No. 1 (2026)

This article examines one of the most important combinatorial optimization problems – three-dimensional packaging. Optimizing three-dimensional packaging reduces costs and improves logistics efficiency, making it relevant for industry. This paper analyzes classical approaches such as greedy algorithms and dynamic programming, as well as widely used methods, including evolutionary algorithms and local search. An analysis of existing methods, including greedy search, dynamic programming, evolutionary algorithms, and local search, revealed their key characteristics and identified suitable areas of application. In the context of this analysis, an overview of the key methods that dominated during certain historical periods is presented. The analysis includes consideration of the application conditions of various methods, their effectiveness for specific types of problems, as well as their advantages and limitations.
A multi-level search algorithm is presented that combines the advantages of traditional and modern optimization methods. This multi-level algorithm improves the accuracy of the packaging problem solution through dynamic parameter adjustment. A software package for solving the three-dimensional packaging optimization problem using bioinspired algorithms has been developed. A computational experiment was conducted on test examples (benchmarks). The packing quality obtained using the developed combined bioinspired algorithm is, on average, 7% higher than the packing results obtained using known algorithms, while the solution time is 7% to 25% shorter, demonstrating the effectiveness of the proposed approach. A series of tests and experiments allowed us to refine theoretical estimates of the time complexity of packing algorithms. In the best case, the time complexity of the algorithms is O(n²), and in the worst case, O(n³).

This paper presents a formalization of the problem of automated monitoring of product availability on retail shelves and compliance with the prescribed planogram, leveraging computer vision and machine learning techniques. The aim of this research is to develop algorithmic solutions for the automatic assessment of product availability in distributed retail environments using computer vision methods, thereby addressing the challenge of maintaining optimal and necessary product assortments through continuous shelf monitoring and supporting data-driven managerial decision-making. A technological pipeline for visual data processing is proposed, comprising the stages of image normalization, segmentation, object localization, and classification, implemented with convolutional neural networks—specifically YOLO and U-Net architectures. An integrated product availability metric is introduced, which jointly accounts for physical, visual, and informational dimensions of availability. An optimization problem aimed at improving overall availability is formulated, and an adaptive neural network fine-tuning mechanism is implemented to enhance the accuracy of image recognition and segmentation, as well as the quality of analytical recommendations. Furthermore, an availability-improvement algorithm is proposed for a decision support system, based on the construction of an optimized merchandiser routing plan that prioritizes products and minimizes time expenditures. This routing problem is reduced to a generalized Traveling Salesman Problem (TSP) with priority-based weights. Methods for evaluating and enhancing product availability are proposed and described in detail. Based on the developed approaches and algorithms, a software system for monitoring and improving product availability has been implemented. Experimental results confirm the effectiveness of the proposed solutions: the average recognition accuracy reached 95.8%, and the integrated availability score achieved A = 0.93. The practical significance of this work lies in establishing an algorithmic foundation for intelligent shelf-monitoring systems that enable more efficient management of retail operations and inventory processes

The article discusses a method for determining the strong connectivity of a periodic fuzzy graph (PFG), which can be used to make decisions in emergency situations such as evacuation. The concept of a fuzzy set of strong connectivity is introduced. The main attention is paid to the development of a method for finding a fuzzy set of strong connectivity, which makes it possible to determine the degree of reachability between the vertices of a graph in a certain number of time cycles. The paper begins with a review of existing approaches to connectivity analysis in fuzzy graphs, emphasizing the need to consider time and fuzzy parameters. The main part of the paper is devoted to the description of key concepts and definitions related to periodic fuzzy graphs. The concepts of fuzzy path, time and degree of reachability, as well as fuzzy set of reachability are introduced. An algorithm for finding a fuzzy set of reachability based on the wave method is proposed, which allows determining the degree and time of reachability between graph vertices. Next, the concept of fuzzy set of strong connectivity PFG is introduced and an algorithm for its determination is proposed. As an example, a specific PFG is considered for which the fuzzy set of strong connectivity is calculated. The proposed method can be useful for choosing a movement strategy during evacuation, especially in conditions where the territory model is represented by a periodic fuzzy graph. In the future, it is planned to explore issues related to finding a discrete-time reachability between vertices with a given degree of reachability

Exploratory Landscape Analysis (ELA) features are numerical descriptors of a problem's fitness landscape, often used to recommend optimal algorithm parameters. This study investigates the critical impact of sampling methods and sample size on the approximation of ELA features and the subsequent performance of machine learning models. The research demonstrates that these feature approximations are not absolute characteristics of the landscape but are significantly influenced by the method used to generate the sample points. While increasing the sample size reduces the variance of feature estimates, the choice of sampling strategy itself introduces substantial bias, leading to statistically different feature values across methods like Mersenne Twister, Latin Hypercube Sampling (LHS), and Faure sequences. The core experiment involved predicting the parameters of the tunable W-model problem using regression models trained on ELA features. The results showed that models trained and tested on data from the same sampling method performed best, highlighting a lack of interoperability between different sampling techniques. Notably, the Faure quasirandom sequences consistently yielded the lowest regression error, outperforming common methods like uniform random sampling and LHS. Furthermore, cross-sampling validation revealed that models, especially those trained on Faure sequences, suffered a significant performance drop when tested on data from any other method, confirming that the sampling strategy imparts a specific "fingerprint" on the feature data. In conclusion, the findings challenge the default use of common sampling methods in ELA. The accuracy of machine learning models for algorithm selection and configuration is highly sensitive to the sampling strategy employed for feature extraction. Therefore, ensuring consistency between the sampling methods used during model training and application is crucial. The superior performance of Faure sequences suggests that low-discrepancy sequences are a promising avenue for future research in making ELA-based models more robust and accurate

Experimental reproducibility constitutes a critical cornerstone of modern machine learning research, yet random initialization seed selection substantially influences final model performance, creating challenges for principled comparison of different architectures and methods. Random seed effects on convolutional time series classifiers were quantified, and a principled comparison criterion was established. Two 1D architectures, FCN and ResNet, were trained on seven public datasets containing different data. 55 independent runs for each combination of model and dataset were performed nder controlled pseudorandomness in Python, NumPy, and PyTorch. Deterministic backends were enabled, and identical hyperparameters were used across runs. Normality of seed-wise accuracy distributions was assessed with the Shapiro–Wilk and Anderson–Darling tests. Accuracy variability attributable to seed choice reached up to 12 percentage points in some settings, with magnitude dependent on dataset and architecture. The distributions were found to be non-normal in most cases, indicating that confidence intervals predicated on normality are unreliable. To enable fair comparison across runs, a reproducibility meta-metric, RM, was introduced that subtracts a dispersion penalty from the mean and depends on the number of runs and a tunable coefficient λ. RM was shown to lie between the empirical minimum and the mean, to approach the lower bound for small sample sizes, and to converge toward the mean as the number of runs increases. Portability of the approach was examined on an additional architecture, DenseNet, confirming expected behavior. Practical value is provided by RM metric rankings reflect both performance and stability. In this way, reproducibility and the credibility of empirical conclusions are strengthened

Coherent optical communication using quadrature phase-shift keying (QPSK) provides high-speed data transmission using in-phase and quadrature (IQ) modulation formats over long distances. A quadrature optical modulator with two Mach-Zehnder interferometers in a two-way configuration with simultaneous phase shift in its arms significantly improves the technical characteristics and ensures the electromagnetic compatibility of digital signal transmission paths. The two-cycle configuration of an optical modulator is the basis for high-speed and noise-resistant data transmission. The research results prove that, after certain corrections to the known coherent optical transmission structures, the system can be used for key distribution. The structure of a coherent two-band optical key sequence formation system is presented. The model for generating optical radiation with modulation by a phase-shifted radio signal is based on a quadrature optical modulator with two parallel Mach-Zehnder interferometers with constant bias voltages on all control electrode arms. Key sequence formation is implemented by modulating the in-phase and quadrature components of radio signals at the subcarrier frequency. A set of two amplitude multiplication coefficients for each bit is specified by the electronic encoding device. The coefficients represent the signals that modulate the amplitude of the in-phase and quadrature components of the subcarrier frequency, respectively. The optical radiation generated at the modulator output provides key distribution (setting the values of zero "0" and one "1" bits in the binary number system) according to a protocol with four phase states in the rectangular and diagonal bases. A module for controlling the encoding of the in-phase and quadrature components of a radio signal at a subcarrier frequency is proposed. The use of quadrature phase shift keying (QPSK) provides noise immunity to external influences at high data transfer rates

The article describes a method for estimating the spatial position of an onboard camera of an aircraft. This method involves comparing aerial photographs and georeferenced remote sensing (RSS) data by using a neural network detector to detect stable spatiotemporal reference points in both datasets. This method then solves the well-known Perspective-n-Point (PnP) problem for estimating rotation and translation matrices that minimize the reprojection error based on the correspondences between 3D world points and 2D points of their projections onto the onboard camera matrix. This approach can be used to solve the pressing problem of aircraft localization in the absence of global navigation satellite system signals. Road intersections are selected as stable spatiotemporal reference points that are clearly visible in RSS data and aerial photographs. Other local semantic image patterns characteristic of a particular area may serve as an alternative. Since direct comparison of remote sensing and airborne images is difficult due to significant differences in shooting conditions, the use of robust landmark detectors based on artificial neural network (ANN) algorithms is proposed. To train the robust detector, a mixed dataset was created using satellite and airborne imagery. The mixed dataset was labeled using a 3D Gaussian function normalized to unity with a apex at the intersection center, the graph of which is projected onto a 2D mask of the training set. The parameters of the Gaussian function are calculated based on the radius of the circle enclosing the intersection. Using a normalized 3D Gaussian function with a apex at the geometric center of the intersection projection allows the network to predict the probability of each image pixel belonging to the intersection, with a maximum at the intersection center, which increases positioning accuracy due to more precise georeferencing of the landmark point in the global 3D dataset. A U-Net-type artificial neural network was trained as an intersection detector. A differentiable analog of the Dice metric was used as a training quality metric. AdamW, coupled with a CosineAnnealingLR cosine learning rate planner, was used as an optimizer. The final section of the paper presents the results of comparing satellite data and airborne imagery using the proposed method.

The propagation and reception of electromagnetic waves in a continuous, heterogeneous medium by mobile sources and receivers is considered. Based on ray theory, the calculation of signal transmission coefficients from the source to the receiver, depending on their positions in space and parameters of medium heterogeneity, is carried out. A set of transmission coefficients for multiple transmitting and receiving points constitute a matrix of MIMO channel coefficients for an information transmission system. Spectral analysis of channel coefficients, depending on motion parameters of transmitting and receiving points, has been performed. The difference in spectrum in homogeneous and heterogeneous media has been determined. The obtained spectra significantly differ from the classical Jakes spectrum, emphasizing the need to account for heterogeneous structure of medium in modeling of modern MIMO systems. Fluctuations in channel coefficients due to spatial heterogeneity of the medium have been investigated, as has the influence of the speed and direction of movement of the transmitter and receiver on the spectral properties of the channel. When moving in a heterogeneous continuous medium, the ergodic capacity of a MIMO information transmission system has been calculated. It has been demonstrated that the complex nature of the medium leads to an unstable distribution of amplitudes and phases for multipath signals, resulting in non-monotonic variations in bandwidth. Maximum capacity has been found to occur when the direction of movement coincides with that of the beam center. It has also been shown that when the velocity vector is orthogonal to the beam direction, the velocity magnitude has a negligible effect on channel properties. The findings of the study allow for a more accurate consideration of the physical characteristics of the environment and the dynamic behavior of sources, which is crucial for developing adaptive signal processing techniques and optimizing MIMO systems of the next generation. The provided dependencies can be utilized to enhance the reliability of data transmission and maximize the throughput in situations where the signal transmitter is mobile within an inhomogeneous and continuous environment

In the context of global industrial digitalization, the development and implementation of corporate information systems (CIS) have become strategically critical for enhancing operational efficiency and competitiveness of enterprises. This study examines the integration case of the ERP system "1C: Enterprise Management" at the AvtoVAZ plant – a key player in the Russian automotive industry. The research aimed to optimize management and production processes through the automation of planning, resource accounting, and coordination of cross-functional interactions. The methodological framework combined analytical, graphical, and comparative approaches, as well as practical testing of solutions in the "1C" software environment. The focus was on designing algorithms for managing production cycles, forming resource specifications, and configuring planning scenarios. The study developed demand forecasting models, analyzed production capacities, and balanced output based on model prioritization (Lada Granta, Vesta, Largus). The system implementation reduced order processing time by 30%, minimized warehouse downtime by 18–22% through synchronized logistics schedules, and improved quality control accuracy via integrated diagnostic tools (CAN-bus, spectrophotometry). Special emphasis was placed on overcoming institutional and technological barriers, including modernizing outdated planning methods, training employees in ERP interfaces, and deploying hybrid cloud solutions to ensure system scalability. The practical significance of the research was confirmed by achieving resource allocation transparency, reducing operational costs, and forming an adaptive production strategy aligned with market dynamics. The results demonstrate that CIS implementation not only optimizes current business processes but also lays the foundation for sustainable enterprise development in the digital transformation era. The acquired experience can be extrapolated to other engineering and industrial enterprises facing challenges in management automation and data integration under competitive pressure. Future research prospects involve analyzing the long-term effects of ERP system adoption, including their impact on innovation potential and supply chain ecosystems.

Ensuring the accuracy and stability of computer simulation of electronic devices is an important problem in their design. The greatest difficulties in simulation of electronic devices arise in the case of the analysis of self-oscillating circuits, since mathematical models of such circuits can be stiff and oscillating at the same time. The aim of this work is to develop an efficient numerical method for solving ordinary differential equations that provides higher accuracy of time domain analysis for various types of autogenerators compared to existing methods. The proposed method is a hybrid method and is based on the well-known Gear and trapezoidal methods used in simulators of electronic circuits. To evaluate the accuracy of the proposed method and known methods a generalized model of a self-oscillating circuit was used for which an analytical solution was determined in the steady-state operating mode. The accuracy of the numerical solution was determined based on the analysis of errors in estimating the main parameters of the oscillatory process – the amplitude and frequency of oscillations. A comparative analysis of errors in estimating the amplitude and frequency of oscillations in autogenerators demonstrates the high efficiency of the proposed hybrid method for analyzing both harmonic oscillators and relaxation oscillators. A further increase in the accuracy of the hybrid method is possible using implicit Runge-Kutta methods (Rado IIA and Lobatto IIIA subclasses), which have L- and P-stability, respectively. It should be noted that with an increase in the order of accuracy of implicit Runge-Kutta methods, the computational complexity of these methods increases, but for the Rado IIA and Lobatto IIIA subclasses the increase in computational complexity will be minimal.

Modern decision support systems (DSS) increasingly rely on heterogeneous data streams from IoT sensors, databases, text messages, and social media, represented in different formats and characterized by diverse semantic models and quality levels. The lack of semantically aligned integration results in inconsistent entity interpretation, duplication, and loss of context, which reduces the quality and timeliness of decisions. The aim of this paper is to systematize methods for semantic analysis and integration of heterogeneous information streams in DSS and to identify their benefits, limitations, and application domains. The study is conducted as an analytical review of publications from 2018–2025 focusing on semantic interoperability, ontologies and knowledge graphs, multi-source data fusion, data federation, and real-time stream processing. The review shows that semantic compatibility is primarily achieved through ontologies and knowledge graphs that define shared entities and identifiers and provide a flexible integration schema. For real-time decision-making, hybrid solutions combining a semantic layer with data fusion algorithms and source trust assessment are the most effective; published case studies report accuracy gains of about 15–20% and response-time reductions of up to 70–80% in multi-source settings. For unstructured streams, NLP and machine learning play a key role by extracting entities and relations and enabling semantic enrichment. The results can be used to design DSS for smart city, industrial, and healthcare domains. Furthermore, the paper highlights the role of standards like SHACL for validation and SPARQL for querying, enhancing the practical applicability of semantic approaches. Future directions include automating ontology alignment to reduce labor costs and integrating with AI for dynamic adaptation to new data sources.

A methodology for analyzing a robust automatic control system for linear dynamic control objects with a delay in the uncertainty of information about the parameters of the mathematical model and interference in the measured signals is described. The system implements a widely used PID controller and a dynamic compensator (DC) for the inertial part of the control object, implemented using estimates of state variables obtained on the basis of a state observer (SO). It is noted that any stability criterion can be used for the asymptotic stability of the controller and observer, but to ensure the maximum degree of stability and the required quality indicators of the transient process, it is convenient to use the maximum degree of stability criterion. In this paper, instead of derivatives obtained by differentiation, it is proposed to use in the dynamic compensator of the inertia of the object estimates of the state variables of the object obtained with the help of a state observer. Another difference from the known ones is the use of sigmoid functions in the corrective feedback of the state observer and the implementation of an additional effect to the main one based on the estimation error, compensating for external disturbance at the input of the object. The parametric synthesis of a state observer was used to study the impact of disturbances and noise on the performance of an automated control system (ACS) for an industrial interval-defined plant with various parameters. A robust typical PID controller with optimal parameters for maximum stability was calculated, taking into account the compensation of the inertial part of the object. For this purpose, a dynamic compensator (DC) was implemented using the parameters of a nominal (calculated) model of the object with the worst combination of parameters obtained on the basis of the interval model of the control object with a delay. The conducted research has established that the structure of the ACS with a typical PID controller with a sequential dynamic compensator for the inertia of the object and a state observer with corrective feedback based on sigma functions ensures the simplicity of the ACS synthesis methodology that is robust to changes in the parameters of the object and the action of unmeasured disturbances and uncontrolled interference.

In modern conditions, ensuring the cyber resilience of enterprise information and telecommunication systems is becoming a priority task that requires accounting for the complex interaction of numerous factors. This paper investigates resilience of information and telecommunication systems through the lens of a process-oriented approach, which focuses on maintaining continuous and secure enterprise operations in the event of cyber incidents. However, one of the current problems lies in the lack of tools for systematic analysis of security factors and their impact on the overall protection of business processes.
The aim of the study is to develop a formalized tool for predicting resilience levels, taking into account functional dependencies between factors. The primary research method is conceptual semantic modeling, which enables the formalization of cause-and-effect relationships between system elements. The scientific novelty consists in the development of a model that captures the mutual influence of threats and countermeasures, allowing for the prediction of information and telecommunication system resilience levels as early as the design stage. Using the method for evaluating the correctness of Industrial Control Systems endpoint parameters as an example, the influence of various factors on the effectiveness of the system’s lower-level protection mechanisms is demonstrated. The results presented in the paper help reduce the gap between the theoretical and practical application of the process-oriented approach. They can also serve as a theoretical foundation for developing software tools to support decision-making in selecting business protection measures, based on a balance between potential damage, expected levels of cyber resilience, risks, and security throughout the entire lifecycle

In the context of the rapid development of national-scale information systems and their evolution into digital ecosystems, new requirements are imposed on the process of ensuring the security of the information processed within them. These requirements include enhancing information availability in user access management while maintaining the required level of confidentiality, and making access decisions to resources based on multiple factors. To meet these requirements, numerous compositional access control models based on roles and attributes have been proposed previously, which have resolved several pressing issues while maintaining administrative convenience and providing flexibility and scalability without role explosion. However, known models still have a significant limitation – the impossibility of their use in information systems where high-sensitivity data is processed. The aim of the study is to develop, within the framework of the subject-object approach methodology in information security theory, a new mandatory role-centric attribute-based access control (MRABAC) model, as well as its formal description using the mathematical apparatus of automata theory. The use of the model will enable dynamic prevention of unauthorized information flows from high-confidentiality objects to low-confidentiality objects during the restriction of the permission set assigned to a role, through the implementation of mandatory access control via a separate attribute-based policy, while preserving the ability to provide users with fine-grained access based on contextual attributes. The application of the model may be particularly useful in large-scale information systems where information of various confidentiality levels is processed simultaneously, and, due to operational characteristics, attribute-based access control is necessary

To use a phased array antenna in the beamforming mode, it is necessary to establish a certain amplitude-phase distribution at the inputs of the emitting elements. Amplitude and phase errors distort the radiation pattern. The paper analyzes the sources of errors in the amplitude-phase distribution of phased antenna arrays, including parasitic phase shifts, nonlinear amplification paths, temperature instability and mutual electromagnetic coupling between the elements. Three methods of calibration and adjustment of phased antenna arrays are described, based on direct measurements of the transmission coefficients in the near zone and the subsequent calculation of the impact vector using inverse and pseudo-inverse matrices of mutual connections, which provides a systematic approach to error elimination. To obtain the initial values, each channel was pre-calibrated along a closed path using a vector network analyzer. Technique 1 implements correction for a set of points in space and one set of states of each channel. To increase the stability of the solution, method 2 uses the regularization of the elements of the matrix of interconnections based on an additional set of measured states of each channel. Method 3 makes it possible to construct a mathematical model of a specific implementation of a phased array antenna based on measurements with a fixed channel state, which ensures the formation of an arbitrary amplitude-phase state without repeated measurements. An experimental setup of an eight-element equidistant linear phased array antenna was carried out. The lattice attenuator/phase shifter modules are based on the PE44820 phase shifter and PE4302 attenuator debugging boards and are controlled by a microcontroller to automatically change phases and amplitudes. The measurements were carried out automatically on a near-field stand in an anechoic shielded chamber using a vector network analyzer. Calibration results are presented, matrices of mutual relationships are constructed and radiation patterns are formed, confirming the operability of the proposed approaches. Since the experimental array is low-element, the results of applying the considered techniques are compared with the results of tuning in the far zone performed using an evolutionary algorithm.

The article is devoted to the actual problem to study of the possibilities of phase cancellation of scattered electromagnetic waves using thin non-absorbing Pancharatnam-Berry (PB) metasurfaces with the generation of vortex waves with orbital angular momentum (OAM) and a spiral phase front. The aim of the work is to design such metasurfaces (MS) based on unit cells of hexagonal shape and traditional square shape and to compare their scattering characteristics and the possibilities of broadband phase cancellation of scattering. Metasurfaces with a dispersionless PB-phase for circular polarized waves
(CP-waves) consist of cells in which the rotation angles of meta-particles change according to a given law. A hexagonal unit cell (like a honeycomb) has six axes of symmetry (instead of four in a square unit cell), which should provide a smaller influence of different rotation angles of adjacent meta-particles of an MS on the properties of the reflection coefficients of the cells. In this paper, a meta-particle in the form of a perforated patch in a hexagonal cell is proposed, which effectively reflects co-polarized CP-waves in the range from 8.5 to 19.7 GHz regardless of the rotation angle of the meta-particle. Four models of metasurfaces from such cells with generation of OAM of different orders were designed. Simulation of scattering of CP-waves by the finite element method confirmed that models with hexagonal cells reduce the backscattering field more effectively (over 10 dB) compared to square cells (only by 8 dB) in an ultra-wide frequency range from 8.5 to 20.8 GHz. Backscattering is reduced due to the generation of funnel-shaped vortex waves with OAM modes of minus the first or plus the third order and a phase singularity of the field on the vortex axis. The obtained results can be useful in choosing the shape of unit cells of metasurfaces intended for broadband scattering reduction

The paper presents the results of a comprehensive study of the characteristics of a microstrip MIMO antenna system intended for application in fifth-generation communication equipment operating within the n79 frequency band (4.4–5.0 GHz). The purpose of the research was to develop a compact 1×2 antenna array on a single dielectric substrate and to evaluate its electrodynamic properties, taking into account factors capable of influencing matching and radiation parameters under real operating conditions.
The proposed design incorporates two radiating elements, four excitation ports, and a modified ground plane, which provides enhanced functional stability and improved operational characteristics. Numerical modelling demonstrated that the developed antenna system ensures stable standing-wave ratio values, high antenna gain, and consistent radiation patterns across the entire operating frequency band. Particular attention was devoted to assessing the influence of external environmental effects, especially the formation of a thin water film on the radiating surface. The study revealed that the presence of moisture results in an SWR increase of more than 0.5, a decrease in antenna efficiency, and degradation of radiation characteristics at higher frequencies, highlighting the necessity of employing protective structural solutions when operating under humidity-prone conditions. Additionally, the feasibility of replacing copper with an ideal conductor in electromagnetic simulations was examined. The difference in SWR between the realistic and idealized models did not exceed 0.006, confirming that such a simplification may be applied during early design stages without significant loss of accuracy. An experimental prototype of the antenna was fabricated and tested under laboratory conditions. The obtained measurements demonstrated a high degree of agreement with the modelling results, validating the proposed mathematical model and confirming the practical applicability of the developed MIMO antenna system in next-generation communication infrastructures

The development of neuromorphic machine vision systems for robotic systems requires the creation of transparent memristive structures that combine optical transparency, stable bipolar resistive switching, and compatibility with crossbar array technology. A key challenge is to establish patterns in the influence of ZnO thin film deposition modes on their structural and electrical properties, which determine the characteristics of memristive structures. The aim of this study was to determine the optimal RF magnetron sputtering power for a ZnO ceramic target, ensuring the formation of transparent ITO/ZnO/ITO memristive structures with stable resistive switching, and to create a crossbar array based on these structures. ZnO thin films were deposited using RF magnetron sputtering at powers ranging from 25 to 100 W. Structural (SEM, AFM) and electrical (Hall effect) studies of the resulting ZnO films were conducted. Transparent ITO/ZnO/ITO memristive structures and a crossbar array of 16 structures with a cell size of 2000 × 2000 nm were fabricated on glass substrates using magnetron sputtering and lithography, and their current-voltage characteristics were measured. Increasing the magnetron sputtering power from 25 to 100 W resulted in an increase in the grain size from 12,8 to 35,7 nm and in the surface roughness of the ZnO films from 2,8 to 11,4 nm. At a sputtering power of 75 W, the charge carrier concentration in the ZnO films reached a maximum value of 2.7 × 1015 cm-3, which is necessary for stable resistive switching of the structure. The obtained ITO/ZnO/ITO memristive structures were shown to exhibit stable bipolar switching for 1000 cycles between the states HRS = 537,4 ± 26,7 Ohm and LRS = 291,4 ± 38,5 Ohm (HRS/LRS ratio ~ 1,8). The fabricated transparent crossbar array showed stable resistive switching for 20000 cycles (LRS = 13,8 ± 1,4 kOhm, HRS = 34,8 ± 2,6 kOhm, HRS/LRS ratio ~ 2,5). The obtained results can be used in the development of technological processes for the fabrication of transparent memristor crossbars for neuromorphic structures of machine vision in robotic systems

The hybrid convolutional neural network architecture with combining spectral and spatial layers, as well as new methods of subsampling (WalsPooling) and convolution (ConvWals) are proposed. The developed system is used to geographical proximity assess of images pair based on their visual similarity. A pair of different sensors obtained images visual similarity determination is complicated by different scales and sensor tilt angles shooting conditions. Based on the low-altitude image fragment, a search in the database of underlying surface images is performed. The search is performed in the surrounding area of a given route based on the vector of image features, which is formed on the last layer of the convolutional neural network. The system uses the Siamese architecture, since a pair of images must be submitted to the input. The relevance of this problem stems from the need to ensure UAV navigation in the absence or unreliability of a GPS signal. The approach to data set formation and its preprocessing is also considered. The database search is performed in the surrounding area of the route, which reduces computational costs. The experiments include an analysis of the applicability of the proposed layers (WalsPooling, ConvWals) and a comparison with traditional pooling and convolution methods. The paper also presents a linear approximation method with trainable parameters for reducing the dimensionality of the convolutional layer. The main advantage of the approach is its resistance to changes in the scale and angle of shooting due to a combination of spectral and spatial features. The results demonstrate the applicability of the method for UAV navigation in conditions of loss of GPS signal is lost or unreliable. The experiment demonstrated that using images reconstructed after spectral transformation yields the best neural network convergence and mean square error. The developed architecture demonstrates robustness to geometric and brightness distortions, and its quality metrics (Precision = 0.728, Recall = 0.800, F1 = 0.872) confirm the effectiveness of the approach for visual localization tasks based on images from a surface database.

The emergence of large language models has become an important milestone in the field of natural language processing, as such models demonstrate impressive results in text generation, transformation, and analysis, as well as in solving a wide range of applied tasks. However, despite significant practical success, large language models possess limited reasoning capabilities. These limitations manifest in difficulties with generalizing knowledge beyond the training distribution, challenges in transferring knowledge to new contexts, and reduced accuracy when performing multi-step logical and mathematical operations. The goal of this work is to examine methods for improving the reasoning abilities of large language models, where reasoning is understood as the process of forming and evaluating inferences based on existing information. The paper discusses the main types of reasoning relevant to large language models: mathematical, logical, and commonsense reasoning. It provides a list of the most commonly used benchmarks applied to assess the reasoning quality of language models. An overview is presented of the methods used to enhance reasoning in large language models at 2025. Depending on the stage of application (during training or during model usage), the work examines approaches to training data preparation, architectural modifications of language models, training and finetuning procedures (including those using specially constructed synthetic datasets), reinforcement learning, various chain-of-thought construction techniques, mechanisms for integrating external tools, and multi-agent approaches. The paper also discusses existing limitations of large language models, which include the lack of conceptual understanding, poor out-of-distribution generalization, and reduced effectiveness as task complexity increases. Finally, the most promising methods aimed at improving the quality and reliability of reasoning in large language models are highlighted.

Problem Statement: The increasing demand for higher quality products in small-scale production and the complexity associated with the early detection of manufacturing defects necessitate the development of innovative approaches to predict and control defects at the early stages of manufacturing complex technical objects. Traditional methods applied in mass production settings are unsuitable for small-scale manufacturing due to the high variability of technological processes and the insufficient data required for conventional statistical analyses. The objective of this study is to reduce the incidence of manufacturing defects by identifying deviations at the preparatory stages of production. The proposed solution involves employing neuro-fuzzy systems capable of adaptively forecasting defects based on historical production data. Methods: To address the early detection of manufacturing defects, neuro-fuzzy components based on the fuzzy neuron proposed by Kwan–Cai were employed, integrating expert knowledge with production data. The system includes a training and fine-tuning subsystem consisting of modules for data preparation, validation and normalization, fuzzification of data, and calculation of forecasting errors. Temporal neuro-fuzzy Petri nets were used as structural forecasting elements, enabling the consideration of temporal aspects and uncertainties inherent in manufacturing processes. Novelty: The novel aspects of this research include the utilization of temporal neuro-fuzzy Petri nets and neuro-fuzzy components based on the Kwan–Cai fuzzy neuron, enabling the early detection of defects and the implementation of proactive measures. Another innovative aspect is the approach for integrating neuro-fuzzy methods into existing production management systems. Results: The implementation of the proposed methods resulted in a 15% reduction in manufacturing defects through early identification of deviations and the timely adoption of corrective actions. Developed software tools provide operational analysis of production situations and defect forecasting in near real-time. Practical Significance: The presented solution has been realized as specialized software integrated into existing production systems. It improves the effectiveness of quality management, reduces the costs associated with defect correction, and can be adapted to various small-scale production environments, significantly enhancing their operational performance.

This paper addresses the problem of automatic recognition of Russian Sign Language (RSL) using computer vision and deep learning methods. The relevance of the study is driven by a steady increase in the number of people with hearing impairments: according to the World Health Organization, there are currently about 70 million deaf and hard-of-hearing individuals worldwide, and this number is projected to reach 630 million by 2035. The development of effective gesture recognition algorithms is an important direction for creating contactless human–machine interaction systems aimed at improving accessibility of information technologies and enhancing the quality of life for people with hearing disabilities. The aim of the study is to develop and experimentally validate an algorithm for real-time recognition of Russian Sign Language alphabet gestures in a video stream using a convolutional neural network. A specialized dataset was created, consisting of 430 images of hand gestures corresponding to the letters of the RSL alphabet, captured from different angles and under varying lighting conditions. The model was implemented using TensorFlow and Keras libraries, while integration with the video stream was performed using OpenCV and a marker-based hand tracking system. As a result of training and testing, the proposed model achieved a recognition accuracy of 99% on the test dataset. A comparative analysis with classical machine learning methods demonstrated the superiority of the convolutional neural network in terms of classification accuracy and robustness to external noise. The obtained results confirm the effectiveness of the proposed approach and its applicability for real-time systems intended for communication, educational, and rehabilitation applications, as well as for the development of advanced human–machine interaction interfaces.