The article evaluates the efficiency of a liquid cooling system designed for high-power components of modern computing systems, including the central processing unit (CPU), memory chips, and graphics processing unit (GPU). To achieve this, a detailed three-dimensional model of a personal stationary computer was developed using the Solid Edge (Siemens) software environment. The study employed the FloEFD (Flow Simulation) module to simulate hydrodynamics and heat transfer processes, enabling accurate analysis of thermal performance. Key parameters such as ambient temperature (ranging from 263 K (-10 degrees C) to 313 K (40 degrees C)), type of refrigerant (water (H2O), EK-CryoFuel, and Koolance LIQ-702), and the number of operational fans (one to three) were systematically investigated to determine their impact on cooling efficiency. The results provide insights into optimizing liquid cooling systems under varying environmental and operational conditions, offering valuable guidance for improving thermal management in high-performance computing environments. This research highlights the importance of considering multiple factors when designing efficient cooling solutions for advanced technical systems.
The integration of the Internet of Things (IoT) and Artificial Intelligence (AI) is reflected in the Internet of Educational Things (IoEd), in which connected devices and digital platforms are used to support learning environments. Research on Machine Learning (ML) in IoEd is fragmented, and comprehensive evaluation is limited. A systematic and bibliometric review of ML-based IoEd studies from 2015 to 2025 is presented, and a structured screening procedure is applied for study selection. A taxonomy that includes supervised, unsupervised, deep, and reinforcement learning is introduced, and applications in adaptive learning, personalized feedback, engagement monitoring, and resource management are examined. Key challenges are identified, i.e., data scarcity, class imbalance, device heterogeneity, limited interpretability, privacy concerns, and constraints in edge and fog computing environments. Future research directions are identified through the integration of generative AI and Large Language Models (LLMs), Explainable Artificial Intelligence (XAI), Federated Learning (FL), Multimo dal Learning Analytics (MMLA), digital twins, and energy efficient ML methods. The findings indicate that the integration of ML and IoEd supports the development of intelligent educational systems and requires interdisciplinary collaboration to ensure transparent, ethical, and scalable solutions.
Artificial intelligence is applied in smart grids to improve efficiency, reliability, and the integration of conventional and renewable energy sources. A state of the art review of artificial intelligence methods in smart grids is presented. A methodology is used for resource identification and systematic review. A taxonomy is proposed to classify machine learning models by method and application domain. Models are compared based on accuracy and computational efficiency. Key applications such as demand response, energy forecasting, fault detection, and grid optimization are analyzed. Artificial neural networks, decision trees, long short term memory networks, support vector machines, convolutional neural networks, and random forest models are identified as the most used approaches. The best performance is reported for convolutional neural network based and random forest based models. Load forecasting and energy management are identified as the most common application areas.
This paper presents the findings of a numerical study investigating the impact of solar and thermal radiation under clear and cloudy sky conditions on the formation of the wind regime in an isolated urban block in Krasnoyarsk city. A numerical simulation was conducted based on the microscale mathematical model of the urban atmosphere. The area under investigation is a residential zone comprising high-rise buildings with single access points, situated among low-rise buildings, which are streamlined by a constant wind flow in winter and summer season. The findings of the numerical simulation have shown a correlation between the wind regime within the urban block and the daily dynamics of solar radiation.
Interdisciplinary education is positioned as a strategic response to complex technological and societal challenges. However, outcome-driven synthesis of complete study plans remains difficult to scale under heterogeneous competency standards and strict academic regulations. Retrieval-Augmented Generation (RAG) improves factual grounding by retrieving external evidence, while Knowledge-Augmented Generation (KAG) extends RAG by integrating structured knowledge representations for relational reasoning and domain robustness. This paper introduces Curriculum-KAG, a new method and implemented information system that mirrors KAG onto the macro-level task of interdisciplinary curriculum synthesis. A curriculum knowledge base integrates (i) a vector index for semantic retrieval and (ii) a curriculum knowledge graph encoding prerequisites, domains, and regulatory constraints. Hybrid retrieval with graph expansion selects candidate courses and enforces prerequisite closure. Constrained synthesis is formulated as multi-objective optimization with strict verification, while bridge modules are generated only under an evidence constraint when critical learning outcomes remain weakly covered. A prototype is reported on a two-domain case study (IT + Forensics) for the "Cyber Investigator" programme, including an 8-semester plan (240 ECTS) and outcome coverage diagnostics (LO1-LO7, with LO5 at 65%). Measured evaluation against baselines indicates improvements in retrieval and mapping (Recall@20=0.91 +/- 0.02, nDCG@20=0.88 +/- 0.02, Macro-F1=0.85 +/- 0.02) while preserving feasibility (0 violations) and reducing redundancy (0.41 +/- 0.03).
Rheumatoid arthritis is a chronic systemic inflammatory disease that affects synovial joints. The Joint pain, swelling, and reduced mobility lead to decreased quality of life. This disease is caused by autoimmune processes that result in inflammation. We understand that the accurate diagnosis is essential for effective treatment. Our present study evaluates the use of motion laboratory methods for early detection of rheumatoid arthritis and the assessment of a personalized movement therapy. We perform gait analysis under assisted and non assisted conditions to obtain objective biomechanical data. Our results indicate that non assisted walks lead to a longer and more natural gait cycle. Furthermore, the stability of the body center of gravity improves and the joint motion becomes more balanced with reduced asymmetry. These changes indicate better coordination and functional mobility. In addition, our motion laboratory analysis provides objective data to monitor disease progression and treatment outcomes. Our findings support the role of independent movement in rehabilitation and contribute to improved treatment strategies and quality of life.
Speech emotion recognition (SER) is recognized as a growing field with important applications in human computer interaction. In this study, attention is given to SER for the Persian language, which is regarded as a relatively underexplored domain. A lightweight architecture is introduced, in which a frozen DistilHuBERT model is used for feature extraction, and a bidirectional gated recurrent unit (Bi-GRU) block with a dense classification head is employed. Optuna is used for hyperparameter optimization, in which the Bi-GRUs structure, learning rate, dropout rate, L2 regularization, and optimizer choice are tuned so that a balance between simplicity and performance is achieved. The proposed model is trained and evaluated with five fold cross validation, and an acceptable accuracy is obtained. These results show that competitive performance is achieved with a compact and optimized model in comparison with larger architectures.
The paper considers the inverse problem of determining the refractive index of optical media using four-angle Hilbert visualization of phase disturbances. The phase functions for each projection (Radon data) are reconstructed using the Gauss-Newton iterative algorithm. The refractive index is determined using the Gerchberg–Papoulis algorithm based on a discrete analog of the central slice theorem for the four-angle case. The results of numerical modeling are presented.
Effective kicking in karate depends on the stability of lower limb joints and the force of muscles, but little is known about how various training protocols influence these factors. Having an understanding of how specific strengthening programs influence the effectiveness of kicking and joint stability can optimize training for young karate players. Therefore, the present study investigates the effect of back-strengthening exercises executed at varying velocities on roundhouse kick performance and lower limb stability in young karatekas. Eight participants were divided into four groups, each following a 20-week, twice-weekly training program: Control (1), Classic (2), Classic with Constant Muscle Tone (3), and Steady Motion Fitness (SMF) (4). Joint angle changes in the kicking and supporting legs were analyzed before and after training using a video motion capture system, with statistical analysis conducted to assess improvements. Among the three static exercises tested, Group 4 SMF showed the most significant enhancements where p < 0.0001. In the right knee, the mean joint angle increased from 43.32◦ to 48.2 ◦ where p < 0.05, while the maximum joint angles improved from 118.49◦ to 133.42◦ where p < 0.0001. In the left knee, the mean joint angle decreased from 42.58◦ to 32.03◦ where p < 0.05, while the maximum angles decreased from 60.07◦ to 40.02◦ where p < 0.0001. These results indicate that SMF is more effective than traditional methods in enhancing lower limb strength and stability which suggests its potential as an optimal training approach to improve martial arts performance.
Medical image classification often fails for two reasons. Rare but clinically important categories create class imbalance. Similarity between classes also makes some diagnoses hard to separate without a loss that focuses on fine patterns. We introduce Validation Adaptive Focal Loss (VALF), a plug and play objective that augments focal loss with per class weights that are initialized uniformly or from a user provided prior and that are adapted during training based on validation feedback. We keep the weights fixed for the initial part of training, then update them after each epoch using per class validation accuracy. We apply a small multiplicative change and then renormalize the mean weight. The loss is class weight times focal factor times cross entropy. VALF needs no architectural changes, no auxiliary network, and no multi stage training schedule. On LungHist700 at 20x and 40x across five backbones, VALF attains the top macro F1 in 8 of 10 settings and yields consistent gains in accuracy, precision, and recall. The largest macro F1 improvement is about +4.0% over the best baseline at 40x. Improvements are robust across models and magnifications, with only minor shortfalls in two 20x cases. These results indicate that simple validation driven and class aware weighting can balance sensitivity and specificity and can serve as a practical drop in for clinical pipelines.
We present a novel matrix commitment scheme, MCTproofs, which allows a committer to commit to a matrix along with its submatrices and generate proofs for an arbitrary subset of matrix elements. The scheme achieves efficient proof generation and verification by employing tree-based structures in its algorithms. MCTproofs builds upon the foundational principles of Matproofs, offering a significant enhancement over the original scheme. Notably, MCTproofs is aggregatable, maintainable, and up datable, achieving an average performance improvement of 10 x in proof aggregation and 3 x in verification. Additionally, the size of aggregated proofs in MCTproofs is O(1), compared to that of Matproofs, which has a complexity of O(min{b, root n}). We further demonstrate the applicability of MCTproofs in payment-only stateless crypto currencies and compare its performance against Matproofs and Hyperproofs. Experimental results show that MCTproofs outperforms both in aggregation and verification while maintaining comparable efficiency in other operations.
The paper compares the results of numerical modeling with experimental data obtained as a result of a series of experiments to study. The pumping of a finite volume of solution through an elongated rectangular domain of a porous medium is modeled. The MIM approach is used to describe the transport of the impurity. The sorption process is described by a nonlinear MIM model with saturation. Density heterogeneity due to concentration differences is taken into account in the Darcy-Boussinesq approximation. Clogging of the medium due to the deposition of dissolved matter on the pore wall leading to a decrease in porosity, which in turn leads to a decrease in permeability. The dependence of permeability on porosity is given by the Kozeny-Carman equation. The model parameters were found for which the modeling results are in good agreement with the experimental data. The influence of model parameters on breakthrough curves was analyzed. The fields of concentration distribution and current function at different time moments has been received, they demonstrate the occurrence and development of concentration convection in the system.
In this paper, an analysis of turbulence models created using machine learning methods is performed for a wide range of fluid flow parameters at low Prandtl numbers in a channel with complex shape. Three different modern machine learning methods are used for the analysis. A developed turbulent flow in the peripheral channel fragment with three round longitudinal rods is considered for Reynolds number Re = 4270 at low Prandtl numbers Pr = 0.025, 0.05, 0.1, 0.2, 0.4, 0.6, 0.71 encountered in heat-released elements where liquid metals and gas mixtures are used as coolant. The importance of input features of turbulent scalar flux models is analyzed for the entire data set using the permutation feature importance assessment and a method based on Shapley values. The importance of using the local Reynolds number invariant for accurate prediction of the transverse turbulent scalar flux components is demonstrated. It is found that the mutual correlation of some invariants within groups of the selected 15 input features generally preserves the significance level of the most important invariants within each of the groups for the universally interpretable machine learning model for the turbulent scalar flux modeling. Using the permutation method, it was found that the most important basis tensor for estimating the transverse components of the turbulent scalar flux in the tensor basis neural network model is the tensor obtained by the inner product of the deformation tensor with itself S2, and the least important is the tensor SR + RS.
A study is conducted to evaluate the effects of equine-assisted therapy on children with autism, which focuses on motor coordination and social development. Motion capture analysis is applied. Significant improvements in stride length are observed in the riding group. As a result, better balance, posture, and movement are achieved. However, no comparable progress is observed in the non-riding group, although physical exercises are performed. Therefore, the observed effects are attributed to equine-assisted therapy rather than general physical activity. The pedagogical curriculum test of psychological and social skills evidenced significant gains in communication, self-care, occupation, and socialization for the riding group and no measurable changes for the non-riding group. These findings suggest that equine therapy has some remarkable rewards on motor and social development in children with autism, though further research is needed to confirm and expand upon these results.
We present the results of mathematical modeling of the motion of a plate immersed to a certain depth in the fluid and free-floating horizontally in a rectangular convective cell. The convective flow is provided by radiative heating at the bottom and cooling at the upper free boundary. It is shown that the character of the plate motion depends essentially on the optical properties of its surface. Three cases are considered: total transmission, absorption, and reflection of radiation. A good qualitative agreement is obtained between the results of the simulations performed and the experiments described in the literature.
Effective kicking in karate depends on the stability of lower limb joints and the force of muscles, but little is known about how various training protocols influence these factors. Having an understanding of how specific strengthening programs influence the effectiveness of kicking and joint stability can optimize training for young karate players. Therefore, the present study investigates the effect of back-strengthening exercises executed at varying velocities on roundhouse kick performance and lower limb stability in young karatekas. Eight participants were divided into four groups, each following a 20-week, twice-weekly training program: Control (1), Classic (2), Classic with Constant Muscle Tone (3), and Steady Motion Fitness (SMF) (4). Joint angle changes in the kicking and supporting legs were analyzed before and after training using a video motion capture system, with statistical analysis conducted to assess improvements. Among the three static exercises tested, Group These results indicate that SMF is more effective than traditional methods in enhancing lower limb strength and stability which suggests its potential as an optimal training approach to improve martial arts performance.
The way we predict service life and model pavement performance gradually evolves due to the popularity of machine learning (ML). We now have better tools to handle complex, multidimensional and big data through applications of recent advancements in deep learning and large language models (LLMs). Thus, the pavement deterioration can be predicted earlier and with greater accuracy because of models' ability to identify patterns that conventional techniques might miss. In addition to support and continuous monitoring through automated analysis of sensor and visual inputs, ML helps through learning from historical data, traffic patterns, and environmental factors. In order to investigate current trends, we performed a literature analysis after conducting a systematic review in accordance with PRISMA guidelines. Our results state that although ML has gained popularity, the pavement engineering is still in its infancy. In this field, more sophisticated techniques, e.g., generative AI and LLMs have not yet been thoroughly investigated. Even though there are still issues, especially with data quality and model transparency, ML presents great potential in enabling intelligent infrastructure management.
This article includes the design and development of an autonomous robotic system for intelligent navigation in a potato farm. An algorithm for stable and accurate navigation using ultrasonic sensors was designed and implemented in a laboratory-scale model. Hence, four ultrasonic sensors and an Arduino-based control system were a solution for the suggested robot to the farm's physical condition. The suggested robot was first developed with a structure that reflects the physical attributes of the farm, including four ultrasonic sensors and an Arduino-based control system. The robot was further simulated in the MATLAB/Simulink environment and ultimately modeled and evaluated under three different physical conditions. For the given scale of the robot, the maximum acceptable deviation is set at 2 cm, whereas in the unstable scenario, the deviation reaches 5 cm. Experiments were conducted with the primary goal of reducing tracking errors and oscillations while maintaining the robot's alignment with the optimal route. In addition, the mechanical component is constructed with a design that corresponds to the actual prototype in terms of structure and operation. Thus, consistency between the model and the physical system is maintained. Moreover, ultrasonic sensors are used instead of LiDAR sensors under the same arrangement of ridges in potato farms. This decision is logical and economical. Consequently, cost efficiency is improved in future agricultural applications.
Resolution is always a key factor for all types of image processing applications. One common application is text retrieval from images. The assumption that all input images are of good quality may not be met in some cases. Low-quality images can be taken from a low-resolution clip, old printed documents, or downloaded from one of the millions of documents available online. Traditional image preprocessing techniques that use a super-resolution algorithm are mostly designed to enhance natural scenes and face detection. One of the biggest challenges they face is the possibility of losing certain textual details in the noise reduction process. In this study, an optimized framework is specifically designed for text retrieval from low-resolution documents. It can recover missing textual content, which is often eliminated by existing denoising methods. The proposed framework extracts useful features from a small number of outputs for the same low-resolution image. The outputs are generated using four interpolation algorithms. The proposed framework is evaluated on test images scanned at different low resolutions. The evaluation proves the superiority of our proposed framework over other text retrieval methods, which is expressed by a better accuracy compared to the best existing method by 11%.
This review looks at the implementation of artificial intelligence and machine learning into employment research. Based upon an extensive search of literature, the study aims to illustrate the main themes of algorithmic management, platform labor, occupational safety, and aesthetic work where model-building techniques such as neural networks and generative artificial intelligence are applied. An initial search concerning publications between 2014 and 2024 revealed 802 studies. Following a rigorous screening procedure according to the PRISMA guidelines, 25 articles were retained for detailed analysis. This review develops a comprehensive taxonomy of machine learning in employment research and demonstrates its role in modeling the employment quality and improving organizational productivity while affecting occupational safety. Furthermore, this study highlights the importance of explainability, transparency and fairness in machine learning applications for employment research in view of the new legal framework adopted by the European Union in 2024. Additionally, this review attempts to classify machine learning applications in employment research according to the new European regulation on artificial intelligence, introducing a conceptual framework to assess contemporary machine learning-enabled employment research for readiness in view of the new legislation. The results highlight the transformative aspects of artificial intelligence on the nature of work. The research contributes to the understanding of the impact of artificial intelligence and machine learning on employees and organizations, deepening the discourse on its implications in restructuring employment relations in the near future.
