The creation of timetables is a complex combinatorial problem involving numerous hard and soft constraints. Traditional methods often prove ineffective when the number of subjects, teachers and student groups increases. This paper proposes a genetic algorithm with weighting coefficients for multi-criteria optimisation, which creates timetables by balancing preferences for time, days, halls and workload distribution. The approach uses a multi-criteria fitness function in which the weights are extracted through surveys, allowing for a balanced satisfaction of preferences for time, days, halls, workload distribution and minimisation of “free windows”. The experiments conducted show that the algorithm ensures a high percentage of satisfaction of the criteria, no violations of hard constraints, and practical applicability in medium-sized higher education institutions. A block diagram of the algorithm is presented, and the results are discussed in terms of its efficiency, flexibility, and potential for future extensions, including dynamic weight adjustment and integration with machine learning.
Virtual Reality (VR) technologies and avatars are reshaping youth identity development, making the online environment a primary vehicle for self-exploration while also increasing susceptibility to radicalization. This paper examines the impacts of VR on minors, focusing on the psychological and neurobiological mechanisms that facilitate the spread of extremist narratives. Conceptual analyses demonstrate how avatars function as "digital alter egos" and "identity probes," amplifying the Proteus Effect in immersive environments, which shape selfperception and behavior. Leveraging the Sympathy-Empathy (SE) system, radicalization progresses through the amplification of emotional contagion and the neurobiological vulnerability of adolescents. Three radicalization trajectories are identified, including the critical “Parochial Empathy”, where strong solidarity toward the in-group is dissociated from the capacity for empathy toward the out-group, justifying violence. We conclude with the necessity of a multi-stakeholder approach to mitigate the risks amplified by immersion and algorithmic amplification.
No-reference image quality assessment (NR-IQA) is crucial for evaluating perceptual quality without reference images. Existing NR-IQA models for contrast-distorted images primarily rely on luminance-based Natural Scene Statistics (NSS), often neglecting chromatic information. This study introduces two perceptually motivated color features—colorfulness (CIELab) and color naturalness (CIELuv)—into the NR-IQA framework. Experiments on three benchmark databases (TID2013, CID2013, and CSIQ) demonstrate that incorporating these color features consistently improves predictive accuracy, with up to 30% higher PLCC and notable reductions in RMSE. These findings confirm that color cues complement luminance-based features and enhance the reliability of contrast-distortion assessment.
Massive multiple-input multiple-output (M-MIMO) underpins 5G and is central to beyond-5G/6G systems. By equipping base stations with large antenna arrays, M-MIMO enables aggressive spatial multiplexing that improves spectral and energy efficiency while increasing capacity. This survey synthesizes principles and developments \emph{with an emphasis on advances from 2021--2025}, covering antenna/RF advances, hybrid beamforming, signal processing, and deployment at scale. We compare beamforming strategies and \emph{quantify representative SE/EE trade-offs under stated power models}. We also discuss cell-free architectures, RIS integration, and THz/near-field operation, highlighting open challenges and future directions for sustainable, high-performance 6G networks.
The emerging fifth-generation (5G) and sixth-generation (6G) technologies require highly efficient wireless communication systems. For this purpose, Multiple-Input Multiple-Output (MIMO) technology is a foundational component. However, conventional MIMO faces challenges in scalability and interference management, often limiting its use to smaller-scale applications. This work positions Massive MIMO as the solution, which utilizes a very large number of antennas at the base station. This architecture dramatically improves capacity, spectral efficiency, and energy efficiency by serving many users simultaneously through precise spatial multiplexing. A detailed comparison between conventional MIMO and Massive MIMO is presented. The comparative metrics analyzed include spectral efficiency, energy efficiency, scalability, user density, and channel state information (CSI) estimation. The analysis confirms that Massive MIMO offers superior performance compared to traditional MIMO technology. Furthermore, we explore the integration of Massive MIMO with advanced technologies such as artificial intelligence (AI), intelligent reflecting surfaces (IRS), and terahertz (THz) communication. Among these, the synergy between Massive MIMO and AI is identified as a particularly promising approach for enabling robust and efficient future wireless communication.
<p>THEJAS32 SoC is based on the VEGA ET1031 processor, a 32-bit single core, in-order, 3-stage pipeline processor developed by C-DAC under the MDP. This work presents the maiden attempt to deploy Tiny Machine Learning (TinyML) on a hardware platform built around the THEJAS32 SoC. A TinyML model was trained and optimized using the TensorFlow framework using Google Colab and deployed to the hardware using the VEGA Software Development Kit (SDK). The model was compressed by as much as 97% compared to the one developed using TensorFlow. This validates the viability of the TinyML-IoT-THEJAS32 platform for TinyML applications.</p><br />&nbsp;
A gate control system using ANPR (Automatic Number Plate Recognition) is very popular, as many companies offer various gate control options. Typically, the ANPR-based gate control system captures images of license plates with cameras and converts the images into characters using OCR (Optical Character Recognition). Then, the extracted number is checked against a database; if it matches, the gate opens; if not, it stays closed. In this paper, we develop a machine learning-based gate control system using ANPR. First, the system captures images of approaching vehicles with a camera. Next, the YOLOv8 algorithm is used to detect license plates and vehicles. Then, a license plate image is extracted and converted to text with OCR. The vehicle number is compared to the stored number in a database. Finally, the gate opens if the vehicle number matches; otherwise, it remains closed. Our machine learning-based gate control system demonstrates high accuracy and effectiveness in detecting license plates and vehicles. It has been thoroughly tested, with 2,395 detections in total, of which 2,370 were correct and 48 were incorrect, achieving an accuracy of 97.99%.
Medical visualization is a fundamental method that guarantees the accuracy of data analysis for the reliable and prompt diagnosis of potential diseases, resulting in improving the public health. Among the instruments of medical visualization, the methods for delineating edges and identifying possible regions of interest are notable. The proposed study examines color medical images of blood smears exhibiting megaloblastic anemia. To perform the requisite analysis, the input image is enhanced using histogram equalization and segmented into multiple color channels of the RGB format. The wavelet theory of the db1 mother wavelet is employed to identify edges in picture objects. The evaluation of the acquired results relies on visual comparison and widely recognized metrics: niqe, brisque, and entropy. The results indicate that decomposing the original image into data corresponding to distinct color channels yields greater information. For example, the value of the entropy parameter for individual color channels exceeds the value of this parameter for the image as a whole, which enhances the findings, both within the context of each color channel and in comparison, to the grayscale format, which is crucial for employing wavelet theory in image processing. The acquired information can be utilized based on the issue formulation necessary for diagnosing the ailment and making educated decisions.
Lung cancer is still among the most common and fatal cancers in the world, and thus, there is a need for early and proper detection methods to enhance the survival rate of patients. The recent past has seen machine learning (ML) arise as a promising method for predictive modeling in medical diagnosis with the capability to automate at high levels of accuracy. The current survey gives an extensive review of work done in machine learning approaches of lung cancer diagnosis. It thoroughly goes through different datasets, preprocessing techniques, feature selection techniques, and ML algorithms from supervised and unsupervised to deep learning architectures. Convolutional Neural Networks (CNN), Support Vector Machines (SVM), Random Forests, and ensembles are particularly referred to. Evaluation metrics like precision, recall, accuracy, F1-score, and AUC are presented in bold font to indicate model performance. Concerns like interpretability, data imbalance, and generalization are elaborately mentioned. Hybrid systems and emerging trends like explainable AI (XAI) and transfer learning are also briefly touched upon. Lastly, we highlight the limitations in the literature and offer directions for future research towards building robust, scalable, and clinically relevant ML-based diagnostic systems for lung cancer.
Voice assistants have become extremely effective interfaces that connect machine operation and human communication. Though their usefulness is frequently restricted to mobile devices, Internet of Things appliances, or cloud-dependent ecosystems, existing implementations show the promise of speechdriven interaction. Conventional assistants made for computer tasks have limited capabilities, inadequate support for multiple languages, and little personalization. In this work, we introduce an AI voice assistant for computers that combines multimodal interaction, transformer-based natural language processing, and sophisticated speech recognition. Complete systemlevel commands, such as file management, software control, and productivity task automation, can be carried out by the suggested assistant. In contrast to previous research, our model uses a hybrid edge&ndash;cloud architecture to minimize latency while maintaining security through voice biometric user authentication and local data handling.While gesture recognition expands interaction beyond voice alone, context awareness and multilingual processing improve accessibility for a wide range of users. When compared to current IEEE models, performance evaluation shows gains in word error rate, latency, and task success rate. The suggested framework emphasizes how next-generation voice assistants have the potential to revolutionize computer interaction in both personal and professional computing environments by making it more efficient, safe, and natural.
<strong><em>Entrepreneurial education plays a critical role in developing the knowledge, skills, and mindset essential for entrepreneurial success. Beyond education, entrepreneurial self-efficacy&mdash;defined as confidence in performing entrepreneurial tasks&mdash;and entrepreneurial attitude&mdash;characterized by risk-taking, perseverance, and innovation&mdash;are key factors shaping an entrepreneurial mindset. This study investigates the effects of entrepreneurial education, self-efficacy, and attitude on the entrepreneurial mindset of Indonesian higher education students. A quantitative experimental design was employed, with data collected from 345 respondents using a structured survey. Multiple regression analysis indicates that entrepreneurial education and attitude significantly influence the entrepreneurial mindset, with attitude emerging as the most dominant predictor. In contrast, entrepreneurial self-efficacy does not exhibit a significant direct effect, suggesting that confidence alone is insufficient without a supportive attitudinal foundation. These findings emphasize the importance of incorporating experiential learning, attitudinal reinforcement, and real-world engagement into entrepreneurship education. The study offers practical implications for educators and policymakers in designing holistic entrepreneurship programs that integrate cognitive, behavioral, and attitudinal dimensions to foster entrepreneurial success.</em></strong>
Federated learning enables collaborative machine learning on decentralized data, but faces a critical privacy challenge from gradient leakage attacks, which can reconstruct sensitive user data from shared model updates. While differential privacy defenses like static noise injection are common, they often establish a poor privacy-utility trade-off by indiscriminately adding noise, thereby degrading model accuracy. Conventional dynamic methods also fall short, as they typically fail to adapt to the fine-grained, contextual dynamics of local training. To overcome these limitations, we propose FedDynaNoise, a novel privacy-preserving framework that introduces a triple-adaptive noise injection mechanism. The noise level is dynamically and intelligently calibrated based on three key factors such as the training round, the layer-wise gradient sensitivity, and the prediction entropy. This multi-faceted approach ensures that the privacy budget is used efficiently and effectively. We conducted a comprehensive evaluation of FedDynaNoise on four image classification benchmarks against gradient inversion attacks. Our experiments show that FedDynaNoise provides robust privacy protection, achieving a high reconstruction mean square error of approximately 0.65. This is a significant improvement over static noise and conventional dynamic noise baselines around 0.13 and 0.31. Remarkably, this strong defense is achieved with minimal impact on model utility, with FedDynaNoise reaching a test accuracy of 93.9%, only a slight decrease from the 95.1% of a non-private model. Our work demonstrates that FedDynaNoise offers a superior privacy-utility balance, presenting a practical and effective solution for building more secure, trustworthy, and accurate federated learning systems.
