Recommendation systems play a key role in the digital environment, providing personalized recommendations in online stores, streaming services, social networks, and educational platforms. This paper presents a comprehensive review of recommendation system models, including content and collaborative filtering, hybrid approaches, and state-of-the-art algorithms based on deep learning, reinforcement learning, and graph neural networks. The advantages and disadvantages of different methods, their accuracy, performance, scalability and adaptability to new data are analyzed. The main challenges such as the cold-start problem, data sparsity, bias of algorithms, the need for explainability of recommendations and privacy assurance are reviewed. Special attention is paid to the prospects of implementing recommendation systems in educational platforms. The importance of using hybrid and intelligent systems to effectively analyze user data and build recommendations tailored to individual needs is emphasized. The conclusion is drawn about further development of recommendation systems, which will be associated with the integration of the latest artificial intelligence technologies, optimization of computational resources and expansion of their application area in various digital ecosystems. The work can be useful for researchers, developers and practitioners working in the field of artificial intelligence and educational technologies.

This study aims to address the challenges of unstructured learning and information overload in digital education by developing and analyzing a digital learning model integrated with artificial intelligence (AI) to enhance the flexibility and accessibility of education. The methodology is based on system analysis and simulation modeling. A mathematical model was created to describe the interaction between learners, the digital platform, and educational content, incorporating key parameters for flexibility, accessibility, and knowledge acquisition. This model was then modified to assess the impact of AI on platform adaptability, content interactivity, student motivation, and learning rate. The simulation was implemented in Python using the NumPy and Matplotlib libraries. The simulation results demonstrate that while AI tools for adaptability and interactivity increase system flexibility, their overall impact on effectiveness is limited by a plateau effect. The most significant improvements in overall effectiveness are driven by AI components that optimize knowledge acquisition (AIk) and student motivation (AIm). Accessibility remained constant throughout the simulations, indicating that it requires separate optimization strategies beyond the pedagogical scope of this model. The model provides practical recommendations for educational institutions, prioritizing investments in AI solutions for personalization and motivation to achieve the greatest impact on learning outcomes. The proposed framework can serve as a foundation for designing and implementing next-generation intelligent learning systems. This study confirms that the strategic integration of AI can significantly increase the flexibility and effectiveness of digital learning. The findings underscore the importance of focusing AI implementation not just on technological features but on core pedagogical drivers like student motivation and optimized knowledge acquisition.

The rapid transition to digital education has revealed the fundamental limitations of linear, «conveyor belt» platforms that offer identical content to all learners. Such systems are unable to take into account individual differences in initial training levels, cognitive styles, learning speeds, and content format preferences, which reduces the overall effectiveness of learning. This article addresses this problem by developing and describing a formalized mathematical model of an adaptive educational system. The proposed approach is based on a vector representation of both the learner and the learning materials in a single multidimensional feature space. The central element of the model is an interpretable utility function (U), which evaluates and selects the optimal next learning module for a specific student in real time. The function finds a balance between several pedagogically significant criteria: potential knowledge gain (G), cognitive load (E), and format compliance (M). A key feature is the built-in machine learning mechanism (online learning), which iteratively updates the learner's knowledge vector based on their results, allowing the system to continuously adapt. To validate the model, a numerical experiment was conducted, which confirmed that the system is capable of making non-trivial choices, finding the optimal balance between all factors, and correctly tracking the learner's progress. Thus, the proposed model represents a transparent and flexible framework for creating a new generation of intelligent learning systems capable of building truly personalized and effective educational trajectories.

В эпоху цифровизации образования возрастает интерес к использованию многоагентных систем (МАС) как эффективного инструмента создания адаптивных и интеллектуальных обучающих сред. Настоящая работа представляет собой систематический анализ 253 научных публикаций, посвящённых применению МАС в образовательных контекстах за период 2005–2025 гг. Анализ выполнен с использованием методологии PRISMA и охватывает ключевые архитектурные и поведенческие модели агентов, педагогические задачи, а также основные вызовы и тенденции. Результаты показывают преобладание иерархических и гибридных архитектур, широкое использование BDI-моделей агентов, а также доминирование таких задач, как интеллектуальный тьюторинг, персонализация и поддержка совместного обучения. Работа выявляет устойчивый рост интереса к теме, особенно в последние годы, на фоне развития технологий искусственного интеллекта. Обсуждаются основные барьеры на пути массового внедрения МАС, включая высокую стоимость разработки, необходимость построения точной модели обучающегося и недостаток лонгитюдных исследований. В качестве перспективных направлений выделены интеграция с глубоким обучением, объяснимым ИИ, аффективными вычислениями и создание инструментов для преподавателей без навыков программирования. Результаты могут быть полезны для исследователей, разработчиков и практиков в области цифрового образования. Ключевые слова: многоагентные системы, интеллектуальные обучающие системы, персонализация обучения, совместное обучение, систематический анализ, образовательные технологии, искусственный интеллект в образовании.

Gram staining remains a primary diagnostic method in clinical microbiology, but manual interpretation of stained smears is time-consuming, subject to inter-operator variability, and impractical for continuous 24/7 monitoring. Automated detection offers faster turnaround, standardized classification independent of operator experience, and scalability for routine water-quality surveillance. This study evaluates seven YOLO-based architectures for four-class detection of Gram-stained bacterial cells (Gram-positive cocci, Gram-negative cocci, Gram-positive rods, Gram-negative rods) in microscopic images at 1000× magnification. Four baseline nano-scale models (YOLOv10n, YOLO11n, YOLO12n, YOLO26n) were trained and compared with three modified architectures based on YOLO26, including the proposed YOLO26-Refined, which extends the feature pyramid to the P2/4 level, applies a CBAM attention module after the P3 stage, disables shortcut connections in neck blocks, and uses reg_max = 16 for Distribution Focal Loss. All models were trained on a dataset of 5994 annotated images (69/21/10 split) using an NVIDIA RTX 4080 GPU. YOLO26-Refined achieved the best overall performance, with mAP50 = 0.934, mAP50-95 = 0.616, Precision = 0.846, and Recall = 0.888, outperforming the YOLO26n baseline (mAP50 = 0.923) by 0.011. Per-class results show the highest accuracy for Gram-negative rods (mAP50 = 0.965) and the lowest for Gram-positive rods (mAP50 = 0.890). With 2.63 million parameters, 9.9 GFLOPs, and an inference latency of approximately 1.8 ms per image on an RTX 4080 GPU (batch = 1, FP16), the model remains suitable for edge deployment, supporting its application in automated water-quality microbiological monitoring under the IRN BR28712227 research grant.