2
RUDN scientist built a neural network to estimate the optimal threshold policy for heterogeneous devices in queueing theory

RUDN scientist built a neural network to estimate the optimal threshold policy for heterogeneous devices in queueing theory

RUDN University mathematician proposed an artificial neural network for solving the queuing theory problems that arise, for example, in computing systems or business processes. Together with classical methods, it will help calculate the optimal queue length, at which it is necessary to use a slower device to minimize the functions of average losses even with a large system load. The results are published in Mathematics.

Queuing theory mathematically describes systems with the need for optimization of the flow of customers, or requests. For example, it is used to calculate the load on servers in computer systems or optimize customer queues in a supermarket (in this case, the “server” is a cashier). The approach by which the requests are distributed depends on each specific task. One of the popular approaches is based on the threshold values with the maximum allowed number of clients in the queue to the server. To find the optimal threshold value, the policy iteration method is used, but it cannot be applied, for example, if there are too many clients. RUDN University mathematicians have shown that in such cases, the classical method can be improved by artificial neural networks.

“Policy-iteration algorithm is a fairly versatile tool for solving various optimization problems. Unfortunately, as is usually the case in practice, this algorithm is not without some limitations, such as the difficulties associated with convergence when the traffic is close to loaded, limitation on the process dimension, and, consequently, on the number of states. Thus, we would like to compensate for some of the weaknesses of this algorithm with other methods for calculating the optimal control policy”, said Dmitry Efrosinin, Ph.D., associate professor of the Department of Probability Theory and Mathematical Statistics at RUDN University.

Mathematicians studied a model of several servers with different operating speeds. The flow of incoming client requests is a Poisson process—the number of requests in any time interval does not depend on how many requests were in another interval. The time it takes to close on each request is distributed exponentially. First, RUDN University mathematicians performed calculations using a classical algorithm and then used the obtained data to teach artificial neural networks. They were then used to calculate the efficiency of the neural network (or the accuracy of the solution).

As a result, it turned out that the neural network gives a fairly accurate result. The value completely coinciding with the theoretical value was obtained with a probability of 80-97%. An almost correct solution, that is, differing by ±1 compared to the theory, was obtained with a probability of 99%. Hence, the mathematicians concluded that, if necessary, the neural network can complement the classical algorithm.

“We strongly believe that the trained neural network can be successfully used to calculate the optimal thresholds for those system parameters for which alternative numerical methods are difficult or impossible to use, for example, in a heavy traffic case. With this study, we confirm that the analysis of controlled queueing systems and the solution of optimization problems using classical Markov decision theory can be successfully combined with machine learning techniques. These approaches do not contradict each other; on the contrary, combining them provides new results”, said Dmitry Efrosinin, Ph.D., associate professor of the Department of Probability Theory and Mathematical Statistics at RUDN University. 

The results are published in the journal Mathematics.

30 Jan 2018
The conference on international arbitration, where law students from European universities simulate court proceedings and alternately defend the interests of the respondent and the orator.
1312
International scientific cooperation View all
16 Oct
530 applications, 90 young scientists from 30 countries. Darya Nazarova, a postgraduate student of RUDN Faculty of Economics, traveled 11,276 km from Moscow to Sao Paulo for the International Scientific School on Technological and Innovation Strategies and Economic Development Policy at the University of Campinas (UNICAMP). Darya Nazarova, a young RUDN scientist, writes about scientific research, rafting and the country of eternal carnival.
83
Similar newsletter View all
16 Oct
Green Diplomacy Center opened in RUDN

A Center for Green Diplomacy was created based on the RUDN Institute of Environmental Engineering. Among the goals is the integration of the results of scientific and practical activities into the development of international relations in the environmental sphere. The center's specialists will also accompany the corporate sector in solving various environmental problems.

131
19 Apr
A huge pizza and a jug of water, why should 5G networks be sliced? The winners of RUDN science competition explain

RUDN summarized the results of the scientific competition "Project Start: work of the science club ". Students of the Faculty of Physics, Mathematics and Natural Sciences have created a project for a managed queuing system using a neural network to redistribute resources between 5G segments. How to increase flexibility, make the network fast and inexpensive and reach more users — tell Gebrial Ibram Esam Zekri ("Fundamental Computer Science and Information Technology", Master's degree, II course) and Ksenia Leontieva ("Applied Mathematics and Computer Science", Master's degree, I course).

183
19 Apr
Lyricists and physicists are now on equal terms: the first humanitarian laboratory opened in RUDN

What is your first association with the word “laboratory”? Flasks and beakers? Microscopes and centrifuges? Yes, many of us would answer the same way.

231
Similar newsletter View all