<?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">guuvest</journal-id><journal-title-group><journal-title xml:lang="ru">Вестник университета</journal-title><trans-title-group xml:lang="en"><trans-title>Vestnik Universiteta</trans-title></trans-title-group></journal-title-group><issn pub-type="ppub">1816-4277</issn><issn pub-type="epub">2686-8415</issn><publisher><publisher-name>State University of Management</publisher-name></publisher></journal-meta><article-meta><article-id pub-id-type="doi">10.26425/1816-4277-2022-11-48-55</article-id><article-id custom-type="elpub" pub-id-type="custom">guuvest-3952</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>СТРАТЕГИИ И ИННОВАЦИИ</subject></subj-group><subj-group subj-group-type="section-heading" xml:lang="en"><subject>STRATEGIES AND INNOVATIONS</subject></subj-group></article-categories><title-group><article-title>Сравнение эффективности методов минимизации нулевого и первого порядка в нейронных сетях</article-title><trans-title-group xml:lang="en"><trans-title>Comparison of the efficiency of zero and first order minimization methods in neural networks</trans-title></trans-title-group></title-group><contrib-group><contrib contrib-type="author" corresp="yes"><contrib-id contrib-id-type="orcid">https://orcid.org/0000-0002-5135-6555</contrib-id><name-alternatives><name name-style="eastern" xml:lang="ru"><surname>Губарева</surname><given-names>Е. А.</given-names></name><name name-style="western" xml:lang="en"><surname>Gubareva</surname><given-names>E. A.</given-names></name></name-alternatives><bio xml:lang="ru"><p>Губарева Елена Алексеевна – кандидат физико-математических наук, доцент кафедры математики и информатики</p><p>Москва</p></bio><bio xml:lang="en"><p>Elena A. Gubareva – Cand. Sci (Phys. and Math.), Assoc. Prof. at the Department of Mathematics and Computer Science</p><p>Moscow</p></bio><email xlink:type="simple">ea_gubareva@guu.ru</email><xref ref-type="aff" rid="aff-1"/></contrib><contrib contrib-type="author" corresp="yes"><contrib-id contrib-id-type="orcid">https://orcid.org/0000-0002-6508-663X</contrib-id><name-alternatives><name name-style="eastern" xml:lang="ru"><surname>Хашин</surname><given-names>С. И.</given-names></name><name name-style="western" xml:lang="en"><surname>Khashin</surname><given-names>S. I.</given-names></name></name-alternatives><bio xml:lang="ru"><p>Хашин Сергей Иванович – кандидат физико-математических наук, доцент кафедры информационных технологий и прикладной математики</p><p>Иваново</p></bio><bio xml:lang="en"><p>Sergey I. Khashin – Cand. Sci (Phys. and Math.), Assoc. Prof. at the Information Technologies and Applied Mathematics Department</p><p>Ivanovo</p></bio><email xlink:type="simple">khash2@mail.ru</email><xref ref-type="aff" rid="aff-2"/></contrib><contrib contrib-type="author" corresp="yes"><contrib-id contrib-id-type="orcid">https://orcid.org/0000-0002-3787-804X</contrib-id><name-alternatives><name name-style="eastern" xml:lang="ru"><surname>Шемякова</surname><given-names>Е. С.</given-names></name><name name-style="western" xml:lang="en"><surname>Shemyakova</surname><given-names>E. S.</given-names></name></name-alternatives><bio xml:lang="ru"><p>Шемякова Екатерина Сергеевна – Ph.D., ассоциированный профессор департамента математики и статистики</p><p>Толидо</p></bio><bio xml:lang="en"><p>Ekaterina S. Shemyakova – Ph.D., Assoc. Prof. at the Department of Mathematics and Statistics</p><p>Toledo</p></bio><email xlink:type="simple">Ekaterina.Shemyakova@UToledo.Edu</email><xref ref-type="aff" rid="aff-3"/></contrib></contrib-group><aff-alternatives id="aff-1"><aff xml:lang="ru"><institution>Государственный университет управления</institution><country>Россия</country></aff><aff xml:lang="en"><institution>State University of Management</institution><country>Russian Federation</country></aff></aff-alternatives><aff-alternatives id="aff-2"><aff xml:lang="ru"><institution>Ивановский государственный университет</institution><country>Россия</country></aff><aff xml:lang="en"><institution>Ivanovo State University</institution><country>Russian Federation</country></aff></aff-alternatives><aff-alternatives id="aff-3"><aff xml:lang="ru"><institution>Университет Толидо</institution><country>Соединённые Штаты Америки</country></aff><aff xml:lang="en"><institution>University of Toledo</institution><country>United States</country></aff></aff-alternatives><pub-date pub-type="collection"><year>2022</year></pub-date><pub-date pub-type="epub"><day>25</day><month>12</month><year>2022</year></pub-date><volume>1</volume><issue>11</issue><fpage>48</fpage><lpage>55</lpage><permissions><copyright-statement>Copyright &amp;#x00A9; Губарева Е.А., Хашин С.И., Шемякова Е.С., 2022</copyright-statement><copyright-year>2022</copyright-year><copyright-holder xml:lang="ru">Губарева Е.А., Хашин С.И., Шемякова Е.С.</copyright-holder><copyright-holder xml:lang="en">Gubareva E.A., Khashin S.I., Shemyakova E.S.</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://vestnik.guu.ru/jour/article/view/3952">https://vestnik.guu.ru/jour/article/view/3952</self-uri><abstract><p>Для минимизации целевой функции в нейронных сетях обычно применяют методы первого порядка, предполагающие неоднократное вычисление градиента. Количество переменных в современных нейронных сетях может составлять многие тысячи и даже миллионы. Многочисленные эксперименты показывают, что время аналитического вычисления градиента функции N переменных примерно в N/5 раз больше времени вычисления самой функции. В статье рассматривается возможность использования для минимизации функции методов нулевого порядка. В частности, предлагается новый метод нулевого порядка для минимизации функции: спуск по двумерным пространствам. Проведено сравнение скоростей сходимости трех различных методов: стандартного градиентного спуска с автоматическим выбором шага, координатного спуска с выбором шага по каждой координате и спуска по двумерным подпространствам. Показано, что эффективность правильно организованных методов нулевого порядка в рассмотренных задачах обучения нейронных сетей не ниже градиентных.</p></abstract><trans-abstract xml:lang="en"><p>To minimize the objective function in neural networks, first-order methods are usually used, which involve the repeated calculation of the gradient. The number of variables in modern neural networks can be many thousands and even millions. Numerous experiments show that the analytical calculation time of an N variable function’s gradient is approximately N/5 times longer than the calculation time of the function itself. The article considers the possibility of using zero-order methods to minimize the function. In particular, a new zero-order method for function minimization, descent over two-dimensional spaces, is proposed. The convergence rates of three different methods are compared: standard gradient descent with automatic step selection, coordinate descent with step selection for each coordinate, and descent over two-dimensional subspaces. It has been shown that the efficiency of properly organized zero-order methods in the considered problems of training neural networks is not lower than the gradient ones.</p></trans-abstract><kwd-group xml:lang="ru"><kwd>Нейронные сети</kwd><kwd>целевая функция</kwd><kwd>минимизация</kwd><kwd>градиент</kwd><kwd>градиентный спуск</kwd><kwd>координатный спуск</kwd><kwd>скорость сходимости</kwd></kwd-group><kwd-group xml:lang="en"><kwd>Neural networks</kwd><kwd>objective function</kwd><kwd>minimization</kwd><kwd>gradient</kwd><kwd>gradient descent</kwd><kwd>coordinate descent</kwd><kwd>convergence rate</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">Гафаров Ф.М., Галимянов А.Ф. Искусственные нейронные сети и приложения: учеб. пособие. Казань: Изд-во Казан. ун-та; 2018. 121 с.</mixed-citation><mixed-citation xml:lang="en">Gafarov F.M., Galimyanov A.F. Artificial neural networks and applications: textbook. Allowance. Kazan: Kazan Publishing House. Un-ta; 2018. (In Russian).</mixed-citation></citation-alternatives></ref><ref id="cit2"><label>2</label><citation-alternatives><mixed-citation xml:lang="ru">Николенко С., Кадурин А., Архангельская Е. Глубокое обучение: серия «Библиотека программиста». СПб.: Питер; 2018. 480 с.</mixed-citation><mixed-citation xml:lang="en">Nikolenko S., Kadurin A., Arkhangelskaya E. Deep Learning: series “Programmer’s Library”. St. Petersburg: Peter; 2018. (In Russian).</mixed-citation></citation-alternatives></ref><ref id="cit3"><label>3</label><citation-alternatives><mixed-citation xml:lang="ru">Chollet F. Deep Learning with Python. 2nd Ed. Shelter Island: Manning Publications Co.; 2020. 450 р.</mixed-citation><mixed-citation xml:lang="en">Chollet F. Deep Learning with Python. 2nd Ed. Shelter Island: Manning Publications Co.; 2020. 450 р.</mixed-citation></citation-alternatives></ref><ref id="cit4"><label>4</label><citation-alternatives><mixed-citation xml:lang="ru">Будума Н. Основы глубокого обучения. М.: Манн, Иванов и Фербер; 2020. 306 с.</mixed-citation><mixed-citation xml:lang="en">Buduma N. Fundamentals of Deep Learning. M.: Mann, Ivanov and Ferber; 2020. 306 p. (In Russian).</mixed-citation></citation-alternatives></ref><ref id="cit5"><label>5</label><citation-alternatives><mixed-citation xml:lang="ru">Schroff F., Kalenichenko D., Philbin J. FaceNet: A unified embedding for face recognition and clustering, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015; P. 815–823. https://doi.org/10.1109/CVPR.2015.7298682</mixed-citation><mixed-citation xml:lang="en">Schroff F., Kalenichenko D., Philbin J. FaceNet: A unified embedding for face recognition and clustering, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015; P. 815–823. https://doi.org/10.1109/CVPR.2015.7298682</mixed-citation></citation-alternatives></ref><ref id="cit6"><label>6</label><citation-alternatives><mixed-citation xml:lang="ru">Yanping Huang et al. GPipe: Efficient training of giant neural networks using pipeline parallelism. arXiv:1811.06965 [cs.CV]. https://doi.org/10.48550/arXiv.1811.06965</mixed-citation><mixed-citation xml:lang="en">Yanping Huang et al. GPipe: Efficient training of giant neural networks using pipeline parallelism. arXiv:1811.06965 [cs.CV]. https://doi.org/10.48550/arXiv.1811.06965</mixed-citation></citation-alternatives></ref><ref id="cit7"><label>7</label><citation-alternatives><mixed-citation xml:lang="ru">Банди Б. Методы Оптимизации. Вводный курс. Пер. с англ. М.: Радио и связь; 1988. 127 с.</mixed-citation><mixed-citation xml:lang="en">Bunday B.D. Basic optimization methods. Trans. from Eng. M.: Radio i svyaz’; 1988. 127 p. (In Russian).</mixed-citation></citation-alternatives></ref><ref id="cit8"><label>8</label><citation-alternatives><mixed-citation xml:lang="ru">Kochenderfer M.J., Wheeler T.A. Algorithms for Optimization. MIT Press; 2019. 520 p.</mixed-citation><mixed-citation xml:lang="en">Kochenderfer M.J., Wheeler T.A. Algorithms for Optimization. MIT Press; 2019. 520 p.</mixed-citation></citation-alternatives></ref><ref id="cit9"><label>9</label><citation-alternatives><mixed-citation xml:lang="ru">Avriel M. Nonlinear Programming: Analysis and Methods. Dover Publishing; 2003. 512 p.</mixed-citation><mixed-citation xml:lang="en">Avriel M. Nonlinear programming: analysis and methods. Dover Publishing; 2003. 512 p.</mixed-citation></citation-alternatives></ref><ref id="cit10"><label>10</label><citation-alternatives><mixed-citation xml:lang="ru">Rosenbrock H.H. An automatic method for finding the greatest or least value of a function. The Computer Journal. 1960;3:175–184.</mixed-citation><mixed-citation xml:lang="en">Rosenbrock H.H. An automatic method for finding the greatest or least value of a function. The Computer Journal. 1960;3:175–184.</mixed-citation></citation-alternatives></ref><ref id="cit11"><label>11</label><citation-alternatives><mixed-citation xml:lang="ru">Городецкий С.Ю., Гришагин В.А. Нелинейное программирование и многоэкстремальная оптимизация. Нижний Новгород: Изд-во Нижегородского гос. ун-та им. Н.И. Лобачевского; 2007. 489 c.</mixed-citation><mixed-citation xml:lang="en">Gorodeckii S.Y., Grishagin V.A. Nonlinear programming and multi-extreme optimization. N. Novgorod: N.I. Lobachevsky State University of Nizhny Novgorod Publ. House; 2007. 489 c. (In Russian).</mixed-citation></citation-alternatives></ref><ref id="cit12"><label>12</label><citation-alternatives><mixed-citation xml:lang="ru">Nocedal J., Wright S.J. Numerical Optimization. 2nd ed. NY: Springer New York; 2006. 651 p. https://doi.org/10.1007/978- 0-387-40065-5</mixed-citation><mixed-citation xml:lang="en">Nocedal J., Wright S.J. Numerical Optimization. 2nd ed. NY: Springer New York; 2006. 651 p. https://doi.org/10.1007/978-0-387-40065-5</mixed-citation></citation-alternatives></ref><ref id="cit13"><label>13</label><citation-alternatives><mixed-citation xml:lang="ru">Акулич И.Л. Математическое программирование в примерах и задачах. М.: Высшая школа; 1986. 352 c.</mixed-citation><mixed-citation xml:lang="en">Aculich I.L. Mathematical programming in examples and tasks. Moscow: Vyshaya shkola; 1986. 352 p. (In Russian).</mixed-citation></citation-alternatives></ref><ref id="cit14"><label>14</label><citation-alternatives><mixed-citation xml:lang="ru">Прокопенко Н.Ю. Методы оптимизации. Н. Новгород: ННГАСУ; 2018. 118 с.</mixed-citation><mixed-citation xml:lang="en">Prokopenko M.Y. Optimization methods. N. Novgorod: NNGASU; 2018. 118 p. (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>
