CC BY
2
Актуальные проблемы авиации и космонавтики - 2022. Том 2
УДК 004.032.26
ИСПОЛЬЗОВАНИЕ НЕЙРОННЫХ СЕТЕЙ
О. В. Куимова, А. А. Гладков, И. В. Маркевич
Научный руководитель - Е. Л. Вайтекунене
Сибирский государственный университет науки и технологий имени академика М. Ф. Решетнева Российская Федерация, 660037, г. Красноярск, просп. им. газ. «Красноярский рабочий», 31
E-mail: olga2008kuimova@mail.ru
В данной статье рассматривается анализ работы нейронных сетей. Алгоритмы сетей способны упростить обработку больших объемов информации. Область применения нейронных сетей расширяется.
Ключевые слова: нейронные сети, информация, алгоритм, метод, объем данных
THE USE NEURAL NETWORKS
O. V. Kuimova, A. A. Gladkov, I. V. Markevich Scientific supervisor - E. L. Vaitekunene
Reshetnev Siberian State University of Science and Technology 31, Krasnoyarskii rabochii prospekt, Krasnoyarsk, 660037, Russian Federation E-mail: olga2008kuimova@mail.ru
This article discusses the analysis of the operation of neural networks. Network algorithms can simplify the processing of large amounts of information. The scope of neural networks is expanding.
Keywords: neural networks, information, algorithm, method, data volume
Difficulties in formalizing real problems were faced in the last century. Neural networks, which appeared around the same time, helped to advance in solving this problem. "Neural networks" are now called any one or another structure that in one way or another simulates processes that imitate the work of the human brain. Thanks to algorithms built on the principles of the brain, the processing of large amounts of information is simplified. These algorithms have the ability to learn. It is this ability that is the main advantage of neural networks over classical algorithms, which makes them a relevant direction in research and development [1].
Neural networks are used where the types of connection between input and output data are unknown. There are two types of learning algorithm: "learning with a teacher", that is, controlled, and "learning without a teacher", respectively - unmanaged. One of them is selected for further work. Usually this is "supervised learning". For the controlled method, a training data set is prepared, which is a group of input signals and a group of output signals corresponding to them. The neural network is trained by creating connections between these two groups. She searches for matches and remembers them. As a rule, the data for training is taken from some historical information.
After the data is loaded, the neural network begins to learn using the selected algorithm. The most commonly used method for this purpose is backpropagation. This method uses loaded values to measure weights and thresholds in such a way that the probability of an error in the prediction on the training set is reduced. Further, using the example of primary data, the network can find an unknown function if the training was successful. Due to the fact that the neural network itself forms
CeiajHg « H h ( | i o p m i a u h o h h o - 3 k o h o m i h h e c k h e c hgtc m i b»
connections between input and output data, it can be used in tasks where the output values are unknown.
Now neural networks have a considerable number of applications in which a person is inferior in efficiency to a machine, and analytical calculations are time-consuming. Such areas are economics, medicine, scientific research, energy, manufacturing, military industry, information technology, etc.
Currently, there are many programs based on neural network technology. The list of issues that they solve includes tasks from face recognition and speech to prediction. The principle of building a neural network is embedded in every problem in which it is possible to build a clear solution algorithm. For example, weather forecast, text recognition, music composition, etc. [2, 3].
The use of neural network technologies is becoming wider every year. The sphere of economy was no exception.
The economy does not stand still and every year brings more and more information that a person cannot, using his capabilities, competently process. Neural networks are already playing a significant role in economic activity.
Today, neurotechnologies in the economy are used to optimize commodity-money flows, predict the dynamics of political ratings, optimize the production process, predict markets, comprehensive diagnostics of product quality, predict markets, and more. One of the tasks solved by neural networks due to the ability to identify hidden dependencies is forecasting.
Neural network forecasting is used in the following aspects of economics: -determination of the reliability of the company; -forecasting sales volume;
-determination of the risk of bankruptcy of the company; -forecasting demand for goods and services; -risk assessment;
-evaluation of the effectiveness of projects, etc.
Neural networks are used in many financial institutions to predict economic parameters, stock indices, and to manage investments.
Citicorp, which is a prime example of the successful use of neural networks, uses neural networks to analyze and predict currency fluctuations. The brokers of the corporation lost in the accuracy of the predictions made by the neural network.
With the help of LBS Capital Management neural networks, it was possible to improve the accuracy of predictions of the S&P 500 stock indices [4, 5].
It is very difficult to create a mathematical model that meets all the requirements, because economic, social, financial systems are quite complex. They are influenced by many factors, events and actions of people. To develop your own neural network, you need huge amounts of data for its training, experienced programmers, and serious computing power. So far, the implementation of this technology is very expensive. But, despite the cost and complexity, it pays off pretty quickly where work is being done on colossal data flows [6].
Neural networks can be used as a replacement for logistic regression or discriminant analysis in data classification problems. Neural networks are able to classify data that is separated non-linearly.
The most common non-linear method is the Bayesian classifier. It builds a quadratic separable surface, while the neural network is able to build a surface of a higher order. The Bayesian classifier requires a large number of examples to accurately estimate the probability for each combination of intervals of variable values. While the neural network is trained on the entire data sample, not dividing it into fragments, which leads to an increase in the adequacy of its settings [7, 8].
Due to their flexibility, neural networks have a wide range of practical applications. This area needs to be studied and developed. Indeed, now there are already effective methods that are used in various fields of activity. But these are rather narrow areas. Neural networks cannot completely replace a person in solving problems. Therefore, these technologies are yet to be explored.
Актуальные проблемы авиации и космонавтики - 2022. Тома 2
References
1. Gainullin, R.N. Forecasting business processes based on neural networks/R.N. Gainullin, I.S. Rizaev, L.M. Sharnin - Text: electronic//Bulletin of Kazan Technological University. - 2017. - №. 3. - S. 56-80. - URL: https:// cyberleninka.ru/ article/ n/ prognozirovanie-biznes-protsessov-na-osnove-neyronnyh-setey (дата обращения 13.01.2022).
2. Katasev A.S., Kataseva D.V., Kirpichnikov A.P. Neural network biometric system for recognizing images of a human face//Bulletin of Kazan Technological University. - 2016. - T. 19. -№. 18. - S. 135-138.
3. McCulloch W.S. A logical calculus of the ideas immanent in nervous activity / W.S. McCulloch, W. Pitts // The bulletin of mathematical biophysics. — Vol. 5, N 4. — 1943. — P. 115-133.
4. Neural Networks: Global Market and Notable Neural Network Developments: B-MAG: Electronic Journal. - URL: https://b-mag.ru/nejronnye-seti-mirovoj-rynok-i-izvestnye-razrabotki-nejrosetej/ (accessed date 13.01.2022). - Text: electronic.
5. NEI INTUIT: [site]. - URL: http:// www.intuit.ru/ studies/ courses/ 3735/ 977/lecture/14689?page=2 (accessed date 13.01.2022).
6. Shiryaev, V.I. Financial markets: Neural networks, chaos and nonlinear dynamics: Textbook/V.I. Shiryaev. - Moscow: LIBROCOM, 2013. — 232 c. - Text: direct.^^^, В. В., & Sheenok, D A. (2012).
7. Estimate software upgrade costs for reliability critical systems. Siberian Journal of Science and Technology, (5 (45)), 62-65.
8. Tynchenko V. S. et al. Methods of developing a competitive strategy of the agricultural enterprise //IOP Conference Series: Earth and Environmental Science. - IOP Publishing, 2019. - Т. 315. - №. 2. - С. 022105.
© Куимова О. В., Гладков А. А., Маркевич И. В., 2022
