All issues

The article presents an approach to configuration of an artificial neural network architecture and a training set size. Configuration is based on parameter minimization with constraints specifying neural network model quality criteria. The algorithm of artificial neural network architecture and training set size configuration is applied to dynamic object artificial neural network approximation.
Series of computational experiments were performed. The method is applicable to construction of dynamic object models based on non-linear autocorrelation neural networks.

An approach to the decrease of norm of the correction in Newton’s methods of optimization, based on the Cholesky’s factorization is presented, which is based on the integration with the technique of the choice of leading element of algorithm of linear programming as a method of solving the system of equations. We investigate the issues of increasing of the numerical stability of the Cholesky’s decomposition and the Gauss’ method of exception.

Efficiency of production directly depends on quality of the management of technology which, in turn, relies on the accuracy and efficiency of the processing of control and measuring information. Development of the mathematical methods of research of the system communications and regularities of functioning and creation of the mathematical models taking into account structural features of object of researches, and also writing of the software products for realization of these methods are an actual task. Practice has shown that the list of parameters that take place in the study of complex object of modern production, ranging from a few dozen to several hundred names, and the degree of influence of each factor in the initial time is not clear. Before working for the direct determination of the model in these circumstances, it is impossible — the amount of the required information may be too great, and most of the work on the collection of this information will be done in vain due to the fact that the degree of influence on the optimization of most factors of the original list would be negligible. Therefore, a necessary step in determining a model of a complex object is to work to reduce the dimension of the factor space. Most industrial plants are hierarchical group processes and mass volume production, characterized by hundreds of factors. (For an example of realization of the mathematical methods and the approbation of the constructed models data of the Moldavian steel works were taken in a basis.) To investigate the systemic linkages and patterns of functioning of such complex objects are usually chosen several informative parameters, and carried out their sampling. In this article the sequence of coercion of the initial indices of the technological process of the smelting of steel to the look suitable for creation of a mathematical model for the purpose of prediction is described. The implementations of new types became also creation of a basis for development of the system of automated management of quality of the production. In the course of weak correlation the following stages are selected: collection and the analysis of the basic data, creation of the table the correlated of the parameters, abbreviation of factor space by means of the correlative pleiads and a method of weight factors. The received results allow to optimize process of creation of the model of multiple-factor process.

Nowadays the random search became a widespread and effective tool for solving different complex optimization and adaptation problems. In this work, the problem of an average duration of a random search for one object by another is regarded, depending on various factors on a square field. The problem solution was carried out by holding total experiment with 4 factors and orthogonal plan with 54 lines. Within each line, the initial conditions and the cellular automaton transition rules were simulated and the duration of the search for one object by another was measured. As a result, the regression model of average duration of a random search for an object depending on the four factors considered, specifying the initial positions of two objects, the conditions of their movement and detection is constructed. The most significant factors among the factors considered in the work that determine the average search time are determined. An interpretation is carried out in the problem of random search for an object from the constructed model. The important result of the work is that the qualitative and quantitative influence of initial positions of objects, the size of the lattice and the transition rules on the average duration of search is revealed by means of model obtained. It is shown that the initial neighborhood of objects on the lattice does not guarantee a quick search, if each of them moves. In addition, it is quantitatively estimated how many times the average time of searching for an object can increase or decrease with increasing the speed of the searching object by 1 unit, and also with increasing the field size by 1 unit, with different initial positions of the two objects. The exponential nature of the growth in the number of steps for searching for an object with an increase in the lattice size for other fixed factors is revealed. The conditions for the greatest increase in the average search duration are found: the maximum distance of objects in combination with the immobility of one of them when the field size is changed by 1 unit. (that is, for example, with $4 \times 4$ at $5 \times 5$) can increase the average search duration in $e^{1.69} \approx 5.42$. The task presented in the work may be relevant from the point of view of application both in the landmark for ensuring the security of the state, and, for example, in the theory of mass service.

Mathematical simulation of unsteady natural convection and thermal surface radiation within a rotating square enclosure was performed. The considered domain of interest had two isothermal opposite walls subjected to constant low and high temperatures, while other walls are adiabatic. The walls were diffuse and gray. The considered cavity rotated with constant angular velocity relative to the axis that was perpendicular to the cavity and crossed the cavity in the center. Mathematical model, formulated in dimensionless transformed variables “stream function – vorticity” using the Boussinesq approximation and diathermic approach for the medium, was performed numerically using the finite difference method. The vorticity dispersion equation and energy equation were solved using locally one-dimensional Samarskii scheme. The diffusive terms were approximated by central differences, while the convective terms were approximated using monotonic Samarskii scheme. The difference equations were solved by the Thomas algorithm. The approximated Poisson equation for the stream function was solved by successive over-relaxation method. Optimal value of the relaxation parameter was found on the basis of computational experiments. Radiative heat transfer was analyzed using the net-radiation method in Poljak approach. The developed computational code was tested using the grid independence analysis and experimental and numerical results for the model problem.

Numerical analysis of unsteady natural convection and thermal surface radiation within the rotating enclosure was performed for the following parameters: Ra = 10³–10⁶, Ta = 0–10⁵, Pr = 0.7, ε = 0–0.9. All distributions were obtained for the twentieth complete revolution when one can find the periodic behavior of flow and heat transfer. As a result we revealed that at low angular velocity the convective flow can intensify but the following growth of angular velocity leads to suppression of the convective flow. The radiative Nusselt number changes weakly with the Taylor number.

In this paper we discuss lower bounds for convergence of convex optimization methods of high order and attainability of this bounds. We formulate a hypothesis that covers all the cases. It is noticeable that we provide this statement without a proof. Newton method is the most famous method that uses gradient and Hessian of optimized function. However, it converges locally even for strongly convex functions. Global convergence can be achieved with cubic regularization of Newton method [Nesterov, Polyak, 2006], whose iteration cost is comparable with iteration cost of Newton method and is equivalent to inversion of Hessian of optimized function. Yu.Nesterov proposed accelerated variant of Newton method with cubic regularization in 2008 [Nesterov, 2008]. R.Monteiro and B. Svaiter managed to improve global convergence of cubic regularized method in 2013 [Monteiro, Svaiter, 2013]. Y.Arjevani, O. Shamir and R. Shiff showed that convergence bound of Monteiro and Svaiter is optimal (cannot be improved by more than logarithmic factor with any second order method) in 2017 [Arjevani et al., 2017]. They also managed to find bounds for convex optimization methods of p-th order for $p ≥ 2$. However, they got bounds only for first and second order methods for strongly convex functions. In 2018 Yu.Nesterov proposed third order convex optimization methods with rate of convergence that is close to this lower bounds and with similar to Newton method cost of iteration [Nesterov, 2018]. Consequently, it was showed that high order methods can be practical. In this paper we formulate lower bounds for p-th order methods for $p ≥ 3$ for strongly convex unconstrained optimization problems. This paper can be viewed as a little survey of state of the art of high order optimization methods.

The paper is devoted to the problem of minimization of the non-smooth functional $f$ with a non-positive non-smooth Lipschitz-continuous functional constraint. We consider the formulation of the problem in the case of quasi-convex functionals. We propose new strategies of step-sizes and adaptive stopping rules in Mirror Descent for the considered class of problems. It is shown that the methods are applicable to the objective functionals of various levels of smoothness. Applying a special restart technique to the considered version of Mirror Descent there was proposed an optimal method for optimization problems with strongly convex objective functionals. Estimates of the rate of convergence for the considered methods are obtained depending on the level of smoothness of the objective functional. These estimates indicate the optimality of the considered methods from the point of view of the theory of lower oracle bounds. In particular, the optimality of our approach for Höldercontinuous quasi-convex (sub)differentiable objective functionals is proved. In addition, the case of a quasiconvex objective functional and functional constraint was considered. In this paper, we consider the problem of minimizing a non-smooth functional $f$ in the presence of a Lipschitz-continuous non-positive non-smooth functional constraint $g$, and the problem statement in the cases of quasi-convex and strongly (quasi-)convex functionals is considered separately. The paper presents numerical experiments demonstrating the advantages of using the considered methods.

In this paper, we consider the problem of restoring the correspondence matrix based on the observations of real correspondences in Moscow. Following the conventional approach [Gasnikov et al., 2013], the transport network is considered as a directed graph whose edges correspond to road sections and the graph vertices correspond to areas that the traffic participants leave or enter. The number of city residents is considered constant. The problem of restoring the correspondence matrix is to calculate all the correspondence from the $i$ area to the $j$ area.

To restore the matrix, we propose to use one of the most popular methods of calculating the correspondence matrix in urban studies — the entropy model. In our work, which is based on the work [Wilson, 1978], we describe the evolutionary justification of the entropy model and the main idea of the transition to solving the problem of entropy-linear programming (ELP) in calculating the correspondence matrix. To solve the ELP problem, it is proposed to pass to the dual problem. In this paper, we describe several numerical optimization methods for solving this problem: the Sinkhorn method and the Accelerated Sinkhorn method. We provide numerical experiments for the following variants of cost functions: a linear cost function and a superposition of the power and logarithmic cost functions. In these functions, the cost is a combination of average time and distance between areas, which depends on the parameters. The correspondence matrix is calculated for multiple sets of parameters and then we calculate the quality of the restored matrix relative to the known correspondence matrix.

We assume that the noise in the restored correspondence matrix is Gaussian, as a result, we use the standard deviation as a quality metric. The article provides an overview of gradient-free optimization methods for solving non-convex problems. Since the number of parameters of the cost function is small, we use the grid search method to find the optimal parameters of the cost function. Thus, the correspondence matrix calculated for each set of parameters and then the quality of the restored matrix is evaluated relative to the known correspondence matrix. Further, according to the minimum residual value for each cost function, we determine for which cost function and at what parameter values the restored matrix best describes real correspondence.

In this paper, a comparative analysis of the exact and heuristic algorithms presented by the method of branches and boundaries, genetic and ant algorithms, respectively, is carried out to find the optimal solution to the traveling salesman problem using the example of a courier robot. The purpose of the work is to determine the running time, the length of the obtained route and the amount of memory required for the program to work, using the method of branches and boundaries and evolutionary heuristic algorithms. Also, the most appropriate of the listed methods for use in the specified conditions is determined. This article uses the materials of the conducted research, implemented in the format of a computer program, the program code for which is implemented in Python. In the course of the study, a number of criteria for the applicability of algorithms were selected (the time of the program, the length of the constructed route and the amount of memory necessary for the program to work), the results of the algorithms were obtained under specified conditions and conclusions were drawn about the degree of expediency of using one or another algorithm in various specified conditions of the courier robot. During the study, it turned out that for a small number of points  $\leqslant10$, the method of branches and boundaries is the most preferable, since it finds the optimal solution faster. However, when calculating the route by this method, provided that the points increase by more than 10, the operating time increases exponentially. In this case, more effective results are obtained by a heuristic approach using a genetic and ant algorithm. At the same time, the ant algorithm is distinguished by solutions that are closest to the reference ones and with an increase of more than 16 points. Its relative disadvantage is the greatest resource intensity among the considered algorithms. The genetic algorithm gives similar results, but after increasing the points more than 16, the length of the found route increases relative to the reference one. The advantage of the genetic algorithm is its lower resource intensity compared to other algorithms.

The practical significance of this article lies in the potential possibility of using the results obtained for the optimal solution of logistics problems by an automated system in various fields: warehouse logistics, transport logistics, «last mile» logistics, etc.

The article considers minimization of the expectation of convex function. Problems of this type often arise in machine learning and a variety of other applications. In practice, stochastic gradient descent (SGD) and similar procedures are usually used to solve such problems. We propose to use the ellipsoid method with mini-batching, which converges linearly and can be more efficient than SGD for a class of problems. This is verified by our experiments, which are publicly available. The algorithm does not require neither smoothness nor strong convexity of the objective to achieve linear convergence. Thus, its complexity does not depend on the conditional number of the problem. We prove that the method arrives at an approximate solution with given probability when using mini-batches of size proportional to the desired accuracy to the power −2. This enables efficient parallel execution of the algorithm, whereas possibilities for batch parallelization of SGD are rather limited. Despite fast convergence, ellipsoid method can result in a greater total number of calls to oracle than SGD, which works decently with small batches. Complexity is quadratic in dimension of the problem, hence the method is suitable for relatively small dimensionalities.