016/j.neunet.2014.09.003Nielsen M. Neural Networks and Deep Learning. Vol. 25. San Francisco, CA, USA, Determination Press, 2015. Available at: https://jingyuexing.github.io/Ebook/Machine_Learning/Neural%20Networks%20and%20Deep%20Learning-eng.pdf (accessed 10 January 2025).Nielsen M. Neural Networks and Deep Learning. Vol. 25. San Francisco, CA, USA, Determination Press, 2015. Available at: https://jingyuexing.github.io/Ebook/Machine_Learning/Neural%20Networks%20and%20Deep%20Learning-eng.pdf (accessed 10 January 2025).

The authors declare that there are no conflicts of interest present.