Russian Journal of Cybernetics

Commemorating the Teacher: 90th Anniversary of Prof. V. Tupchiev

2025-05-17T10:41:00+03:00

The Choice of Correctness Classes

2025-04-04T11:24:44+03:00

we present a general topological approach for analyzing a broad class of inverse problems that arise in processing information flows. These problems often involve the lack of continuity in the inverse operator due to the a priori assignment of topologies in the space of admissible solutions and given data. To address such ill-posed problems, researchers widely use A.N. Tikhonov’s regularization method. We propose an approach to identifying correctness classes for inverse problems by aligning the topologies of source data and generalizing the concept of the solution space. This ensures the continuous dependence of the solution on the source data. We also examine correctness classes based on the Zermelo selection principle. The proposed approach can support the development of stable artificial neural networks (ANN) for pattern recognition tasks and may extend to artificial intelligence (AI) system technologies.

Effects of Presynaptic Synapses on the Recognition Rate of a Neural Network

2025-04-04T11:22:59+03:00

we studied the properties of a neural network where the rank of the coupling tensor is greater than two. In this case, in addition to the matrix of synaptic connections, there are presynaptic synapses. Such coupling tensors occur in hardware implementations of neural networks based on crossbar arrays. An inherent property of the crossbar scheme is the presence of parasitic currents: a signal from one wire going to a specific neuron flows through memory cells (synapses) to the wires of all other neurons. As a result, noise is superimposed on the input signal of the neuron, weakening signals going to all other neurons. In this scenario, the conductivity of the analog crossbar cell changes proportionally to the noise signal, causing the output signal of the cell to acquire a nonlinear relationship with the input. We showed that with a specific type of coupling tensor, the quality of the neural network algorithm improved significantly. We used a network similar to the Hopfield network as an example.

Tsunami Simulation on a PC

2025-04-04T11:35:02+03:00

we designed hardware code acceleration architectures for solving the nonlinear system of shallow water equations, enabling to run it on a PC or other device. We achieved sufficient computational performance to conduct numerical experiments on transoceanic tsunamis. We proposed a method to adjust source parameters based on wave profile data from deep-sea pressure sensors. By optimizing the computational pipeline and implementing the McCormack finite-difference scheme with second-order approximation, we computed values at seven consecutive time steps within a single clock cycle. Using a Xilinx Virtex-7 VC709 chip as a co-processor on a computational grid of 9,601×6,781 nodes, we performed 36,000 time steps with a 3-second interval, simulating 30 hours of wave propagation in just 1,352 seconds (22.5 minutes). We validated our numerical solution against known exact solutions, analyzed how global bathymetric data affect calculations, and justified the use of the nested grid method. The proposed technology enhances our ability to study tsunamis.

Point Cloud (Irregular Grid) Interpolation with the Curved Point-Normal Triangles

2025-04-04T11:37:54+03:00

we examine interpolation from a point cloud, where data is arranged in an irregular grid, as it plays a crucial role in both scientific research and practical applications. We developed an algorithm for interpolating irregular data using a parameterized curved point-normal triangle. We also address the challenge of selecting vertex normals for the surface.

Possible B-Computer Implementations of Artificial Intelligence

2025-04-04T11:42:32+03:00

we explore the prospects for computing and artificial intelligence (AI) development based on B-technology, which operates on a different principle than the A-technology used in modern computing. B-technology enables us to create computing devices for any number system, ensuring maximum reliability and speed while using a simplified element base. B-computers offer remarkable capabilities, including a high density of function-computing devices and a variable structure device (VSD) that allows us to generate new computing devices within the computer.
Like neural networks, B-computers synthesize computational devices from fragments of mappings. However, unlike neural networks, the synthesized device is fully known, making its associated mapping completely transparent. B-computers self-develop and infinitely expand their computing power not only by improving technical parameters but also by generating new computing devices within themselves.

Implicit Communities in Social Networks and Indoctrination

2025-04-07T22:03:24+03:00

we studied the problem of assessing the quality of implicit community detection in a graph constructed from data imported from social networks and instant messengers. Additionally, we analyzed methods for identifying indoctrination within such networks. We considered two approaches to evaluating the accuracy of community detection. The first approach, based on information theory, involved calculating normalized mutual information (NMI) and asymmetrically normalized mutual information (ANRMI). The second approach examined three methods for assessing the quality of implicit community detection using text analysis. We determined and compared pairwise rank correlation coefficients for dictionaries derived from different text arrays. We also applied correspondence analysis to study corpora of community texts. The third method involved calculating the psycholinguistic characteristics of text arrays associated with implicit communities. Using real data, we demonstrated the applicability of these methods for evaluating the partitioning of a social network graph and analyzing information influence within the network.

Simulation of the Collective Behavior of a Miniature Robot Swarm

2025-04-04T11:55:54+03:00

we developed a two-dimensional model to simulate the collective behavior of a miniature robot swarm. The model incorporates size, mass, and other experimental characteristics. It represents the robots as discrete bodies that interact through an effective potential as the interaction of sensors and emitters with a microcontroller. Each robot generates heat and moves along the gradient of the global temperature field, calculated using the 2D heat conduction equation. We present numerical simulations for interactions ranging from a few robots to swarms of up to 500 robots. The software simulates various self-organization and collective behavior mechanisms, including the self-assembly of robot swarms. To optimize runtime, we implemented the Nvidia CUDA parallel computing architecture. We used gnuplot for visualization of the results.

Computer Simulation of the Total Energy of a Diatomic Silicon Molecule Using the First-Order Perturbation Theory

2025-04-04T12:03:50+03:00

we developed an application to simulate the total energy of a diatomic molecule using wave functions that approximate the solutions of the Hartree-Fock equation for isolated atoms. The paper presents a simulation of the total energy of a diatomic silicon molecule using computer simulation based on the first-order perturbation theory. We found the parameters through a numerical solution of a self-consistent system of equations.

Initial Magnetic Field Amplification in the Geodynamo Model

2025-04-07T23:26:32+03:00

in a hydromagnetic dynamo, the self-excitation mechanism generates the magnetic field by amplifying a weak initial field through specific configurations of conductive fluid flows. These flows sustain the field in a stationary or quasi-stationary state, preventing its attenuation. Demonstrating this amplification effect requires computational magnetohydrodynamics methods. In this paper, we analyze a geodynamo model with vacuum boundary conditions for the magnetic field at the outer boundary of a spherical layer. The dimensionless parameters that determine the problem are: Pr is the Prandtl number, Pm is the magnetic Prandtl number, E is the Ekman number, and Ra* is the modified Rayleigh number. The initial level of the magnetic field depends on the parameter Λ, the Elsasser number. We conducted computational experiments using our MHD code, which we adapted for hybrid computing systems with graphics processors. Our code employs a finite-volume method to solve resistive magnetohydrodynamics problems for a viscous incompressible fluid. A weak (Λ = 10⁻² ) homogeneous magnetic field directed along the axis of rotation of the spherical layer was considered as the initial magnetic field. When conducting computational experiments, for the values of the parameters defining the problem Pr = 1, Pm = 5, E = 5 · 10⁻⁴ , Ra* = 200, we obtained a quasi-stationary solution in which the initial level of the magnetic field energy increases by three orders of magnitude during the generation process. In the obtained solution, the temperature and velocity fields are symmetrical relative to the equatorial plane, and the dipole component predominates in the external magnetic field.

Quantum Singular-Value Decomposition

2025-04-08T11:44:08+03:00

real-valued calculations in artificial neural networks captures the amplitudes of neural signals but neglects their phases, which are critical parameters controlling the composition of cognitive waves in natural brains. To address this limitation, we present a complex-valued modification of singular matrix decomposition, a conceptual precursor to the tensor algebra underlying modern neural networks. In this approach, we generalize the diagonal matrix of singular values to a self-adjoint complex matrix, analogous to the density matrix in quantum theory. Within low-dimensional “semantic” spaces, the additional non-diagonal elements of the complex matrix account for the non-stationary logic of the cognitive systems that generate the data. As in the standard formulation, the density matrix is sandwiched between two orthogonal real-valued matrices, unfolding semantic regularities into the space of observable events. The squared modulus of the resulting set of complex-valued amplitudes then produces observable real-valued data, following the quantum-mechanical Born rule. By introducing a minor increase in the number of parameters, our method significantly enhances the precision of classical singular value decomposition. This improvement highlights the efficiency of wave-like and quantum-inspired principles in natural cognition, as expressed in the proposed algebra. The method provides new opportunities for semantic data analysis and offers pathways to advance modern neural network architectures.

Applicability of the Displacement Discontinuity Method to Curvilinear Crack Growth and Interaction Problems

2025-04-04T12:27:08+03:00

we studied the applicability of numerical methods based on the discontinuous displacement method to the simulation of curvilinear crack growth and interaction under complex loads. We examined how the resolution of the numerical method and the accuracy of stress intensity factor approximation affect the accuracy of a predicted crack growth path. We considered a low computational complexity method under the assumptions of linear elasticity, plane-strain conditions, and quasi-static crack growth. We selected the zero-order accuracy discontinuous displacement method and approximated stress intensity factors using displacement discontinuity values in elements closest to the crack tip. We demonstrated high accuracy in determining stress intensity factors for single-crack cases. For multiple cracks, we obtained criteria of the method applicability: high accuracy in approximating stress intensity factors occurs when the distance between cracks exceeds the size of the boundary element. The numerically calculated growth trajectory coincided with the experimental one for a plexiglass sample under shear load. We observed a weak dependence of the crack trajectory on the boundary element size and the crack length increment at each step under quasi-static growth. We also showed that the crack trajectory remains stable under small deviations in the presence of stress gradients. Based on these results, we concluded the applicability limits of zero-order methods for modeling the growth and interaction of curvilinear cracks.

Domestic 2D and 3D CAD Systems for Students

2025-04-04T12:31:22+03:00

we explored the use of domestic drafting software. This approach simplifies teaching students the fundamental principles of understanding and making part and assembly drawings. It also improves their ability to solve problems creatively.

Modern Problems in Biocybernetics

2025-04-04T12:37:21+03:00

in 1999, V. L. Ginzburg published a seminal article on the challenges in physics and astrophysics, where he uniquely addressed the issue of totalitarianism in science. He specifically discussed Lysenkoism, which not only shaped a dominant scientific direction but also triggered strong opposition to emerging fields such as genetics and cybernetics, both of which were labeled ”pseudosciences” and subjected to political persecution. A similar pattern has emerged in the case of cybernetics and biocybernetics. Today, the contributions of N. A. Bernstein, W. Weaver, and the Eskov-Zinchenko effect are often dismissed or ignored, despite their potential to advance biomedicine, psychology, and ecology beyond what genetics and cybernetics have achieved. What are these overlooked scientific facts? What new science did Weaver envision? And what is the key message of our study? We explore these questions and present our findings.

Efficiency of Conducting Technical Inspections on Corporate Vehicle Fleets within Limited Timeframes

2025-04-04T12:46:41+03:00

we reviewed modern approaches to monitoring the technical condition of motor vehicles (MV), focusing on risk-based methods. We identified the main problems related to time constraints for inspections, lack of data on previous inspections, and coordination issues between employees. We proposed methods for optimizing inspections, including the use of failure prediction algorithms and automated diagnostic systems. We conducted a comparative analysis of time costs and the efficiency of the proposed solutions. Our conclusions allow enterprises to minimize diagnostic costs, increase the safety of MV operations, and improve production performance.

A Model of Interaction and Competition Between Two Types of Autonomous Agents

2025-04-04T12:58:52+03:00

this work constructs and studies a model of interaction between two types of autonomous agents. The model assumes that strong, predatory agents and relatively weak defensive agents coexist in a cellular world. Defensive agents typically do not attack others but can collectively resist a predator. They occupy a specific area of the world known as the ”defensive agent domain”. Each agent has its resources, and defensive agents can replenish their resources by eating portions of food. Predator agents can attack, kill, and eat defensive agents, significantly increasing their resources when they succeed. A large group of defensive agents can pose a threat to a predator that enters their area. To coordinate their actions, defensive agents elect a leader. Upon encountering a predator, the leader commands the defensive agents to gather in a specific location and then collectively threaten the predator, driving it out of their area. This work demonstrates that a large group of relatively weak defensive agents can successfully resist strong predator agents.

Neural Network Applications to the Reconstruction of Coulomb Field in Potential Flows

2025-04-08T12:10:50+03:00

we apply an artificial neural network (ANN) to solve the Cauchy problem by modeling the flow of heavy impurities in potential flows of an incompressible fluid generated by a harmonic potential. We construct an ANN that does not require training to represent the gradient of the Coulomb potential, demonstrating high approximation accuracy. We establish the connection between iterative methods for solving the Cauchy problem and ANNs. Specifically, we design an ANN that implements the Runge-Kutta method, including a fourth-order accurate version, where weight coefficients are predefined at the design stage, eliminating the need for training. We perform a test calculation to compare impurity dynamics simulated using the developed ANN with results from a software package for solving the Cauchy problem. We analyze different initial data configurations and visualize the simulated flow of heavy impurities. Additionally, we demonstrate the practical application of the ANN for locating sources of the Coulomb potential.

Lack of Psychoemotional Content in Training Texts

2025-04-08T12:26:49+03:00

this article addresses a key issue in the development of educational materials: the lack of attention to the psychoemotional component, which affects how students perceive educational texts. This issue becomes particularly significant with the rise of personalized and independent learning approaches that use generative language models to create educational content. We analyze both traditional and innovative methods for developing educational texts and highlight the need for tone analysis techniques to shape their emotional impact. Implementing a system for analyzing the emotional tone of educational texts enables the adaptation of materials to specific conditions, improving their readability and helping students overcome psychoemotional barriers in learning.

The PyPharm Library for Solving Pharmacokinetic Problems

2025-04-04T13:46:06+03:00

recently, simulation has gained significant popularity in the creation of medical products, particularly in the study of pharmacokinetics for developed drugs. Determining the parameters of these models involves a multidimensional optimization problem, which evolutionary algorithms can effectively solve. This study introduces a new open-source Python library designed for pharmacokinetic modeling and optimizing the parameters of pharmacokinetic models using evolutionary algorithms.

System Approach to Decision Support in Medicine

2025-04-08T14:26:21+03:00

we examine the application of a software solution in medical institutions to support medical decision-making and assess disease risks. This solution integrates various diagnostic algorithms with medical information systems.
The proposed software automatically informs doctors about potential disease risks in examined patients based on available medical and social data. It applies diagnostic algorithms even when patients seek medical attention for unrelated conditions.

Modern Image Recognition Technologies : A Review

2025-05-17T10:45:30+03:00

in this paper, we provided an overview of the methods and technologies used in the construction of artificial neural networks (ANN). Biometric technologies based on a person’s unique physical and behavioral characteristics have become a key tool for personal identification. We conducted a comparative analysis of various face recognition models, including Haar cascades, dlib, MTCNN, and FaceNet algorithms. We based the analysis on nine criteria, including accuracy, completeness, processing time, and computing resource consumption. To evaluate the effectiveness of the algorithms, we used a set of 250 images classified according to various conditions. The results demonstrated that each algorithm has its own advantages and disadvantages, depending on the specific tasks and operating conditions.