This article presents the development and validation of DigSME, a tailored Digital Maturity Model designed to support the digital transformation of small and medium-sized enterprises (SMEs) in Ukraine. The scientific novelty lies in the creation of a customizable, context-sensitive self-assessment tool that addresses the digital inequality faced by SMEs in transitional and post-conflict economies. Unlike existing models developed for mature markets, DigSME integrates strategic, technological, and human resource indicators that reflect the unique socio-economic conditions and institutional constraints of Ukrainian SMEs. The article provides a comprehensive analysis of 13 global digital maturity models and highlights their structural limitations when applied to the Ukrainian context. Based on this analysis and stakeholder consultations from EU-funded and UK-funded research initiatives, a new three-block architecture − Scope, Technologies, and Integration − was designed. The DigSME model includes 13 indicators and 52 sub-indicators, offering a nuanced diagnosis of SMEs’ digital capabilities. An online platform was developed for testing the model, enabling SMEs from Ukraine and the UK to conduct a self-assessment and receive personalized recommendations and strategic roadmaps. The model computes a Digital Maturity Index (DMI) based on weighted technical and organizational scores. Visualization tools, benchmarking capabilities, and downloadable guidance documents enhance its practical applicability. The results of pilot testing confirmed the model’s usability and relevance. Moreover, the platform allows for clustering SMEs by maturity levels and regional contexts, facilitating data-driven policymaking. The study concludes that DigSME serves as both a diagnostic and developmental tool, supporting more equitable digital integration of SMEs into the global economy and informing future digitalization policies.
In this work-in-progress paper, we present Extractomat – our three-lingual Automated Term Extraction (ATE) framework for English, German, and Ukrainian. The framework follows a hybrid iterative approach for ATE in multilingual and cross-domain settings. The approach is tailored to extracting terms from scientific texts in scholarly domains. We report the results of our initial evaluation experiments with Extractomat over our OTRT dataset and an out-of-the-shelf Named Entity Recognition (NER) model over the English subset of the ACTER dataset. The results of these experiments are comparable to the best-performing solutions for multilingual ATE and NER. These findings indicate that the iterative combination of linguistic, statistical, and neural ATE methods, when fully integrated in Extractomat, has the potential to improve the State of the Art (SotA) in the mentioned settings.
Industry 4.0 is characterized by increasing automation of processes and the introduction of intelligent technologies into industry and production. The Internet of Things, cloud computing, artificial intelligence and machine learning, cybersecurity, virtual reality, etc., are all examples of the fourth industrial revolution. However, the presence and implementation of technologies in production do not guarantee the quality of the process. Future specialists must be ready for the challenges of the labor market, new technologies and, in particular, have the necessary set of skills and competencies for their success and demand. The purpose of the study is to analyze the readiness of young people for changes in the labor market of Ukraine and Poland in the context of Industry 4.0. The study was based on a survey of more than 1,000 future specialists from the two countries in various specialties regarding their readiness to work in the era of the fourth industrial revolution, the difficulties they face in finding a job, as well as their vision of the necessary skills and competencies to meet the modern realities of the labor market. Job analysis shows that a modern specialist, in addition to mastering digital tools and subject area orientation, must also have a certain set of soft skills. Our findings show the readiness of young people and the market to cooperate in the context of employment and the difficulties and gaps in the context of Industry 4.0.
This paper presents the development and implementation of a machine learning-based web service designed to predict residential rental prices in the Ukrainian real estate market. The study employs a comparative analysis of three regression algorithms - Multiple Linear Regression, Decision Tree, and Random Forest - to identify the most effective approach for rental price prediction based on apartment characteristics, including area, number of rooms, floor, building height, proximity to metro stations, pet allowance, and distance to city center. Using data collected from DOM.RIA during November-December 2024, the research demonstrates that the Linear Regression model outperforms more complex algorithms, achieving a Mean Absolute Percentage Error of 4.96
The article highlights the problem of using numerous artificial intelligence (AI) services effectively in educators’ and academic staff’s research activities. The Aixploria catalogue has over 5,000 tools, but academic professionals often lack time to view, test and select them for specific research needs. A tool for the personalised selection of AI services using the author’s WPadV4 software has been proposed to overcome this problem. On this basis, it is possible to create educational packages that enable the formation of structured collections of services according to selected criteria. The work implements the selection and classification of AI tools from the Aixploria catalogue following the main stages of scientific and pedagogical research: literature analysis and problem statement, data collection and processing, analysis and interpretation of results, peer review, and dissemination. For each stage, relevant services are to be selected, and their brief descriptions are created and presented in tables automatically converted into web resources – personalised educational packages ready for direct use by educators, academic researchers, and professional development providers. It avoids the routine processing of large amounts of unstructured information and focuses on meaningful aspects of research. The described solution was tested within the framework of international educational and scientific events, particularly in the V4+ EDUPORT project (2022–2023) and at the AISE 2024 conference. To further assess the effectiveness of the developed approach, an expert survey among specialists in the field of ICT in education was conducted.
Nowadays, most of the data obtained using image and video sensors are discrete in nature and represent mappings of continuous processes onto metric spaces with a given dimension. The discretized data can be naturally represented as point arrays on regular integer grids of the corresponding dimension. The obtained arrays of points in a given metric space can be studied from the point of view of their spatial characteristics and relations between the points of the set, which, together with the growth of the amount of data obtained, lead to the need to develop specialized geometric data structures and computer algorithms for their effective processing. The paper presents a unified approach to the processing of discrete point data, which is based on their indexing in a space of a given dimension for solving computational geometry and image processing tasks. A time-efficient bijective geometric hashing method for grid-based data and an approach to organizing interaction of indexed structures have been developed. The proposed techniques have been implemented as computer software in the Python language, and computer experiments have been conducted to assess the effectiveness of the developed approach.
Higher education institutions operating in hybrid warfare environments or located in high-risk regions play a critical role in safeguarding educational, scientific, and governmental stability. Based on the wartime experience of Ukrainian universities, this research proposes a university resilience architecture designed to maintain educational and administrative operations despite infrastructure damage, electricity blackouts, and cyber threats. In this study, an extensive analysis of international standards relevant to enhancing the resilience of universities was conducted, leading to the development of a comprehensive resilience architecture. The proposed model includes geo-replicated data centers, cloud-based backup infrastructures, redundant and satellite communication channels, autonomous energy systems, cyber-resilience measures, and caching server centers to support online education platforms. Future steps of this research may include integration with Digital Twin technologies, cooperation with international and governmental cyber-resilience organizations, and strengthening measures aligned with global security and resilience standards.
This article presents the experience of developing a model for constructing an individual learning trajectory for learning the English language. The model is implemented using Notation3 (N3) and N3 Logic Rules. It takes into account the student's initial knowledge level, university curricula, elements of non-formal education, and external courses from MOOC platforms such as Udemy and Coursera. The system dynamically generates personalized learning paths by aligning educational content with ontological representations and semantic reasoning.
Climatic factors play a primary role in crop‐yield fluctuations, with their effects exhibiting significant nonlinearity. This study examines the influence of air temperature throughout the growing season on wheat yield in the Steppe zone of Ukraine. Weather and climatic conditions in April, May, and June are critical for determining the subsequent wheat harvest. Based on a dataset of 150 climate records and using machine learning techniques, we constructed a statistically significant quadratic regression model to relate yield to temperature indicators. The resulting quadratic model served as the objective function, defined over a constrained domain of admissible temperature values. To locate the maximum of this yield function, optimization methods for multivariate nonlinear functions were employed. The coordinates of the maximum define the optimal temperature trajectory that ensures the highest wheat yield in the Steppe zone. To verify the results obtained, a comparison was made with optimization results produced by other machine learning methods. The proposed methodology enables early yield forecasting with a lead time of approximately three months and can be adapted to forecast the yields of other agricultural crops.
The article focuses on developing a web application that automates the creation of test tasks for engineering students. This development utilizes computer vision and machine learning technologies. The proposed approach involves analyzing images that contain text, formulas, diagrams, and graphs, followed by generating questions using artificial intelligence (AI) tools and a conceptual-thesis model. The web application has been built on the low-code platform FlutterFlow and includes three functional modules designed to process text data, mathematical formulas, and images. This structure enables the adaptation of test tasks to align with educational objectives and the knowledge levels of students. The paper also examines the potential of integrating AI to reduce the routine workload for tutors, enhance the accuracy of educational content analysis, and personalize the learning experience. The findings may help optimize pedagogical activities and improve adaptive educational technologies.
Customer segmentation remains a critical challenge for businesses operating in dynamic market environments. This paper presents a novel hybrid approach integrating reinforcement learning (RL), clustering, and classification methods with extended RFM-D analysis for dynamic customer segmentation. The proposed methodology combines 4 key behavioral metrics: Recency, Frequency, Monetary value, and Diversity of purchases to create comprehensive customer profiles. Our approach employs 3 complementary machine-learning strategies: Q-learning for adaptive discount optimization, k-means clustering for unsupervised customer grouping, and ensemble methods for supervised classification. RL component enables real-time strategy adaptation based on customer response patterns, while Diversity metric captures purchasing breadth beyond traditional RFM parameters. Experimental validation demonstrates the effectiveness of the hybrid approach. RL method achieved optimal profit maximization, while k-means clustering successfully identified four distinct customer segments with targeted discount strategies. XGBoost model showed superior performance. Results indicate that Diversity metric strongly correlates with purchase Frequency, enabling more precise customer targeting. The adaptive discount strategy dynamically adjusts based on price elasticity, resulting in personalized customer engagement that maximizes customer satisfaction and business profitability in evolving market conditions.
This paper presents a novel approach for generating synthetic license plate (LP) images using Latent Diffusion Models (LDMs). The generation of synthetic data is crucial in the domain of LP recognition (LPR) due to strict privacy regulations limiting access to real-world datasets. The study leverages a dataset of 250,000 Ukrainian LP images to train an LDM conditioned on LP text. To that end, multiple Variational Autoencoders (VAEs) are compared in terms of their performance, and the resulting LDM is evaluated based on visual realism and control accuracy. Results demonstrate that the LDM outperforms traditional generative models, such as Generative Adversarial Networks (GANs), in producing realistic and diverse LP images while maintaining high control over the generation process. More precisely, our model yields a Fréchet Inception Distance of 16.15, compared to 31.73 obtained by the baseline. The proposed method shows promise in enhancing the quality of LPR systems while mitigating data scarcity issues.
We present a monadic library in Scala that extends the capabilities of logical programming by integrating machine learning feedback into declarative search. Built upon the dotty-cps-async implementation of the Logic monad, our approach introduces result reordering mechanisms for logical streaming operations. This enhancement allows more flexible and efficient evaluation strategies in logical computations. A key contribution of our work is the extension of the LogicalMonad interface with facilities for directed search—enabling the scoring of results during backtracking. This scoring mechanism serves as a bridge between logic-based search and machine learning models, which typically output evaluation metrics rather than deterministic answers. By combining this scoring with a monadic unification framework, we develop a universal and composable architecture for declarative programming in Scala. Our framework enables developers to guide logical search processes based on external or learned preferences, making it well-suited for applications that require probabilistic reasoning or adaptive feedback. Moreover, the modular design of the system allows for the gradual integration of machine learning components into existing traditional software systems without full rewrites. This approach opens new possibilities for hybrid systems that leverage both symbolic reasoning and statistical learning within a unified and expressive programming paradigm.
This paper explores the development of diagnostic models within intelligent educational systems, with a focus on mathematical disciplines. A comparative analysis of discrete and continuous knowledge tracing approaches is presented, highlighting their strengths in adapting to students’ cognitive profiles and uncovering latent skill dynamics. In parallel, the paper introduces a novel sense-economic model that aims to enhance student motivation by recognizing meaningful contributions to learning. Educational progress is measured not solely through correctness but through deeper indicators such as analytical reasoning, conceptual understanding, creative problem-solving, and peer support. A central component of this approach is the implementation of a unified blockchain-based token (SenseCoin), which operates within a decentralized infrastructure of validated educational values—termed “senses.” These tokens are awarded based on transparent, configurable criteria and can be exchanged for intrinsically valuable educational opportunities, such as access to premium learning resources, project-based modules, or mentorship-driven activities. The model supports a dynamic ontology of sense types, allowing flexible expansion across domains and disciplines. The paper argues that the combination of diagnostic precision and value-sensitive tokenization offers a promising direction for the future of educational systems—one grounded in personal meaning, autonomy, and integrity rather than coercion or formal compliance.
The paper examines approaches to organizing the educational process in Ukrainian higher education institutions during martial law, emphasizing the integration of modern information and communication technologies with a pedagogical focus. Virtual learning environments and learning management systems have become essential for maintaining educational continuity despite unprecedented challenges. The impact of war on students’ general conditions and learning experiences has been analyzed, revealing that these digital platforms offer flexibility and accessibility for both students and faculty. They enable academic activities to continue despite physical displacement and disruptions. However, the rapid transition to online learning has exposed complexities in integrating educational software, particularly the need for students and academic staff to develop the necessary digital skills. Digital literacy gaps and technical support challenges have been identified as significant barriers to the full utilization of these platforms. Addressing these issues requires targeted measures to enhance digital competencies, ensuring participants can effectively navigate virtual learning environments. Based on the experience of Kherson State University, several initiatives have been proposed to improve digital literacy and ensure the effective use of educational platforms. Establishing dedicated digitalization assistant positions, implementing comprehensive training programs, and providing continuous technical support are key measures to facilitate this transition. The integration of artificial intelligence assistants can further enhance accessibility by offering guidance when human support is unavailable. Findings suggest that a combination of technological infrastructure and pedagogical support is crucial for the long-term resilience of higher education in Ukraine. Continued investment in digital skills development will help mitigate future disruptions, ensuring that higher education institutions can maintain stability and adaptability in crisis conditions.
It is well-known that hybrid automata are mathematical models for cyber-physical systems. Unfortunately, the most of problems in behavior analysis for stochastic and probabilistic hybrid automata of general form are algorithmically unsolvable. For this reason, much effort has been focused on the development of probabilistic high-level models intended to solve specific problems in behavior analysis for fairly narrow classes of cyber-physical systems. Among them, models based on Markov processes are widely used. An important, and the easiest to analyze, subclass of these models is formed by Markov chains. In the given paper we illustrate the application of the following three classes of such models intended for solving problems of the behavior analysis of cyber-physical systems: models based on iterating random functions on the state space, models based on finite discrete-time Markov chains, and models based on finite continuous-time Markov chains.
The transition toward Software-Defined Vehicles is reshaping the automotive industry, with vehicle functionalities increasingly determined by software rather than hardware. This paradigm, enabled by Hardware Abstraction Layers, improves software portability and upgradability; however, it also introduces challenges in timing analysis for safety-critical functions that demand an estimation of their reaction times. Existing timing analysis techniques, including those designed for frameworks like ROS 2, often neglect uncertainties in execution times, fail to attribute delays to specific software components, and provide limited insights into the causes of timing variability—hindering system optimisation. This paper presents a novel timing analysis framework for SDVs that overcomes these limitations by leveraging a minimal yet sufficient set of timing information exposed by the HAL. The framework accounts for uncertainty in both hardware and software, models logical timing requirements of components, and estimates reaction time and its variability while explaining their root causes. We demonstrate the applicability and benefits of our approach through a running example, showing how it enables more transparent and informed system design.
The article explores the role of artificial intelligence in the context of three-subject didactics 2.0 – a modern pedagogical concept that redefines the interaction between the learner, the teacher, and the digital educational environment. Based on the three-subject didactics model, the authors analyze the stages of didactics transformation: from a teacher-centered model to an interactive one, and ultimately to the updated three-subject system in format 2.0, where AI becomes an equal participant in the educational process. The article emphasizes the ability of AI to detect logical inconsistencies and analyze dialogues between participants in the educational process to improve the effectiveness of interaction. The paper identifies the levels of interaction between AI and higher education students, analyzes the advantages and challenges of integrating AI into the educational process, and addresses ethical aspects, such as explainability, human involvement, and the risks of reduced social interaction. Special attention is given to the importance of regulatory frameworks and institutional policies, particularly with examples from Ukrainian higher education institutions, such as Kherson State University. In conclusion, it is argued that AI can significantly enhance the effectiveness of learning through adaptive trajectories, real-time feedback, dialogue analytics, and outcome prediction, provided that a human-centered approach is followed in its implementation.
Training deep learning models on large-scale datasets is often constrained by computational cost, time, and energy requirements. Coreset selection offers a promising solution by constructing compact and representative data subsets that preserve the full dataset’s essential characteristics. This paper proposes a novel coreset selection method based on class-wise Neural Principal Component Analysis (NPCA). The method identifies dominant components for each class and selects samples that most strongly express these components. We evaluate the approach on a subset of the ImageNet dataset and on the CIFAR-10 dataset and compare it against established coreset selection methods, Herding, and K-Center. Experimental results demonstrate that the NPCA-based method performs competitively, especially in low-data regimes, while offering improved interpretability through visual analysis of selected samples. Our findings suggest that the proposed method is a viable and efficient strategy for training under resource constraints.
The changeability of the modern world, individual needs and priorities, and societal requirements for higher education graduates are forcing universities to seek the best digital models to move beyond a one-size-fits-all approach to training and provide students with a tailored learning experience. The purpose of the research was to develop and evaluate innovative digital solutions to ensure a transparent, user-friendly, and institutionally efficient system for building personalized learning paths. The research is conducted using the data obtained in the process of testing the developed module of the KSU24 e-platform of Kherson State University (tested with 378 first-year master’s students). The study focuses on the following components of designing a student’s tailored learning experience in the university environment: 1) institutional frameworks and regulatory measures to ensure the dynamic design of tailored learning experience in universities within the changing student mindset and technological progress; 2) digital tools and algorithms, including artificial intelligence and adaptive systems, for personalised learning paths to be automated and adjusted; 3) management initiatives facilitating the integration of decentralised decision-making mechanisms and end-to-end analytics, providing an open and transparent system for selecting elements of the student curriculum; 4) empowering students and improving their engagement by providing them an autonomy in choosing learning goals, academic resources and services; 5) principles of internal peer learning, including methods of horizontal knowledge sharing, teachers’ adaptation to digital technologies and integration of mentoring practices to support the quality implementation of personalised learning paths. The main focus is on the implementation of innovative solutions through the KSU24 platform, including academic performance prediction models, visualisation tools for learning paths, and API integration with learning platforms. This contributes to the creation of a flexible and personalised student curriculum, an analytical framework for making effective decisions, and ensuring their systematic monitoring. The study offers a set of innovative institutional and digital solutions that can be scaled to the unique needs of other higher education institutions for supporting student autonomy in designing their learning paths in the extreme conditions of martial law and limited financial resources of the universities.
