The unprecedented demand for seamless multimedia services necessitates novel approaches to optimize caching and computing. This paper presents an innovative framework using digital twin (DT) technology to create a dynamic, virtual representation of multimedia service infrastructure, enabling real-time optimization for user-centric applications. By leveraging DT's predictive capabilities and integrating them with advanced caching and computing strategies, the proposed system improves resource allocation, reduces latency, and enhances Quality of Experience (QoE) for users. Our simulation results demonstrate significant performance improvements over traditional approaches, establishing the potential of DT-based systems in next-generation multimedia services.

5G-based mmWave wireless positioning has emerged as a promising solution for autonomous vehicle (AV) positioning in recent years. Previous studies have highlighted the benefits of fusing line-of-sight (LoS) 5G signals with an Inertial Navigation System (INS) for an improved positioning solution. However, the highly dynamic environment of urban areas, where AVs are expected to operate, poses a challenge, as non-line-of-sight (NLoS) communication can deteriorate the 5G mmWave positioning solution and lead to erroneous corrections to the INS. To address this challenge, we exploit 5G single-bounce reflections (SBRs) and LoS signals to improve positioning performance in dense urban environments. In addition, we integrate the proposed 5G-based positioning with a low-cost inertial measurement unit (IMU) and a wheel encoder. Moreover, the integration is realized using an unscented Kalman filter (UKF) as an alternative to the widely utilized extended Kalman filter (EKF) within the 5G-based positioning research community. We performed two test trajectories in the dense urban environment of downtown Toronto, Canada. For each trajectory, quasi-real 5G measurements were generated using a ray-tracing tool incorporating 3D map scans of real-world buildings, allowing for realistic NLoS and multipath scenarios. For the same trajectories, real motion data were collected from two different low-cost IMUs. Our integrated positioning solution was capable of maintaining a level of accuracy below 30 cm for approximately 97% of the time, which is superior to the accuracy level achieved when SBR signals are not considered, which is only around 92% of the time.

Knowledge distillation (KD) allows smaller models to leverage knowledge from larger models, commonly used in image classification, natural language processing, and speech recognition. Traditional logit-based methods prioritize high probability classes, neglecting low-probability predictions, which hinders performance on complex tasks. Although feature-based distillation enhances the use of intermediate representations, its fixed or random feature selection fails to capture the teacher’s rich multilayer representations, leading to incomplete learning. To address these issues, the dual-kernel feature standardization and multilayer attention transfer (DKFSMAT-KD) method is proposed in this study. Due to the focus on high-probability classes, information from low-probability classes is often neglected, resulting in the student model being unable to effectively learn the complete knowledge, a dual-kernel correlation based feature normalization preprocessing strategy (DKCFN) is proposed, achieving nonlinear mapping with first-order and second-order kernel functions, laying the foundation for the student network to fully learn the knowledge of the teacher network. To enable the student network to fully learn the teacher network’s knowledge, a multi-round dynamic filtering strategy (MRDF) is proposed to select the important features using the Pearson correlation coefficient and removes redundant features via mutual information filtering and dynamic thresholds. Moreover, a multilayer attention transfer (MLAT) loss function is proposed to measure the degree of matching between key features of the student and teacher networks. Experiments show that the DKFSMAT-KD method proposed in this paper outperforms the majority of existing knowledge distillation algorithms. Moreover, the experimental results indicate that the DKFSMAT-KD method significantly improves accuracy on CIFAR-100 and ImageNet.  

As Financial Technology (Fintech) continues to evolve rapidly, financial institutions recognise the need to integrate innovative solutions to enhance their risk management strategies. This acknowledgement stems from the potential benefits that Fintech offers in mitigating financial risks. However, despite this growing recognition, empirical research, particularly in emerging markets such as Jordan, remains scarce. To this end, this study aims to evaluate how Fintech applications impact the enhancement of financial risk management practices in Jordan. A questionnaire survey was developed and administered to 11 financial risk experts in Jordan to assess the effects of Fintech adoption on financial risk management. Participants evaluated the potential of Fintech applications, such as digital lending and credit, mobile banking, mobile payments, cryptocurrency, blockchain, and online trading, to improve financial risk management practices. The study findings indicate a perceived positive impact of Fintech adoption on enhancing the efficacy of financial risk management practices in Jordan. This study significantly contributes to the literature by shedding light on the positive impact of Fintech adoption on financial risk management, particularly in emerging markets like Jordan. Moreover, this study has important implications for policymakers, regulators, and banking practitioners, highlighting the importance of creating an enabling environment for Fintech adoption to address financial risks proactively.

A DW is needed to efficiently and continuously prepare data for practical real-time data analysis and forecasting with machine learning (ML) algorithms. This paper discusses extraction, transformation, and loading process (ETL) automation and the critical subsystems integration for algorithmic trading ML modelling and stock market forecasting. Additionally, we discuss a new API designed for direct access to the DW. It opens vast opportunities for performance and data processing time to become critical nonfunctional requirements.

Battery management systems rely on a model of open-circuit voltage (OCV) and state of charge (SOC) relationship of batteries to compute various internal states and parameters crucial for optimizing battery operation. Although existing OCV models offer high accuracy, they have a few limitations, such as non-invertibility, sensitivity to rounding errors, and violation of the monotonic assumptions of the OCV-SOC characteristic. An effective solution is to approximate the OCV-SOC characteristic using piecewise linear segments and store them as a lookup table, simplifying runtime operations into simple lookup operations. This paper presents a dynamic programming optimization to reduce the experimental OCV-SOC data to a minimal number of breakpoints in a table. The optimization follows a simple recursion to yield an optimal breakpoint set without requiring derivatives or tedious calculations while providing complete flexibility in model tuning-adjusting the number of breakpoints to adjust model accuracy. We demonstrate the optimization using low-rate cycling data from six different Lithium-ion cells, showing that about 10 breakpoints can effectively approximate the OCV-SOC curves within a few millivolt error margin, while closely capturing inverse characteristics, which common empirical modeling methods often fail to achieve. Furthermore, we present a feature correlation analysis based on the differential voltage curves of a battery to evaluate the effect of piecewise linear approximation on the linear trend of common aging features.

The application of artificial intelligence (AI) continues to expand across various industries, especially in enhancing user experience and optimizing business processes. Through deep learning and machine learning algorithms, companies are able to analyze user behavior data and provide personalized recommendations, which effectively improve customer satisfaction and loyalty. This data-driven approach enables businesses to stand out in a highly competitive market. This paper explores the key role of machine learning-based personalized recommendation systems in improving user experience and highlights the importance of behavioral data-driven UI design for business success. Analyzing the practices of industry leaders such as Netflix and Amazon reveals how data-driven decisions can be leveraged to deliver tailored content and product recommendations that enhance user engagement and satisfaction. Research shows that successful recommendation systems not only rely on advanced technology applications, but also need to deeply understand user needs to optimize user interface design and promote effective user interaction. As technology continues to advance, personalized recommendation systems will become more intelligent, and companies should actively explore these innovative ways to increase user engagement and brand loyalty to achieve sustainable business growth.

Current and upcoming data-intensive Mission Critical (MC) applications rely on high Quality of Service (QoS) requirements related to connectivity, latency and network reliability. Beyond 5G networks shall accommodate MC services that enable voice, data and video transfer in extreme circumstances, for instance in occurrence of network overloads or infrastructure failures. In this work, we describe the specifications of the architectural framework that enables the roll-out of MC services over 5G networks and beyond, considering recent technological advancements of cloud-native functionalities, network slicing and edge deployments. The network architecture and the deployment process is described in three practical scenarios, including a capacity increase in the service load that necessitates the scaling of the computational resources, the deployment of a dedicated network slice for accommodating the stringent requirement of a MC application and a service migration scenario at the edge to cope with critical failures and QoS degradation. Furthermore, we illustrate the implementation of a Machine Learning (ML) algorithm that is used for overload prediction, validating its ability to predict the capacity increase and notify the components responsible to trigger the appropriate actions, based on a real dataset. To this end, we mathematically define the overload detection problem, as well as generalized prediction tasks in emergency situations and examine the key parameters (proactiveness ability, loockback window, etc.) of the ML model, also comparing its predictions abilities (∼ 93% accuracy in overload detection) against multiple baseline classifiers. Finally, we demonstrate the flexibility of the ML model to achieve reliable predictions in scenarios with diverse requirements.

This study explores the influence of both laser spot size and sizing threshold in Surface Scanning Inspection Systems (SSIS) on cluster identification. Reducing the laser spot size enables the detection and differentiation of smaller point defects while simultaneously limiting the specified search distance, collectively influencing cluster identification. A smaller sizing threshold introduces more point defects, enhancing the likelihood of cluster identification. However, this also leads to a shift between clusters and scratches, as well as cluster merging, ultimately resulting in a decrease in the overall quantity of clusters.

To support the development of wearable medical devices for remote monitoring and treatment of cardiovascular diseases, we tackle the data scarcity problem that hinders the application of machine learning methods. We propose a self-supervised approach applied to cardiological signals, which benefits from existing datasets despite differences between them and inconsistencies within them. We develop a specific implementation: a diffusion autoencoder with a semantic encoder based on linear recurrent units, trained on ECG signals (various leads mixed together) without any annotations. The semantic encoder is evaluated as a feature extractor by measuring classification metrics of a logistic regression on a dataset not included in the self-supervised training. We obtain promising results and propose future directions.

Differentiating between a Relaxed State and a Physical Stress Stage is a useful technology for wearable devices. Currently, an accurate physical stress detection system utilizing the Electrodermal Activity Signal is not available. Using the Electrodermal Activity physiological signal, we develop a novel approach to this problem. Assessing physical changes with Electrodermal Activity/physiological signals is a field that has seen significant progress. The Electrodermal Activity signal tries to observe the changes in the electrical output of the eccrine sweat glands. These changes are performed by the sudomotor nerves, part of the nervous system, and by assessing the changes in these nerves (through Electrodermal Activity), it would be possible to develop a technology like that described above. We have used a dataset from a previous experiment, which has 5-minute stages, to provide the data for our experiment. The data for this experiment was taken with a wearable device. The data were pre-processed using a bandpass butterworth filter and a minmax normalizer. After this, feature extraction and selection were performed. For the model training, multiple models were trained for the stratified and non-stratified dataset. Multiple of them had excellent performance, but the Random Forest Classifier especially stood out for the non-stratified dataset, with a macroaverage cross-validated F-1 score of 0.80. After establishing that the best model was the Random Forest Classifier, Shapley values were used to understand and explain the feature importance of the model.

The rapid expansion of machine-generated text applications has highlighted critical limitations in adaptability and context awareness, particularly when models are exposed to diverse or unfamiliar scenarios. Contextual Hyper-Modulation introduces a groundbreaking framework that dynamically adjusts neural pathways to enhance contextual adaptability, offering significant advancements in accuracy, coherence, and real-time responsiveness. Through the integration of modulation layers that reconfigure internal representations, the framework addresses inherent deficiencies in static pre-trained architectures, achieving superior performance without extensive computational overhead. Empirical evaluations across a range of linguistic tasks demonstrate substantial improvements in context sensitivity and robustness, establishing the framework's efficacy in diverse applications. The findings reveal a transformative approach to augmenting language models with adaptive mechanisms, redefining the boundaries of contextual understanding and model scalability.

Training data sets for unsupervised anomaly detection, expected to be anomaly-free, often contain anomalies (i.e., contamination) which can significantly degrade model performance. Most contamination-robust anomaly detection methods do not generalize across different architectures and rely on contamination ratio information which may not be available. In this paper, we propose a novel self-contained preprocessing anomaly detection approach focusing on data cleaning, i.e., identification and treatment of contamination within a training data set for a more effective unsupervised anomaly detection, all without relying on contamination ratio. Our approach is based on dynamic analysis of the cosine similarity between each data point and its reconstruction during training of an auto-encoder, anticipating low density in most anomaly regions within the dynamic space. This approach enhances data quality through hard cleaning: by removing all identified anomalous candidates from training data. Data cleaning allows generalization across various architectures as the improvements made at the data level benefit any anomaly detection model. This process yields up to 100% of contamination reduction, resulting average improvements of 11.2% in F1 score and 7.3% in accuracy across 8 models and 10 benchmark data sets.

Understanding a product's image within a specific market is essential for evaluating and developing effective internal and competitive strategies to enhance product performance. This article discusses the application of Machine Learning (ML) techniques for understanding and proactively predicting a drug's performance rating across multiple dimensions. The ML models developed in this research can analyze shifting patient sentiments toward drugs over time, enabling pharmaceutical organizations to take proactive measures in response to changes in patient sentiment within a given market. Multiple ML algorithms were explored for this purpose, including Support Vector Machine (SVM), Deep Neural Networks, and Bidirectional Encoder Representations from Transformers (BERT). The performance of each algorithm was evaluated on both training and testing datasets sourced from the UCI Machine Learning Repository. The algorithm demonstrating the best performance has been selected for implementation as a front-end application for real-time drug rating prediction based on review text.

The evolution of cyber threats targeting sensitive data and systems has necessitated the development of increasingly advanced detection techniques to counteract their sophistication and adaptability. Dynamic Encrypted Payload Analysis introduces a groundbreaking methodology, addressing limitations in existing approaches through real-time monitoring and pattern recognition of malicious encryption behaviors. Through leveraging entropy analysis and machine learning models, the framework achieves superior accuracy and scalability while reducing false positives, making it highly effective against adaptive ransomware strategies. Comprehensive experimental evaluations have validated its performance across diverse environments, showcasing significant improvements in detection speed, resource efficiency, and adaptability to emerging threats. The findings demonstrate the transformative potential of the proposed framework, providing a resilient, scalable, and efficient solution for securing digital infrastructures against evolving ransomware attacks.

The growing sophistication of cyber threats has intensified the challenges faced by security systems, particularly in detecting malicious activities that exploit advanced encryption and behavioral evasion tactics. Dynamic Crypto-Behavioral Profiling (DCBP) offers a novel framework that combines realtime cryptographic monitoring with adaptive machine learning to address the limitations of traditional and static approaches. Through the continuous analysis of execution patterns, cryptographic operations, and network behaviors, the proposed system demonstrates exceptional accuracy in identifying both known and novel ransomware variants. The system's ability to adapt dynamically to evolving threats while maintaining high performance across diverse environments underscores its potential to redefine automated cybersecurity defenses. Comprehensive experimental evaluations illustrate its scalability, efficiency, and robustness, establishing a transformative direction in the field of malware detection and prevention.

Virtual Reality (VR) is increasingly used in neuropsychological assessments due to its ability to simulate real-world environments. This study aimed to develop and evaluate the Trail Making Test in VR (TMT-VR) and investigate the effects of different interaction modes and gaming skills on cognitive performance. A total of 71 young female and male adults (aged 18–35) with high and low gaming skills participated in this study. Participants completed the TMT-VR using three interaction modes as follows: eye-tracking, head movement, and controller. Performance metrics included task completion time and accuracy. User experience, usability, and acceptability of TMT-VR were also examined. Results showed that both eye tracking and head movement modes significantly outperformed the controller in terms of task completion time and accuracy. No significant differences were found between eye tracking and head movement modes. Gaming skills did not significantly influence task performance using any interaction mode. The TMT-VR demonstrates high usability, acceptability, and user experience among participants. The findings suggest that VR-based assessments can effectively measure cognitive performance without being influenced by prior gaming skills, indicating potential applicability for diverse populations.  

With the widespread adoption of switched power converters within the power grid as both sources and sinks, the necessity arises for establishing verifiable stability conditions. We assume here that power sources can be modelled as synchronous generators-this encompasses virtual synchronous generatorswhile power sinks are operating as constant power loads. We provide a small-signal stability criterion for a model that uses time-varying phasors, characterized by a lossless transmission network and constant power loads distributed throughout the network. A key concept in this study is the matrix H, representing the partial derivatives of the generator powers with respect to their relative angles, computed at an equilibrium point. Our main result is that H > 0 implies the local asymptotic stability of this equilibrium point. Moreover, the equilibrium point is unstable if H is full-rank and not positive definite. To demonstrate the practical significance of our theorem, we conduct a case study, highlighting its applicability in real-world scenarios.