Computer systems and information technologies

PCA METHOD IMPACT ANALYSIS ON DEEP NEURAL NETWORKS ACCURACY AND ARCHITECTURE FOR IMAGE CLASSIFICATION

Anastasiia NESKORODIEVA — Thu, 26 Mar 2026 00:00:00 +0200

The growth of visual data in modern information systems creates a need to reduce the computational complexity of classification methods. This paper presents the impact study of data preprocessing by the Principal Component Analysis (PCA) method on the effectiveness of training and classification ability of deep neural networks. The focus is on a comparative analysis of two architectures — the Multilayer Perceptron (MLP) and the Convolutional Neural Network (CNN). The work is aimed at solving the urgent scientific and practical problem of optimizing computing costs in computer vision systems operating in conditions of limited hardware resources. Based on the analysis of experimental data obtained on the MNIST image set, a comparative analysis of four architectural approaches was performed: classical MLP, classic CNN, hybrid PCA+MLP, and hybrid PCA+CNN. The effect of loss of spatial locality during image transformation by the PCA method, which leads to deterioration of the results of CNN-based models, as opposed to reduction of computational complexity of MLP-based models while maintaining classification accuracy, is studied. The paper provides a combined image classification method, an analysis of the obtained accuracy metrics and loss function, as well as a justification of the observed phenomena. The conducted study allows us to draw conclusions about the feasibility of using the PCA method in image classification problems in combination with MLP. The results show the importance of aligning data preprocessing methods with the architectural features of machine learning models. In both hybrid models, the number of parameters was reduced by 4 times, while the PCA+MLP training time was reduced by 2 times with an accuracy of 97.87%, and the PCA+CNN training time was reduced by 8 times with an accuracy of 95.02%. The PCA+MLP hybrid model is found to have 4 times lower computational complexity and 12 times less training time than CNN, with a small accuracy loss of 1.5%.

SEAMLESS TILING OF QUASI-PERIODIC TEXTURES VIA AN OPTIMAL CYCLIC SHIFT ON A DISCRETE TORUS

Anna BEDRATYUK — Thu, 26 Mar 2026 00:00:00 +0200

In practical computer vision and computer graphics pipelines, it is often necessary to repeatedly replicate a single texture sample to construct a large canvas, background, or regular covering. When the mosaic is not strictly periodic, visible seams appear at the boundaries during repetition, disrupting the perceptual continuity of the texture and often manifesting as a regular grid of artifacts. Such seams not only degrade visual quality but can also alter local gradients and spectral components, which is critical for subsequent processing stages. Common seamless stitching methods increase computational complexity, introduce additional hyperparameters, and modify the local image statistics, which is undesirable in reproducible pipelines and in tasks where the invariance of pixel values is essential. The goal of this work is to propose a simple, reproducible, and computationally efficient method for seam reduction in quasiperiodic textures by selecting an optimal cyclic shift of the pattern that minimizes the energy of mismatch between opposite boundaries. The tile is modeled as a function on the discrete torus . A cyclic shift group is introduced, acting as a permutation of pixels. For each shift , the boundary seam energy is computed in a band of width for opposite boundary pairs, and the minimizing shift is selected. When needed, the evaluation is accelerated via cyclic correlations and FFT. Experiments on synthetic and real textures show that the optimal cyclic shift significantly reduces seam energy and the visual prominence of boundaries during tiling without modifying pixel values. For strictly periodic tiles, the method does not degrade the result. The proposed approach is a lightweight baseline tool for seamless tiling: it does not perform stitching but selects the best cut of the torus. The method is easy to integrate into production pipelines and can be used as a preprocessing step before further processing.

DEVELOPMENT AND RESEARCH OF MULTIMODAL NEURAL ARCHITECTURES FOR HETEROGENEOUS UNBALANCED DATA IN CLASSIFICATION TASKS

Serhii MINUKHIN, Valerii RUDOI — Thu, 26 Mar 2026 00:00:00 +0200

The article presents a comprehensive study of modern multimodal neural architectures for integrating heterogeneous and partially unbalanced data in classification tasks. It considers early and late fusion approaches, hybrid architectures with cross-modal attention, and transformers that allow the formation of consistent latent spaces of visual, auditory, and textual features. Particular attention is paid to contrastive learning (CLIP-like approaches, multimodal InfoNCE), which ensures semantic consistency of representations and improves classification accuracy in the presence of uneven data distribution and rare classes. A model is proposed that combines early and late fusion with cross-modal attention and contrastive learning to form a coherent joint latent space. Features of each modality are processed by specialized encoders, and fusion is performed with adaptive weighting, which minimizes the impact of heterogeneous data imbalance and enables the efficient processing of signals of different natures and intensities. The use of pruning, quantization, and knowledge distillation has reduced computational costs without losing accuracy, ensuring stable model performance in real-world streaming scenarios with limited resources. The results of applying the proposed model to the BDD100K and CMU-MOSEI datasets confirmed the model's high efficiency in processing heterogeneous and unbalanced data. For BDD100K, Accuracy 0.953, F1-score 0.956, ROC-AUC 0.947 were achieved, and the integral indicators Micro F1, Macro F1, and Weighted F1 were 0.953, 0.949, and 0.955, respectively; For CMU-MOSEI, Accuracy 0.956, F1-score 0.969, ROC-AUC 0.968, and the integral indicators Micro F1, Macro F1, and Weighted F1 were 0.956, 0.962, and 0.968, respectively. A comparative analysis of metrics with classical methods,SOTA solutions and AutoML (B-T4SA proved that the developed model provides consistently higher accuracy and consistency of classification for all classes, including rare ones, confirming its ability to effectively adapt to high variability and imbalance of heterogeneous data in real conditions.

MODELING THE PROCESS OF RECOGNITION OF PACEMAKER DYSFUNCTION

Dmytro MEDZATYI, Illia HRYSHCHUK — Thu, 26 Mar 2026 00:00:00 +0200

The article presents a comprehensive study aimed at solving a pressing scientific and practical problem - modeling and designing information technology for recognizing pacemaker dysfunctions to increase the efficiency of diagnostics and reliability of life support systems. The relevance of the work is due to the rapid growth of the number of cardiovascular diseases in the world and in Ukraine in particular, which leads to an increase in the number of operations for implanting pacemakers, the functioning of which requires continuous and high-precision monitoring. The authors analyzed the world experience in using modern diagnostic tools, including neural networks for analyzing radiographs, mobile applications for remote monitoring, and machine learning algorithms for ECG analysis, which revealed the lack of integrated solutions that would combine different methods for detecting technical and clinical failures. The proposed approach is based on the use of multimodal input data, such as information about the patient's symptoms (dizziness, arrhythmia, weakness), device hardware reports (pacing rate, battery status, intracardiac signals), ECG and Holter monitoring results, as well as data from physical activity and intracardiac pressure sensors. The scientific novelty of the study lies in the development of a mathematical model of the process of recognizing pacemaker dysfunction, presented as a sequence of tuples and transformations that provide data preparation, selection of the most informative signs of cardiac activity and direct recognition of the system state. Special attention is paid to the stages of signal normalization and artifact filtering, which guarantees high accuracy of classification of disorders even under difficult operating conditions or during physical exertion of the patient. The practical significance of the work is confirmed by the creation of a structure of output results, which include not only automated fixation of anomalies, but also the formation of specific recommendations for changing pacemaker settings and instant notification of medical personnel, relatives and the patient himself. The proposed technology allows to ensure a continuous monitoring cycle, minimize the risk of human error when interpreting complex diagnostic data and significantly improve the prognosis for patients with high dependence on an artificial pacemaker. Thus, the results obtained create a reliable foundation for building modern information technologies for cardiac care.

HYBRID INTEGRATION OF EXPERT SYSTEMS INTO AVIATION LOGISTICS MANAGEMENT STRUCTURES

Oleh SYDORENKO, Nataliia LYSA — Thu, 26 Mar 2026 00:00:00 +0200

In this study authors considered the integration of expert systems into management of aviation logistics. The complexity of this integration is in the management structures and in the context of the operational complexity which is constantly increasing and the continuous supply of chain operations. It was determined that the existing systems and methods to make a decisions are not full sufficient with big data processing and do not taking into account the dynamic changes or does not fully smooth out the risk factors in aviation logistics. It was agreed that the traditional methods are not enough effective when the tasks are to process a big amount of data in conditions of dynamic changes. It was detected that the integration of expert systems which are able to simulate professional knowledge and operational expertise making a big impact on the decisions quality and reduces the human factor in critical operational time and provides adaptive and flexible planning. In this study authors are also analyze and compare several modern systems, such as NAVBLUE, Ramco Aviation Suite, Veronte Autopilot, ARTAS, FANS-1/A, TCAS II, RUN.S.A.F.E., SAREX, AIR Cargo handling System Assessment Model and Astrea UAV Integration Platform. The authors found that, regardless of the individual functional capabilities of each system, modern aviation expert systems are quite fragmented and do not have unified architectural approaches. This leads to the need to develop a hybrid integration model that will be focused on combining expert system cores with centralized logistics planning. In this article, the authors proposed new architectural and methodological approaches to building a modular, standardized architecture that unites distributed knowledge, synchronizes analytical processes, and provides unified integration into digital platforms such as ERP, MES, TMS. The practical meaning of this research is to find possible transformations of existing aviation logistics practices using unified hybrid decision support systems and in managing the design of future digital aviation infrastructures.

ENERGY–AWARE MODELLING OF IOT NETWORK LIFE–CYCLE UNDER INDUCED FALSE–EVENT FLOWS

Chen YU, Viacheslav KOVTUN — Thu, 26 Mar 2026 00:00:00 +0200

An energy-aware analytical model of the IoT network life-cycle under induced false-event traffic is proposed. The study considers event-driven clustered IoT networks operating under external influences that generate false event messages and thereby cause unnecessary sensing, transmission, reception, and forwarding operations. Unlike conventional approaches that usually treat traffic behaviour, communication energy consumption, and network geometry separately, the proposed model integrates these components within a unified formal framework. Within this framework, the energy cost of a single false event is formalised and linked to the network-level energy balance, false-event arrival parameters, and node mobility. On this basis, closed-form analytical expressions are derived for the temporal evolution of residual energy and for the duration of the network life-cycle under fixed spatial topology. The model thus establishes an explicit relationship between induced false-event intensity and the depletion of network resources. The analysis shows that the intensity and regularity of false-event arrivals significantly affect the degradation trajectory. In particular, more regular induced traffic changes the early-stage depletion pattern, whereas at high intensities different traffic regimes converge. The model is validated by simulation for a LEACH-based clustered IoT network. The simulation results confirm the analytical dependencies over the investigated range 1 ≤ λ_f ≤ 10 and show that the proposed formulation remains in close agreement with the simulation reference. In the comparative analysis, the prediction error of the proposed model remains within about 0%-2%, whereas the traffic-only baseline reaches about 10.2% and the classical LEACH baseline about 20.4%. These results demonstrate that the proposed model provides a more adequate estimate of network life-cycle because it explicitly incorporates induced-loss effects ignored or oversimplified in conventional approaches.

TECHNOLOGY FOR SELECTING MUSICAL GENRES TAKING INTO ACCOUNT HUMAN MENTAL HEALTH BASED ON MACHINE LEARNING

Vitalii ALEKSEIKO, Olena PETIAK, Bohdana BONDARCHUK, Tamara PETRUK — Thu, 26 Mar 2026 00:00:00 +0200

The research considers the technology of intelligent selection of musical genres taking into account human mental health based on machine learning methods. The relevance of using music therapy as an effective non-drug approach to improving emotional state, reducing stress, anxiety and preventing psychoemotional disorders is substantiated. Modern scientific research on the influence of music on mental health is analyzed, as well as existing approaches to recommender systems in the field of healthcare. A comparative analysis of the performance of various machine learning methods for the task of classifying and predicting the user’s psychoemotional state is conducted, based on the results of which the most relevant algorithm is selected. Hyperparameters are selected and optimized in order to increase the accuracy, stability and generalization ability of the model. A concept of a system that provides personalized selection of musical genres according to the individual psychological characteristics of the user is proposed. The compliance of the developed technology with the principles of explanatory and responsible artificial intelligence is outlined, in particular with regard to the transparency of decisions, the ethics of data use and the minimization of potential risks. The consistency of the research results with the UN Sustainable Development Goals is shown, in particular in the context of ensuring well-being, mental health and access to innovative digital technologies in the field of healthcare.

MULTI-FACTOR HYBRID APPROACH FOR BLOCKCHAIN-ENABLED IOT EDGE COMPUTING

Leonid CHEPEL, Yuriy BOYKO — Thu, 26 Mar 2026 00:00:00 +0200

The integration of Internet of Things (IoT) devices with edge computing supports applications in industry, healthcare, and smart cities. Edge computing moves data processing closer to the data source, reducing reliance on central cloud infrastructure. However, managing heterogeneous and geographically distributed systems introduces challenges in coordination, software distribution, and trust establishment. Centralized approaches create single points of failure and limit scalability. Blockchain technology addresses trust and coordination issues. Nonetheless, existing implementations lack on-chain coordination with real-time resource metrics and do not support load-aware task assignments.

This work introduces a hybrid blockchain-IoT edge computing platform. The platform integrates multi-factor node selection based on node reputation and resource availability. The developed approach includes timeout management, queue capacity control, thread-safe concurrent transactions, and event-driven task orchestration with staleness detection. Off-chain computation is separated from on-chain coordination. Node selection uses a scoring function that combines success rate with real-time resource metrics. Coordination logic is implemented through Ethereum-based smart contracts, ensuring transparency and immutability. Validation was conducted in a simulated Docker network environment.

Testing shows that nodes with higher scores are selected in all observed cases. However, when multiple tasks are submitted within the same inter-block period, node scores do not update as expected. Selecting nodes based on the following block rather than the current one improves selection reliability. The platform addresses gaps identified in existing literature and can be used in industrial edge computing scenarios.

CROSS-LINGUAL TRANSFORMER-BASED SCREENING OF POST-TRAUMATIC STRESS DISORDER BASED ON COMPARATIVE ANALYSIS OF BERT AND XLM-ROBERTA WITH MACHINE TRANSLATION ADAPTATION FOR UKRAINIAN LANGUAGE

Andrii FEDORYCHKO, Victoria VYSOTSKA, Lyubomyr CHYRUN — Thu, 26 Mar 2026 00:00:00 +0200

The paper presents a comprehensive study on automated screening of post-traumatic stress disorder (PTSD) using transformer-based natural language processing models in a cross-lingual setting. The research aims to evaluate the feasibility of deploying an intelligent PTSD detection system in Ukraine under conditions of limited, localised training data. A balanced corpus of 4,822 text records was constructed by aggregating publicly available PTSD-related datasets, including 2,042 PTSD-positive texts and 2,780 control texts representing neutral content and other psychological conditions. The study compares the performance of the English-language BERT (bert-base-uncased) model and the multilingual XLM-RoBERTa (xlm-roberta-base) model applied to a Ukrainian-language corpus generated via machine translation using the Google Translate API and large language models for complex structures. Word cloud visualisation and semantic analysis confirmed preservation of core psychological markers during translation. Experimental results demonstrate high predictive performance for both architectures. The English-language model achieved an Accuracy of 0.90 and an ROC-AUC of 0.962. In contrast, the Ukrainian-language model achieved an Accuracy of 0.85 and an ROC-AUC of 0.940, significantly outperforming existing Ukrainian multi-class stress detection models (Accuracy ~0.45) and exceeding standard multilingual mental health benchmarks (0.78–0.82), establishing a robust state-of-the-art baseline for Ukrainian clinical NLP. Importantly, Recall remained identical (0.88) across both language settings, indicating strong sensitivity to PTSD markers despite translation-induced lexical noise. The minimal AUC degradation (2.3%) confirms the robustness of transformer architectures to cross-lingual adaptation. The findings validate the viability of combining machine translation with multilingual transformers for the rapid deployment of mental health screening systems in low-resource language environments. The proposed pipeline enables scalable and cost-effective digital PTSD monitoring while maintaining clinically relevant diagnostic sensitivity.

ADAPTIVE NEURAL STABILIZATION OF ILL-POSED SPECTRAL PROBLEMS

Fedir SAIBERT, Yurii BILAK — Thu, 26 Mar 2026 00:00:00 +0200

The article develops a mathematical model of adaptive neural regularization and creates an algorithmic method for its implementation with a neural network for dynamic determination of the regularization parameter. The developed ANR model combines classical variational stabilization methods with neural network prediction of the regularization parameter and eliminates the key limitation of traditional Tikhonov regularization, associated with the need to manually select the smoothing coefficient, which makes the solution unstable in the case of variable noise conditions and spectral correlation. The ANR neural network subsystem analyzes the statistical features of the spectrum and adaptively selects the optimal α, ensuring a natural balance between reconstruction accuracy and solution smoothness. Numerical experiments demonstrate that adaptive neural-guided regularization provides a reduction in the root mean square error of reconstruction. For a baseline noise level of 2%, RMSE=0.11 mM, while at a noise level of 10%, it increased to 0.32 mM. It is shown that the use of neural-guided regularization allows reducing the reconstruction error by 20–40% compared to the classical Tikhonov regularization. The model retains its versatility and can be integrated into a wide range of spectroscopic methods, from analytical spectrophotometry to optical materials diagnostics. The work outlines prospects for further development, in particular, the extension of ANR to multichannel and hyperspectral systems and the application of physically informed neural networks to solve more complex inverse problems.

IMPROVED SMALL AGRICULTURAL PLANT SEGMENTATION USING POST-TRAINING ADAPTIVE NEURAL NETWORK

Iryna UDOVYK, Kostiantyn KHABARLAK — Thu, 26 Mar 2026 00:00:00 +0200

Precision agriculture enables an automated and data-driven way of improving agricultural crop yields. Soil nutrition, spraying pesticides against pests and diseases can be applied at a large scale. However, defeating weeds poses a significant challenge. High weed localization precision is required as herbicides kill both weeds and crops, cutting or laser removal also should be performed with care. Robotic devices have been proposed to remove weeds, all relying on neural-network-based weed localization. Robots typically perform on-device processing, without Internet connection. Typically, weed removal should be performed at early stages of growth, so the plants occupy a small part of the image, which makes the segmentation task difficult. Existing weed segmentation approaches have insufficient ratio of quality to computation complexity for edge deployment. In the meantime, it is estimated that weeds are accountable for 31.5% yield loss. To solve the problem of on-device weed segmentation, we propose PAN+PTA semantic segmentation neural network, computational complexity of the network can be adjusted after training in a range from 13,08 to 18,12 GFlops. Consequently, the network can be adapted to a wide range of devices without additional training or costly redeployment. We achieve this by 1) integrating the Post-Train Adaptive (PTA) network as encoder in Pyramid Attention Network (PAN); 2) introducing width multipliers to configure initial capacity of the PTA network. To train and evaluate the neural network we use WE3DS dataset, which contains annotations of 7 crops and 10 weeds. The lightest configuration of PAN+PTA achieves higher Dice Score compared to PAN with MobileNetV2 encoder, while reducing the number of computations by a factor of 1.9. Additionally, the trained network in heavy configuration with width multiplier of 1.5 has Dice Score of 0.5112 and computational complexity can be adjusted in range of 32.34%, which is a substantial improvement over existing U-Net+PTA network (Dice Score: 0.4348, range: 3.66%), while reducing inference GFlops by 80%.

ADAPTIVE MULTICRITERIA MODEL FOR SUPPORTING DECISION-MAKING IN SPORTS SELECTION PROBLEMS

Alina HNATCHUK, Yaroslav HNATCHUK — Thu, 26 Mar 2026 00:00:00 +0200

In the current environment of information technology development, intelligent data analysis, and decision support systems, there is a need to formalize and automate sports selection processes. Existing models of multi-criteria assessment of athletes are usually static in nature and focused only on analyzing the current level of preparedness, which limits their effectiveness in selecting young candidates for whom development potential is important. This paper proposes an adaptive multi-criteria decision support model that integrates the assessment of the current level of candidates' characteristics with formalized consideration of their development dynamics over time. A mathematical apparatus for forming an integral assessment of prospects has been developed, based on the hierarchical aggregation of normalized indicators and the application of the hierarchy analysis method to determine weight coefficients. For the first time, a generalized characteristic indicator has been introduced, which combines normalized indicator values and the rate of their change, allowing not only current results but also development prospects to be assessed. An experimental study on a sample of 24 candidates showed a heterogeneous structure of the distribution of integral assessments, which confirms the feasibility of using adaptive thresholds to classify candidates according to their level of prospects. The sensitivity of the model to changes in the α parameter, which allows adjusting the balance between current indicators and development dynamics, was analyzed. The results confirmed the stability of the model and the consistency of the overall ranking structure of candidates when changing weight parameters. The proposed model ensures transparency, reproducibility, and interpretability of the assessment, which makes it suitable for use in intelligent decision support systems for the selection of young athletes. The practical benefit of the model is that it allows coaches and sports selection specialists to effectively and objectively identify candidates with high development potential, taking into account both their current level of preparedness and the dynamics of their characteristics. The model can be integrated into intelligent decision support systems for planning the training process and strategically forming teams of young athletes.

RESEARCH ON THE AWS LAMBDA PERFORMANCE FOR USER AUTHENTICATION IN CROSS-PLATFORM CLOUD APPLICATIONS

Yurii Gunchenko, Alla Kamienieva, Maryna IEPIK — Thu, 26 Mar 2026 00:00:00 +0200

The paper is devoted to investigating the effectiveness of using the serverless AWS Lambda architecture to implement a scalable user authorization system in a cloud environment. The authorization function is deployed as a cross-platform component that interacts with the managed Amazon Aurora PostgreSQL relational database and uses the AWS Systems Manager service to securely store configurations. The function's performance depending on environment parameters (memory capacity, cold/warm start) was investigated, as well as the JWT token generation implementation for user identification. The results can be applied in the development of cloud applications focused on high scalability, portability, and secure database operations.

THE METHOD OF IDENTIFYING KEY ELEMENTS OF A DIGITAL IMAGE IN THE DECISION-MAKING PROCESS OF CLASSIFICATION BY A NEURAL NETWORK

Oleksandr DORENSKYI, Oleksandr Drieiev, Hanna Drieieva — Thu, 26 Mar 2026 00:00:00 +0200

This paper addresses the problem of evaluating the quality of data used for training neural networks by identifying significant elements of digital images, which the neural network algorithm relies on for classification. As is well known, image classification systems are widely used in computer vision, where entire images are assigned to a single class without distinguishing individual objects within them. This is a typical problem in computer vision. For classification tasks, pre-trained convolutional neural networks (CNNs) are often employed, trained on labelled datasets. However, the unresolved issue remains as to which specific elements the neural network relies on when making a particular decision. The paper presents a method based on a competitive gradient descent process to extract details (elements) that were significant during the classification process, i.e., key elements. This method involves the competition between the process of image detail degradation and the preservation of classification results. Using a self-trained neural network, the authors analyse the presence of details in a classified digital image by visually assessing the elements directly related to the classified object after applying the proposed method. Thanks to this approach, the degradation of digital image details while preserving classification quality is achieved, and the network architecture may be arbitrary. This allows for a comparison of attention areas in neural networks with different architectures: convolutional architecture and mixing architecture. Based on the research findings, a method for localizing neural network attention with arbitrary architecture is proposed. The preserved elements on the degraded image can provide additional information about the validity of the classification performed by a specific neural network. This can be assessed by localizing the preserved elements. For example, the presence of these elements in the classified object indicates a high probability of correct neural network performance. In other words, if the preserved elements do not belong to the classified object, it can be concluded that the training data is not representative (for example, one of the objects may more frequently appear against a characteristic background). Experimental studies demonstrated the advantages of the proposed method over existing alternatives: accuracy in localizing significant details (elements), the presence of information about global significant elements of the digital image and their shape, as well as its applicability for both convolutional and other types of neural networks.

CONVOLUTIONAL NEURAL NETWORK-BASED SOUND SOURCE SEPARATION IN THE TIME-FREQUENCY DOMAIN

Oleh TOMASHEVSKYY, Orest TKACHUK — Thu, 26 Mar 2026 00:00:00 +0200

This paper addresses the problem of sound source separation in mixed audio signals in the time-frequency domain. The study considers the application of convolutional neural networks for isolating individual acoustic components from complex audio mixtures where multiple sources overlap in both time and frequency. The presence of such overlap significantly complicates the separation process and increases the requirements for stability and structural consistency of the applied models. The proposed approach is based on transforming audio signals using the Short-Time Fourier Transform and representing audio mixtures as spectrograms that preserve both temporal and spectral characteristics of sound components. A binary masking strategy is applied to the resulting representations to structurally simplify the separation task. A convolutional neural network is employed to predict masks corresponding to individual sound sources such as vocals, bass, drums, and other components. This masking formulation enables selective extraction of spectral regions associated with specific sources and supports the implementation of a hybrid processing scheme that combines elements of classification and regression within a unified neural architecture. The research methodology includes the design of the network architecture, preparation of spectrogram-based input data, model training on multi-source audio mixtures, and validation of separation quality using reconstruction consistency criteria. Particular attention is paid to ensuring stable convergence of the model and preserving meaningful acoustic patterns within the predicted masks. The findings demonstrate stable isolation of sound components and consistent performance across training and validation datasets. Quantitative evaluation shows separation accuracy of 0.772 for vocals, 0.766 for drums, 0.944 for bass, and 0.764 for other sources, with corresponding mean squared error values ranging from 0.044 to 0.203 across evaluated categories. The highest performance was achieved for bass isolation due to the distinct low-frequency spectral structure of this source. Signal-level evaluation using SI-SDR, SDR, and SNR metrics produced values ranging from -1.24 to 4.10 dB (SI-SDR), -0.26 to 4.59 dB (SDR), and 1.16 to 5.09 dB (SNR), with the highest values observed for bass and vocal sources, consistent with the accuracy-based results. The results confirm the effectiveness of integrating binary masking with convolutional processing of spectrograms for computationally efficient sound source separation. The proposed approach, implemented using a compact neural architecture with 323,233 trainable parameters, can be applied in music production systems, speech enhancement solutions, intelligent audio analysis platforms, and other audio processing environments requiring reliable and lightweight separation mechanisms.

ANALYSIS OF INFORMATION TECHNOLOGIES AND METHODS FOR AUTOMATIC UPDATING OF THREAT DETECTION MODELS IN COMPUTER SYSTEMS

Tymur ISAIEV, Olha ATAMANIUK — Thu, 26 Mar 2026 00:00:00 +0200

The development of intelligent adaptive information technologies for automatic updating of threat detection models in computer systems is one of the most important directions in modern research on information technologies. Computer systems today operate in environments that are constantly changing, influenced by new software, evolving hardware, and diverse data processing methods. Traditional static approaches, which rely on fixed rules or predefined models, often become outdated quickly and fail to provide the necessary adaptability.

Existing approaches to detection in computer systems have been studied extensively, and while they provide valuable insights, they also demonstrate clear limitations. Signature-based methods depend heavily on known patterns and therefore struggle to identify new or unexpected phenomena. Heuristic analysis allows for broader generalization but is frequently associated with high rates of false positives, which reduces its practical usefulness. Behavioral monitoring can capture dynamic changes in system activity, yet it requires significant computational resources and may slow down performance. Machine learning models offer adaptability and the ability to learn from data, but they demand large amounts of training information and careful tuning to avoid errors. Hybrid approaches attempt to combine the strengths of multiple techniques, but they often face difficulties in seamless integration and optimization within existing infrastructures.

Because of these limitations, researchers are increasingly focused on developing frameworks that incorporate automatic updating mechanisms. Such frameworks are designed to be self-adaptive, meaning they can evolve continuously in response to new conditions without requiring manual intervention. Real-time adaptation is a central feature of these systems, enabling them to improve accuracy, reduce false positives, and optimize the use of computational resources.

By integrating intelligent updating mechanisms, information infrastructures can achieve higher levels of stability and efficiency. This not only enhances the overall performance of computer systems but also ensures that they remain relevant and effective in environments where change is constant. The ability to evolve automatically, without relying on outdated static methods, positions these technologies as a cornerstone of future developments in information systems.

The continuous evolution of computational environments demands solutions that are flexible, intelligent, and capable of real-time adaptation. By embracing adaptive frameworks, researchers and developers can create systems that are not only more accurate and efficient but also more resilient and scalable. This marks a decisive step toward the next generation of computer systems, where adaptability and automation are essential for long-term reliability and success.

SURVEY OF TOOLS AND TECHNOLOGIES FOR PSYCHOEMOTIONAL SCREENING AND DETERMINING THE STATUS OF PATIENTS WITH DEPRESSION

Maksym PYTLYAK — Thu, 26 Mar 2026 00:00:00 +0200

The article is devoted to a comprehensive analysis of the current state and prospects for the development of information technologies for psychoemotional screening of patients with depressive disorders. The relevance of the study is due to the global increase in the prevalence of mental disorders, which in the conditions of modern challenges, in particular martial law, is becoming a critical threat to public health and economic stability. The work systematizes scientific sources, which made it possible to identify key trends in the field of digital psychiatry. The main attention is paid to a comparative analysis of existing methods according to eight fundamental criteria that determine the suitability of the technology for real clinical implementation. Among them, the availability of decision-making algorithms, patient routing mechanisms in the primary care setting, the use of validated psychometric tools, integration with electronic medical records, real-time notification systems, adaptation to the individual user norm, ethical transparency, and research on objective behavioral markers. The results of the analysis indicate a significant fragmentation of existing solutions - with a high interest of researchers in the use of biomarkers (voice, eye tracking, electroencephalography and locomotor activity) and artificial intelligence, there is an almost complete absence of systems integrated into the state medical infrastructure. It was found that most of the existing mobile applications and cyber-physical systems operate in isolation from the primary care level, which complicates timely diagnostics and continuity of treatment. The work places special emphasis on the importance of digital phenotyping, which allows objectifying the patient's condition through monitoring motor activity, but it is proven that such data must necessarily be combined with classical clinical protocols. It is substantiated that the lack of integration with electronic medical records and formalized routing algorithms are the main barriers to creating an effective national screening system. Based on the identified "blank spots" in world scientific practice, the author has proven the need to develop a unified information technology that would act as a full-fledged link in the medical process. The analytical basis of the article serves as a theoretical basis for designing a new information technology capable of providing a closed cycle of "monitoring - diagnostics - routing - treatment". The scientific novelty of the work lies in the systematic approach to evaluating screening technologies, which allows us to clearly identify the vectors of further research in the direction of creating information technology adapted to the needs of the modern healthcare system.

INTEGRATING ARDUPILOT SITL FOR DRONE BEHAVIOUR SIMULATION IN GAMIFIED TRAINING

Pavlo PONOMARENKO, Vyacheslav KHARCHENKO — Thu, 26 Mar 2026 00:00:00 +0200

New educational technologies, as well as the challenges of the modern world, encourage the search for ways to improve and develop training methods for specialists to operate in dangerous spaces using unmanned swarm systems. Using technologies such as Unity and ArduPilot, it is possible to create a simulated learning environment that, through gamification mechanics, provides a sense of reward and challenge while learning to interact with a swarm of drones, which requires extensive interdisciplinary expertise. The subject of the study is the interaction module of the Unity game engine and the ArduPilot SITL simulation engine. The object of the study: a software system for training specialists to operate in dangerous spaces using drone swarms. The purpose of the article is to suggest ways to integrate the Unity game engine and the ArduPilot SITL simulation engine in context of a gamification of training specialists to operate in dangerous spaces. To achieve this objective, the paper addresses the following tasks: to formulate the concept of an application that simulates missions with drone swarms in dangerous spaces and provides progression and assessment through gamification mechanics; to justify the choice of a technological stack, in particular ArduPilot, SITL, MAVLink, Unity; to describe the architecture of the Unity ↔ SITL interaction, using TCP/UDP communication, MAVLink message processing, as well as the separation of responsibilities between layers and components. As a result of the research, the architecture was substantiated, and an interaction module between Unity and ArduPilot SITL was implemented using the MAVLink protocol, providing two-way data exchange and scalable integration of multiple simulated devices within a single application. To evaluate the scalability of the system, an experiment was conducted to analyze CPU resource usage and model update time depending on the number of simulated drones. The results showed an almost linear relationship between system load and the number of drones, and no bottleneck was observed in Unity’s main thread.

HYBRID METHOD OF ADAPTIVE CONTROL OF VARIABLE MODE OF UNMANNED AERIAL VEHICLES WITH INTELLIGENT ONLINE COMPENSATION OF DISTURBANCES

Stepan TANASIICHUK — Thu, 26 Mar 2026 00:00:00 +0200

The article resolves the current scientific and technical contradiction between the need to increase the accuracy of navigation control of autonomous unmanned aerial vehicles (UAVs) and the strict resource constraints of on-board computing systems. An intelligent-robust control architecture is proposed, based on the synthesis of the adaptive alternating mode method (ASMC) and the recurrent neuro-fuzzy network RSEFNN. The scientific novelty of the work lies in the improvement of the hybrid approach, which, unlike classical robust methods, uses an intelligent observer for online identification and compensation of nonlinear components of dynamics and external disturbances. This made it possible to significantly reduce the gain coefficients of the discontinuous part of the controller, minimize the "rattling" effect, and increase the energy efficiency of actuators. Mathematical proof of the stability of the closed-loop system using the direct Lyapunov method confirmed the asymptotic convergence of trajectory tracking errors to zero and guaranteed the numerical stability of the neural network training processes. An important practical contribution is the implementation of methods for suppressing high-frequency oscillations by replacing the discontinuous control function with its smooth approximation based on the adaptive boundary layer and hyperbolic tangent. To ensure the determinism of the computational cycle in real time, optimization using the Padé method was applied, which allowed minimizing algorithmic latency and achieving a control frequency of up to 1000 Hz on embedded CPUs without specialized accelerators. The results of the comparative analysis confirmed the high robustness of the developed method under conditions of intense wind loads. In particular, the use of the ASMC+RSEFNN controller allowed to increase the positioning accuracy in steady state by 10.2–12.6 times compared to classical PID controllers. The integrated neuro-fuzzy identifier provided effective compensation for systematic wind shear, which is a critical factor for performing UAV precision guidance tasks in difficult meteorological conditions.