All issues

The paper is devoted to convex-concave saddle point problems where the objective is a sum of a large number of functions. Such problems attract considerable attention of the mathematical community due to the variety of applications in machine learning, including adversarial learning, adversarial attacks and robust reinforcement learning, to name a few. The individual functions in the sum usually represent losses related to examples from a data set. Additionally, the formulation admits a possibly nonsmooth composite term. Such terms often reflect regularization in machine learning problems. We assume that the dimension of one of the variable groups is relatively small (about a hundred or less), and the other one is large. This case arises, for example, when one considers the dual formulation for a minimization problem with a moderate number of constraints. The proposed approach is based on using Vaidya’s cutting plane method to minimize with respect to the outer block of variables. This optimization algorithm is especially effective when the dimension of the problem is not very large. An inexact oracle for Vaidya’s method is calculated via an approximate solution of the inner maximization problem, which is solved by the accelerated variance reduced algorithm Katyusha. Thus, we leverage the structure of the problem to achieve fast convergence. Separate complexity bounds for gradients of different components with respect to different variables are obtained in the study. The proposed approach is imposing very mild assumptions about the objective. In particular, neither strong convexity nor smoothness is required with respect to the low-dimensional variable group. The number of steps of the proposed algorithm as well as the arithmetic complexity of each step explicitly depend on the dimensionality of the outer variable, hence the assumption that it is relatively small.

The gradual incorporation of automated vehicles into the global transport networks leads to the need to develop tools to assess the impact of this process on various aspects of traffic. This implies a more organized movement of automated vehicles which can form uniformly moving platoons. The influence of the formation and movement of these platoons on the dynamics of traffic flow is of great interest. The currently most developed traffic flow models are based on the cellular automaton approach. They are mainly developed in the direction of increasing accuracy. This inevitably leads to the complication of models, which in their modern form have significantly moved away from the original philosophy of cellular automata, which implies simplicity and schematicity of models at the level of evolution rules, leading, however, to a complex organized behavior of the system. In the present paper, a simulation model of connected automated vehicles platoon dynamics in a heterogeneous transport system is proposed, consisting of two types of agents (vehicles): human-driven and automated. The description of the temporal evolution of the system is based on modified rules 184 and 240 for elementary cellular automata. Human-driven vehicles move according to rule 184 with the addition of accidental braking, the probability of which depends on the distance to the vehicle in front. For automated vehicles, a combination of rules is used depending on the type of nearest neighbors, regardless of the distance to them, which brings non-local interaction to the model. At the same time, it is considered that a group of sequentially moving connected automated vehicles can form an organized platoon. The influence of the ratio of types of vehicles in the system on the characteristics of the traffic flow during free movement on a circular one-lane and two-lane roads, as well as in the presence of a traffic light, is studied. The simulation results show that the effect of platoon formation is significant for a freeway traffic flow; the presence of a traffic light reduces the positive effect by about half. The movement of platoons of connected automated vehicles on two-lane roads with the possibility of lane changing was also studied. It is shown that considering the types of neighboring vehicles (automated or human-driven) when changing lanes for automated vehicles has a positive effect on the characteristics of the traffic flow.

In this paper, we investigate different alternative ideas for the design of digital predistortion models for radiofrequency power amplifiers. When compared to the greedy search algorithm, these algorithms allow a faster identification of the model parameters combination while still performing reasonably well. For the subsequent implementation, different metrics of model costs and score results in the process of optimization enable us to achieve sparse selections of the model, which balance the model accuracy and model resources (according to the complexity of implementation). The results achieved in the process of simulations show that combinations obtained with explored algorithms show the best performance after a lower number of simulations.

Nowadays, active development of 4th generation nuclear reactors with liquid metal coolants takes place. Therefore, simulation of their elements and units in 3D modelling software are relevant. The thermal-hydraulic analysis of reactor units with liquid metal coolant is recognized as one of the most important directions of the complex of interconnected tasks on reactor unit parameters justification. The complexity of getting necessary information about operating conditions of reactor equipment with liquid-metal coolant on the base of experimental investigations requires the involvement of numerical simulation. The domestic CFD code FlowVision has been used as a research tool. FlowVision software has a certificate of the Scientific and Engineering Centre for Nuclear and Radiation Safety for the nuclear reactor safety simulations. Previously it has been proved that this simulation code had been successfully used for modelling processes in nuclear reactors with sodium coolant. Since at the moment the nuclear industry considers plants with lead-bismuth coolant as promising reactors, it is necessary to justify the FlowVision code suitability also for modeling the flow of such coolant, which is the goal of this work. The paper presents the results of lead-bismuth eutectic flow numerical simulation in the heat exchange tube bundle of NPP steam generator. The convergence studies on a grid and step have been carried out, turbulence model has been selected, hydraulic resistance coefficients of lattices have been determined and simulations with and without $k_\theta^{}$-$e_\theta^{}$ model are compared within the framework of fluid dynamics and heat exchange modeling in the heat-exchange tube bundle. According to the results of the study, it was found that the results of the calculation using the $k_\theta^{}$-$e_\theta^{}$ turbulence model are more precisely consistent with the correlations. A cross-verification with STAR-CCM+ software has been performed as an additional verification on the accuracy of the results, the results obtained are within the error limits of the correlations used for comparison.

A review of the problems of preparing accurate geometric models of steel ropes based on mathematical models without significant simplifications, taking into account the intended purpose of the model, is carried out. Possible approaches to the generation of precise geometric models of steel ropes that have no fundamental limitations on their integration in computational domains and the subsequent construction of finite element models based on them are shown. A generalized parameterized geometric model of single and double twist ropes and its algorithmic implementation using the OpenCASCADE Core Technology geometric modeling kernel in the Gmsh environment (open source software) is considered. The problems of using generic tabular data from steel rope assortment standards as initial data for constructing geometric models are considered. Methods of preliminary verification of collisions of a geometric model based on the initial data of a geometric model are given. Post-verification methods based on Boolean operations over rope wire bodies are given to identify incorrect results of generating models of wire bodies with curvilinear side surfaces based on the algorithm of sequential hierarchical construction of individual wires of single strand and sequential copying of it. Various methods of the process of constructing geometric models of rope wires by extrusion are shown: through a sequence of generatrix with the formation of a body limited by curvilinear surfaces, through a sequence of generatrix with the formation of a body limited by linearly approximated surfaces, and extrusion of one generatrix along a single guideline. The computational complexity of the geometric model generation and the required volume of RAM for the two most universal methods of creating a body of wire are investigated. A method for estimating the value of the step of the arrangement of the generatrix of a single wire is shown, and the influence of its value on the computational complexity of the procedure of wire construction is investigated. Recommendations are given for choosing the value of the radial gap between the layers of wires. An algorithmic implementation of the method for searching for collisions of a geometric model of a steel rope in a non-interactive mode is shown. Approaches to the formation of procedures for processing collisions are proposed. Approaches presented in the article can be implemented in the form of software modules for execution in the Gmsh environment, as well as for another environment using the OpenCascade Core Technology geometric modeling kernel. Such modules allow automation of the construction of accurate geometric models of steel ropes in any configuration without fundamental restrictions on subsequent use, both stand-alone and in the form of objects (primitives) suitable for integration in a third-party model.

The paper investigates a potential solution to the problem of Self-Interference Cancellation (SIC) encountered in the design of In-Band Full-Duplex (IBFD) communication systems. The suppression of selfinterference is implemented in the digital domain using multilayer nonlinear models adapted via the gradient descent method. The presence of local optima and saddle points in the adaptation of multilayer models prevents the use of second-order methods due to the indefinite nature of the Hessian matrix.

This work proposes the use of the Mixed Newton Method (MNM), which incorporates information about the second-order mixed partial derivatives of the loss function, thereby enabling a faster convergence rate compared to traditional first-order methods. By constructing the Hessian matrix solely with mixed second-order partial derivatives, this approach mitigates the issue of “getting stuck” at saddle points when applying the Mixed Newton Method for adapting multilayer nonlinear self-interference compensators in full-duplex system design.

The Hammerstein model with complex parameters has been selected to represent nonlinear selfinterference. This choice is motivated by the model’s ability to accurately describe the underlying physical properties of self-interference formation. Due to the holomorphic property of the model output, the Mixed Newton Method provides a “repulsion” effect from saddle points in the loss landscape.

The paper presents convergence curves for the adaptation of the Hammerstein model using both the Mixed Newton Method and conventional gradient descent-based approaches. Additionally, it provides a derivation of the proposed method along with an assessment of its computational complexity.

While working on activity recognition using wearable sensors for healthcare applications, the main issue arises in the classification of activities. When we attempt to classify activities like walking, sitting, or running from accelerometer and gyroscope data, the signals often overlap and noise complicates the classification process. The existing methods do not have solid mathematical foundations to handle this issue. We started with the standard magnitude approach where one can compute $m =  \sqrt{a^2_1 + a^2_2 + a^2_3}$ from the accelerometer readings, but this approach failed because different activities ended up in overlapping regions. We therefore developed a different approach. Instead of collapsing the 6-dimensional sensor data into simple magnitudes, we keep all six dimensions and treat each activity as a rectangular box in this 6D space. We define these boxes using simple interval constraints. For example, walking occurs when the $x$-axis accelerometer reading is between $2$ and $4$, the $y$-axis reading is between $9$ and $10$, and so on. The key breakthrough is what we call a separability index $s = \frac{d_{\min}^{}}{\sigma}$ that determines how accurately the classification will work. Here dmin represents how far apart the activity boxes are, and $\sigma$ represents the amount of noise present. From this simple idea, we derive a mathematical formula $P(\text{error}) \leqslant (n-1)\exp\left(-\frac{s^2}8\right)$  that predicts the error rate even before initiating the experiment. We tested this on the standard UCI-HAR and WISDM datasets and achieved $86.1 %$ accuracy. The theoretical predictions matched the actual results within $3 %$. This approach outperforms the traditional magnitude methods by $30.6 %$ and explains why certain activities overlap with each other.

This study develops a previously proposed Method of Computer Analogy (MCA) based on formalization of digital computer operations. The paper discusses the position of the proposed approach among other well-known methods. It is emphasized that the primary objective is to derive analytical solutions, although in some cases they have to resort to semianalytical approximations. The paper focuses on constructing solutions for systems which, for certain parameter values, demonstrate the deterministic chaos behavior, namely Lorenz, Marioka – Shimitsu and R¨ossler systems. The paper also considers obtaining solution for Van der Pol equation (reduced to a nonlinear system). The aim of the study is to construct semi-analytical solutions represented as a segment of a power series in a step size of approximating difference scheme. To prevent overflow, authors formalize rank transfer operation. The authors apply a convergent difference scheme, referred to as the “guiding” scheme, to advance to the next step of the independent variable. The resulting approximation by a sum with only a few terms provides an approximation to the solution with any accuracy in accordance with the accuracy of the governing difference scheme. The senior digits in the resulting approximation exhibit probabilistic properties that can be modeled by known distributions, thereby enabling the derivation of analytical and semi-analytical approximations. The paper presents linear approximations that are the base for a complete approximations of solutions and provide important qualitative as well as some quantitative properties of solutions of considered systems. This work describes approximations of various orders, including those that do not guarantee convergence to the exact solution, but simplify the analysis of certain properties of nonlinear equations and systems. In particular, for the Van der Pol equation, authors demonstrate that its corresponding system has a cyclic solution and provide an estimate of its scale. A modification of the MCA that has features of the Monte Carlo method makes it possible to remove recurrent sequences and construct complete solutions in simple situations. The authors mention a promising approach for representing the solution using branched continued fractions.

Monte Carlo (MC) reinforcement learning suffers from high sample complexity, especially in environments with sparse rewards, large state spaces, and strongly correlated trajectories that reduce the statistical efficiency of return estimation. These well-known limitations often lead to slow convergence and unstable learning dynamics, particularly in settings where only a small fraction of collected trajectories is actually informative for policy improvement. A key challenge is therefore to identify a compact yet diverse subset of episodes that contributes most to the accuracy of value estimates while preserving sufficient exploration of the environment. To address this challenge, we reformulate episode selection as a Quadratic Unconstrained Binary Optimization (QUBO) problem and solve it using quantum-inspired sampling techniques. Our method, MC+ QUBO, inserts a combinatorial filtering step into the standard MC policy-evaluation pipeline: given a batch of trajectories, it selects a subset that maximizes cumulative reward and encourages broad state-space coverage. This selection procedure is expressed as a QUBO model, where linear terms favor high-return episodes, quadratic terms penalize redundancy between trajectories, and additional coupling terms can be used to enforce coverage-related constraints or promote structural diversity. Within this framework, we investigate two black-box QUBO solvers: Simulated Quantum Annealing (SQA), which emulates tunneling-based exploration of the search landscape, and Simulated Bifurcation (SB), a dynamical-systems-based iterative optimization method. Both solvers demonstrate the ability to efficiently navigate the combinatorial structure of the trajectory-selection problem and to handle batch sizes that are otherwise computationally expensive for exhaustive or deterministic search. Experiments in a finite-horizon GridWorld environment show that MC+QUBO consistently outperforms vanilla MC in convergence speed, stability of return estimates, and final policy quality. These results highlight the promise of quantum-inspired optimization as a practical decision-making subroutine within reinforcement-learning algorithms, offering a scalable way to improve sample efficiency without modifying the underlying learning paradigm.

The aim of research is to develop software system for solving gas dynamic problem in multiply connected integration domains of regular shape by high-performance computing system. Comparison of the various technologies of parallel computing has been done. The program complex is implemented using multithreaded parallel systems to organize both multi-core and massively parallel calculation. The comparison of numerical results with known model problems solutions has been done. Research of performance of different computing platforms has been done.