We consider single-channel blind separation of two PSK signals with unknown parameters. This problem is particularly important for communication systems where two signals occupy the same frequency band. It is shown that the blind separation problem can be reduced to a multi-class classification problem, which can be solved using modern machine learning methods. A four-layer neural network is proposed to solve this problem. The effectiveness of this method has been verified using BPSK and QPSK signals under various amplitude ratios and noise levels. It is also shown that in the absence of one of the signals, the proposed separation method effectively demodulates the present signal.
We formulate a fuzzy two-stage continuous–discrete optimal set partitioning problem, which generalizes, on the one hand, the classical finite-dimensional transportation problem to the case where production volumes at given locations are unknown in advance and are determined as the solution to the corresponding fuzzy continuous problem of optimal partitioning of a set of consumers (suppliers of a continuously distributed resource) into fuzzy subsets (the service areas of these points), and, on the other hand, discrete two-stage production–transportation problems for the case of a continuously distributed resource. Theorems regarding the conditions for the existence and form of the optimal solution to the formulated problem are proved. A method for solving it is developed based on the synthesis of the theory of optimal set partitioning and fuzzy set theory.
The convergence of a stochastic optimal control problem with Markov switches in an averaging scheme is analyzed. A combination of the standard Wiener process and a uniformly ergodic Markov process is considered, allowing us to describe the system’s evolution under the influence of diffusion noise and random mode switches. The ergodic properties of the fast process and the conditions for the regularity of the coefficients are presented, which guarantee stable behavior of the system on average. The convergence in probability of the trajectories of the original system to the solutions of the marginal averaged system is proved (through Lp-estimates). The obtained results enable the solution of stochastic optimization and optimal control problems.
To improve the positioning accuracy and management efficiency of laboratory equipment, this study proposes an efficient positioning management system based on an improved Bat Algorithm (BA) and Radio Frequency Identification (RFID) technology. Findings demonstrate that the improved BA converges after 120 iterations, with its accuracy and recall approaching 0.982 and 0.956. In the efficient positioning management system for laboratory equipment, its positioning accuracy is superior to the comparative system, while its response time is only 0.031 s. This research can avoid equipment idleness or overuse, thereby optimizing the allocation and utilization efficiency of laboratory resources.
The authors investigate the methods for solving the transcendental Rankine–Hugoniot equation for determining the characteristics of a supersonic and hypersonic input device. The lower and upper bounds for the shock wave angle have been established, which define the existence interval of the unique solution to the Rankine–Hugoniot equation. The convergence rates of the dichotomy, chord, and secant methods, as well as the Newton method, are compared, and it is found that the secant and Newton methods are the most efficient in terms of the number of iterations. Experimental results show that the secant method should be preferred, as its iteration complexity is lower than that of the Newton method.
A general methodology is developed for identifying the spatial characteristics of low-dimensional systems with arbitrary geometric configurations using machine learning methods applied to datasets obtained from experimental measurements. A numerical approach is developed that implements mathematical models for spectral problems and allows for the unified representation of training dataset inputs in the form of normalized tabular dependencies. A neural network trained on 560 datasets is applied to identify the precision parameters of nanosystems and their confinement. The results obtained, compared to those based on direct numerical methods, are characterized by a 30- to 40-fold reduction in computation time and consistently high accuracy. The developed approach has prospects for application as a tool that automates work with large volumes of experimental and computational data, in nanotechnology and electronics.
The authors propose a zero-shot prompting-based source code generation technology and introduce an overall metric for evaluating its overall quality. The metric comprises four groups of indicators: full model confidence, semantic quality, structural integrity, and dynamic code execution. An empirical study of two types of prompt structures is conducted. The proposed metric enables a reproducible, comprehensive quality indicator that integrates functional correctness with maintainability and code reliability. The results provide a foundation for the further standardization of prompt engineering and the development of objective evaluation methodologies for LLM-generated source code in real-world information systems.
The Ukrainian National Standard DSTU 9041:2020, adopted in 2020, defines a hybrid encryption algorithm that is secure against attacks aimed at recovering the key and the message, but is neither IND-CCA-secure nor secure against small-subgroup attacks. The paper presents a modification of this algorithm that is secure against both IND-CCA and small-subgroup attacks and is also consistent with the current National Standards of Ukraine.
This paper investigates and compares methods for converting long-wave infrared (LWIR) images into visible-spectrum (RGB) representations using a baseline reconstructive U-Net model and the conditional GAN model pix2pix with a local discriminator. The study implements a software platform as a reproducible image-to-image translation pipeline based on the U-Net and pix2pix architectures, applied to supervised learning tasks of the “image-to-image” type. The paper presents the results of an experimental study of LWIR-to-RGB image conversion using U-Net and pix2pix models trained on paired data from the KAIST dataset. The experiments evaluate and compare composite loss functions with weighting coefficients that incorporate the structural similarity index and the mean absolute difference between predicted and reference images. The proposed loss functions combine global image context awareness with the preservation of local features. The study presents training curves for the U-Net and pix2pix models, providing qualitative insight into training dynamics and convergence behavior. The results show that the U-Net model achieves high-quality structural reconstruction. The adversarial component of the pix2pix model, implemented with a PatchGAN local discriminator, generates perceptually realistic images at multiple levels of detail while preserving structural fidelity. The paper also outlines directions for improving the reliability of converting single-channel thermal images into three-channel RGB representations based on the investigated models.
Based on the Ito–Skorokhod stochastic differential equation, the authors propose the construction of a model of a two-threshold process and an approximate maximum likelihood method based on the approximation of the logarithmic likelihood function of observations. Estimates of the parameters of a three-regime threshold jump process with discretely selected data are obtained. The model presented in this study takes into account three components: drift, diffusion, and jumps, which make it possible to estimate both moderate and abrupt changes occurring during the process. The process is divided into three regimes, and the parameters are estimated for each interval. This approach allows analyzing both simple and complex processes with hierarchical dynamics. The developed algorithm is based on maximum likelihood estimation. The algorithm is constructed as an iterative procedure that calculates the parameters at each step, taking into account the current threshold values. The calculation ends upon convergence.
Based on the conducted research, it has been shown that diffusion models have significant potential for image generation tasks. However, in their basic form, these models often produce outputs in which key object characteristics are not well aligned with the target domain. To address this issue, the study investigates the use of Low-Rank Adaptation (LoRA) to improve alignment between generated content and domain-specific requirements, using datasets of explosive ordnance images as a case study. The experimental results include hyperparameter tuning, which demonstrates the effectiveness of specific learning-rate ranges. In addition, the study examines how the adaptation rank and related parameter a influence the stability of the adaptation process. Several limitations were identified, including the dependence of generation quality on dataset composition, difficulties in controlling visual artifacts, and the limited expressiveness of standard evaluation metrics. Finally, the study outlines directions for future research, namely automating parameter selection, developing more reliable methods for assessing generated outputs, and evaluating the practical effectiveness of synthetic datasets for training explosive ordnance detection models.
Mobile ad-hoc networks (MANETs), which stipulate operations with random topology and decentralized management, are widely used both for organizing military communication and in the civilian sphere, in particular, during the elimination of consequences of hostilities, natural disasters, etc. An important problem when designing new radio communication systems (means) or improving existing ones is to create an effective management system for MANETs (in what follows, mobile radio networks), which is much harder compared to the centralized management of classic communication systems . The article substantiates the principles of constructing a system of complex operational management of a mobile radio network, proposes its functional model, determines the purpose of individual components, and formalizes the algorithm of work.
Core-construction methods reduce the number of alternatives subject to comparison and improve the reliability of their subsequent ranking. This paper describes a tri-criteria Euclidean core method, on the basis of which a comprehensive method for constructing the core of alternatives is developed. The proposed method enables the construction of four cores within a single algorithm by employing different combinations of comparison criteria. Its practical applicability is demonstrated through a numerical example. It is also shown that the presence of Pareto-inefficient alternatives in the initial set, as well as the choice of normalization approach, may affect the resulting solutions.
The paper makes a short overview of advanced systems analysis methods, models and modeling tools being developed at IIASA (International Institute for Applied Systems Analysis, Laxenburg, Austria) and within NASU (National Academy of Sciences, Ukraine) and IIASA joint project “Integrated modeling for robust management of food–energy–water–social–environmental nexus security and sustainable development.” Emerging systemic risks in interdependent Food–Energy–Water–Environment (FEWE) systems can be managed through a two-stage coherent decision-making framework: ex-ante (anticipatory) and ex-post (adaptive), using integrated models to balance proactive risk reduction (e.g., resilient infrastructure, diversified resources) with reactive crisis response (e.g., emergency planning, technological and financial backstops) for increased resilience, as highlighted by IIASA and NASU joint research. This approach, using two-stage stochastic optimization, aims for robust management by keeping options open while preparing for inevitable uncertainties in these complex, interconnected systems. Truly integrated modeling often requires rescaling (down- and up-scaling) of models’ data and results. The mismatch of scales creates a major source of uncertainties, which calls for the identification of proper indicators, new measures of uncertainties and risks, and goodness criteria for disaggregation and aggregation. To represent information in locations, the procedures rely on an appropriate optimization principle, e.g., generalized cross-entropy maximization, and combine the available samples of real observations in the locations with other “prior” hard and soft data (expert opinion, scenarios), pseudo-sampling models, evidences on the related variables that exist in the form of equations and constraints. A key issue is treatment of uncertainties in priors and parameters of available constraints. Approaches to down-scaling in the presence of uncertain priors are outlined. The approaches are being further developed at IIASA and the NASU–IIASA joint project. Distributed models’ optimization and linkage methods enable to establish relationships and dialogues between separate models of FEWE systems for the analysis of coordinated solutions without requiring to share or reveal systems-specific information, i.e., under asymmetric information. The problem is illustrated with an example of linking models of individual producers emitting GHGs (emitting entities or parties) into a prototype model of an emission trading market when information about parties may not be available and joint safety constraints on emissions (when individual parties’ emissions are uncertain) have to be fulfilled. The outlined methods and tools pursue the goal to develop and implement advanced systems analysis and integrated modeling approaches allowing coherent planning of FEWE systems under joint constraints, asymmetric information and uncertainties about the sectoral models. Explicit modeling of linkages allows evaluation and treatment of such risks under standard independent planning of sectors. Therefore, the models and methods aim for systems analysis of FEWE nexus security under exogenous risks and risks affected (intentionally and unintentionally) by decisions of various agents. The methods and tools involve the concept of robustness and robust solutions, which are, in a sense, optimal for any scenario of potential uncertainties.
A general approach to determining the parameters of a collateralized protocol is considered, along with the development of an appropriate mathematical framework that accounts for all major risks. The following problems are solved: all key factors that traditionally affect asset price and the size of payouts during the position liquidation procedure are identified; risks are incorporated and their potential impact on the price is assessed; the statistical data that must be collected and processed to obtain numerical characteristics of these risks is specified; relevant statistical data is collected across various decentralized platforms and, based on them, a liquidation condition for a collateral position is formulated and justified; the LTV is determined and substantiated; numerical results for LTV are obtained for different pairs of assets.
The purpose of this research is to explore the complex dynamics and impact of vaccination in controlling malware outbreaks. The classical epidemic compartmental model is formulated by introducing vaccination nodes. Initially, the proposed model is analyzed quantitatively. The basic reproduction number is computed, and its numerical values are estimated using the next generation method. The sensitivity analysis is performed to analyze the contribution of the model-embedded parameters in the transmission of the malware. Further, the equilibrium points are computed, and the local as well as global stability is discussed. The numerical simulations are performed describing the impact of various scenarios of vaccine efficacy rate, and other patches are measured. Further, on the basis of sensitivity analysis, the proposed model is restructured to obtain an optimal control model by introducing a time-dependent control variable for vaccination efficacy and treatment enhancement of the network. Finally, the graphical interpretation of each case is depicted and discussed in detail. The simulation results revealed that the developed vaccination e-epidemic compartmental model was able to provide an appropriate defensive mechanism as well as security in the network. Deployment of this model leads to the minimization of malware propagation in the network.
An empirical comparative study of four document segmentation strategies is presented, namely, of fixed windows of 256, 512, and 1024 tokens, and semantic segmentation based on a large language model. Experiments were conducted on long semantically coherent texts from the SQuALITY dataset. The evaluation was performed on 225 question-answer pairs using Precision@5 and Recall@5 (top-5 retrieval metrics), answer quality metrics (Exact Match and token-level F1), and average retrieval latency. The results reveal a clear trade-off between retrieval precision and recall driven by granularity where smaller fragments provide higher precision, whereas larger fragments substantially increase recall and improve answer quality in terms of F1. Within this experimental setting, semantic segmentation demonstrates competitive results, but does not show a consistent advantage over fixed windows of 512–1024 tokens. A reduction in retrieval latency is observed when using larger segments, which can be explained by lower vector-index density. A reproducible evaluation procedure and practical recommendations for selecting a segmentation strategy for efficient RAG systems are provided.
A method and a corresponding algorithm for iterative partitioning of images into rectangular pixel blocks with different color models, along with their compact representation in progressive hierarchical compression, are proposed. A procedure for selecting an effective differential color model with integer coefficients from a set of alternative base models is developed for both the entire image and individual rectangular blocks. The selection is based on an entropy analysis and aims to achieve near-optimal compression performance. It is emphasized that applying differential color models to rectangular blocks in archivers maximizes the efficiency of lossless image compression.
A two-stage formalized model for monitoring and early warning of complex social crises is proposed, combining system-dynamic modeling of an integral crisis pressure index C (t ) and its probabilistic interpretation based on a dynamic Bayesian network. The model accounts for the nonlinear interactions of social, economic, security, and informational factors and the damping role of social resilience. A formalized mechanism for transitioning from temporal dynamics C (t ) to multilevel early warning signals is proposed. Testing the model on Ukrainian data for 2022–2025 confirmed its suitability for identifying phases of latent escalation and quantitatively assessing crisis risks.
A generalized Poisson integral in an orthogonal system of polynomials is presented as a mathematical model of admissible control strategies under the influence of a three-dimensional control vector. An optimality criterion is constructed in the form of an extremal problem on the deviation of a generalized Poisson-type operator from the boundary optimal state, which is modeled by functions of the Hölder class. The solution to the optimization problem is presented as an asymptotic equality that describes the behavior of the controlled dynamic system over the entire set of admissible control strategies.
