This study presents a novel framework for self-supervised motif discovery in long temporal data, an approach focusing on enhancing explainability and identifying intrinsic differences between motifs. Our research fills a critical gap in motif analysis by introducing a three-pronged methodology: extracting motifs, comparing their signatures, identifying them, and interpreting their differences. The proposed self-supervised technique is designed to ascertain the expected number of motifs in a dataset and to extract them effectively. This approach is reinforced by a robust algorithm that identifies unique motifs and their signatures and a proper distance metric for comparing partially similar motifs. This metric is crucial in determining the similarity between motifs of varying lengths or those containing noise. The application of our framework to ECG data demonstrates its effectiveness in distinguishing between normal and irregular heartbeat patterns without supervision. This capability is a significant advancement in motif analysis, offering a scalable and noise-resistant solution that simplifies the process of segmenting, identifying, and understanding complex temporal patterns. Our results of over 99.9% accuracy underscore our approach's potential in a wide range of applications, particularly in healthcare diagnostics, where accurate and explainable motif analysis is paramount.

Novel cybersecurity threats are constantly emerging and posing significant security challenges to organisations; therefore, it is important for organisations to proactively analyse existing and emerging cybersecurity threats against their systems. Threat modelling methods are very effective in proactively analysing cybersecurity threats and enhancing organisational security policies and defence mechanisms against these cybersecurity threats. Several threat modelling methods have been proposed, and it is important for security experts to select the appropriate threat modelling method for an organisation according to their specific security challenges and cybersecurity threats. This paper will present a comparative analysis of six threat modelling methods: STRIDE, DREAD, VAST, PASTA, OCTAVE, and LINDDUN. It will provide a concise description of all the aforementioned threat modelling methods, and subsequently, a comparative analysis of these six threat modelling methods for highlighting their relative strengths and limitations.

The proliferation of cybersecurity threats is posing substantial security risks to organisations; therefore, it requires robust countermeasures and defence mechanisms for organisational IT systems, applications and data. Threat modelling is a process of identifying, analysing, prioritising and mitigating cybersecurity threats and their associated vulnerabilities in a system or network. Understanding the threat modelling process, as well as its benefits and limitations, whilst selecting an appropriate threat modelling method that may assist cybersecurity experts in their comprehensive security assessments. The assessments are designed to uncover security gaps and potential threats, to develop robust countermeasures against these potential threats and strengthening the security of organisational IT systems, applications and data. This paper will present a comprehensive study concerning threat modelling including the phases involved in threat modelling, types of threat models, benefits and challenges of threat modelling. Therefore, this comprehensive study concerning threat modelling will simplify the essential terminologies of threat modelling to users in a clear and concise manner.

Local electricity markets are increasingly seen as a key factor in democratising the energy system. While there has been a significant amount of research on the technical and economic aspects of these markets, it is still not clear how well they align with the social acceptance of potential participants. The concept of fairness in these markets has been gaining attention, but there has not been a clear discussion or definition of what constitutes fairness in a local electricity market context. Furthermore, some fairness indicators have become popular in the literature on local electricity markets, but have yet to be fully analysed on how well they capture fairness in these markets. This study aims to address these issues by formally defining distributive energy justice in the context of local electricity markets and evaluating how well popular fairness indicators perform based on this definition. The results of this extensive evaluation lead to proposed adjustments to the indicators to make them more suitable for local electricity markets, and a discussion on future research directions for fairness in local electricity markets.

Various applications in maritime scenarios, including offshore operation, navigation, and search and rescue, require reliable high-speed communication. However, wireless maritime communication remains challenging due to limited available radio frequency (RF) bandwidth. Additional challenges are imposed by ducting and water surface reflections that lead to high packet loss rates. This work aims to provide a freespace optical (FSO) communication-based solution to address the limited bandwidth issues. An unlicensed 6 GHz RF link is used as a backup for the FSO link during outage events. The hybrid link deployment was conducted on an island located 19 km offshore at the Red Sea shore. The location was chosen based on the need for construction operations led by Red Sea Global (RSG). In addition, the deployment area is considered a challenging environment for FSO due to its high temperatures, humidity, wind speed, and sand/dust storm events. Our findings indicate that the FSO solution successfully achieves a 20 Gbps full duplex with an availability of 98.71%. The overall connectivity remained at 99.999% with the RF as a backup mode. These results highlight the potential of using FSO for high-speed maritime communication and thus connecting remote and unconnected areas.

We consider target parameter estimation for automotive Phase-Modulated Continuous-Wave (PMCW) radar equipped with one-bit modulo analog-to-digital converters (ADCs) with known folding-counts (Mod-ADC-K). One-bit Mod-ADC-K can save energy and overcome the dynamic range problems of one-bit or low-bit conventional ADCs. However, each one-bit Mod-ADC-K introduces an additional folding making it necessary to devise new target parameter estimation algorithms. The Iterative Adaptive Approach (IAA) is a robust nonparametric spectral estimation algorithm with high resolution. We introduce an extended IAA algorithm, referred to as Mod-IAA for one-bit Mod-ADC-K based target parameter estimation. We incorporate the folding-count data generated by Mod-ADC-K into the likelihood function. This likelihood function is then used to construct a proper cost function. Cyclic and Majorization-Minimization (MM) approaches are employed to iteratively optimize this cost function, to estimate the target parameters. Moreover, we employ a relaxation-based maximum likelihood (ML) approach to iteratively refine the results obtained from Mod-IAA for enhanced off-grid target parameter estimation. We also derive the Cramér-Rao bound (CRB) for the target parameter estimation using one-bit Mod-ADC-K based PMCW radar. Furthermore, to determine the number of targets, we derive the Bayesian Information Criterion (BIC) for the one-bit Mod-ADC-K based PMCW radar system. Finally, we provide numerical examples to demonstrate the advantages of using onebit Mod-ADC-K for target parameter estimation using PMCW radar.

Cardiovascular diseases (CVDs) remain a leading cause of mortality worldwide, necessitating advancements in early disease detection and intervention strategies. Among the various manifestations of CVDs, arrhythmias pose significant challenges due to their potential life-threatening nature and the complexity of timely diagnosis. Traditional diagnostic methods often struggle to identify the subtle symptoms and intermittent occurrence of arrhythmias, highlighting the need for more advanced diagnostic tools. This paper presents a novel approach that uses convolutional neural networks (CNNs) to enhance the classification of arrhythmias from electrocardiogram (ECG) data. The proposed CNN model uses the intricate patterns in the data to accurately identify different types of arrhythmias. The dataset comes from the MIT-BIH Arrhythmia Database containing records for 48 patients ranging from 23 years to 89 years of age from Beth Israel Hospital Arrhythmia Laboratory between 1975 and 1979. Preprocessing steps such as noise filtering, normalization, and data augmentation were employed to improve model robustness. The CNN architecture, optimized for 1D time-series data, extracts hierarchical features that capture both the voltage signals and temporal patterns of arrhythmic beats achieving a model accuracy of 98.66%. Compared to standard diagnostic techniques, our approach demonstrates superior sensitivity and specificity, particularly in detecting critical events like premature ventricular contractions and atrial premature beats.

As the most prevalent subtype of breast cancer, Invasive Ductal Carcinoma (IDC) poses significant challenges in early detection and diagnosis. This study introduces a computer-aided diagnosis system that effectively detects IDC from histopathology images, utilizing advanced machine learning and deep learning techniques. A key feature of this system is the implementation of a sliding window approach for generating high-resolution heatmaps across whole slide images, allowing for precise localization of IDC-positive regions. The model's convolutional neural network architecture is optimized through hyperparameter tuning, and it employs image stitching and heatmap overlay techniques to ensure clear and interpretable visual outputs. These enhancements enable pathologists to better understand the model's predictions. The proposed system achieved a balanced accuracy of 89.06% and an F1-score of 86.68%, surpassing existing models in the literature. These findings demonstrate the potential of the sliding window approach and imaging techniques to significantly enhance diagnostic accuracy, positioning the model as a valuable tool for clinical practice and improving patient outcomes in IDC detection.

This paper probes the validity up to 15 kHz of the analytical formulae published in the CIGRE Technical Brochure 908 related to the calculation of the armour losses in singlearmoured AC cables, which have been tested in the Brochure at 50 Hz only. After briefly reviewing the theoretical background, and carefully explaining the approximations performed by the Brochure, it is shown that for a typical cable the analytical formulae are reasonably accurate in the estimation of losses up to 15 kHz, but the absolute error increases linearly with frequency.

Kurdish handwritten character recognition (KHCR) is an emerging field, which deals with the complications brought forth in recognizing Kurdish handwritten text due to the peculiar characteristics of various scripts and dialects of Kurdish. A review on the status of KHCR was discussed here; important challenges in recognition were listed as data scarcity, intricacies in cursive writing, and variability in the individual's handwriting style. The existing works within this area have been involved in the development of neural network models and datasets; however, there are still a lot of limitations concerning the completeness of available datasets and the precision of the recognition system. The future directions in KHCR research put weight on developing largescale and diverse datasets, including all forms of Kurdish letters, using advanced techniques in deep learning, and realizing realtime recognition systems. This research has come up with a hybrid model combining the traditional approaches of feature extraction with the modern methods of machine learning, including user adaptation in order to improve the accuracy, and dialectal variability of Kurdish handwriting-all these will be very important in the future. By focusing on these areas, KHCR can significantly enhance digital communication and accessibility for Kurdish speakers while facilitating the preservation and digitalization of the Kurdish language. This continuous research in the domain carries the potential to create an inclusive digital environment that will also enable further expansion in use and application related to Kurdish handwriting recognition technologies.

A document by Daniel Chaver. Click on the document to view its contents.

This paper delves into the exploration and application of advanced generative AI models, particularly Generative Adversarial Networks (GANs), in the field of fraud detection and prevention within the FinTech sector. As financial institutions are increasingly leveraging sophisticated technology to address the ever-growing threat of fraudulent activities, the integration of cutting-edge deep learning techniques into these systems is of paramount importance. The focus of this research lies in the development and implementation of deep learning models that are capable of analyzing real-time financial transactions, identifying anomalies, and detecting fraud with unprecedented accuracy. By employing adversarial networks, these models can learn from vast amounts of transaction data, simulating both normal and fraudulent behaviors, thereby enabling the detection of even the most subtle deviations from legitimate patterns.This paper introduces a comprehensive framework for incorporating advanced generative AI models into existing financial systems, offering a robust solution for fraud detection that not only enhances security but also significantly reduces the incidence of false positives. Traditional fraud detection systems often face limitations in balancing accuracy and speed, leading to the misidentification of legitimate transactions as fraudulent, which can negatively impact user experience and incur operational costs. By utilizing the unique capabilities of GANs, which consist of a generator network that simulates fraudulent activities and a discriminator network that distinguishes between legitimate and fraudulent transactions, the proposed framework achieves a more efficient and precise identification of suspicious activities in real time. This adversarial learning process improves the system’s ability to generalize across a wide range of financial behaviors, adapting dynamically to new and evolving fraud tactics.The integration of these generative models into FinTech ecosystems also offers significant advantages in compliance with evolving regulatory standards. Financial institutions are subject to stringent regulatory requirements aimed at mitigating fraud and safeguarding consumer assets. The proposed framework ensures that institutions remain compliant by enhancing the precision and robustness of their fraud detection capabilities, thereby aligning with regulations designed to prevent money laundering, financial crimes, and terrorist financing. Furthermore, the ability of GANs to learn from imbalanced data, where legitimate transactions vastly outnumber fraudulent ones, enhances the detection capabilities even when fraudulent patterns are rare or previously unseen.A key aspect of this research is the real-time deployment of the proposed models, which is critical in financial environments where timely detection of fraudulent activities can prevent substantial losses. The models presented in this paper are designed to operate within milliseconds, ensuring that transactions flagged as suspicious can be addressed immediately without disrupting the flow of legitimate financial activities. This efficiency is achieved by leveraging advanced deep learning architectures that are optimized for high-speed processing and can be integrated seamlessly with existing financial infrastructure, including cloud-based and on-premise systems.Another central challenge addressed by this paper is the trade-off between model complexity and interpretability. While advanced generative models like GANs offer superior performance in detecting fraud, their black-box nature often raises concerns regarding transparency, particularly in sectors as highly regulated as finance. The framework introduced here incorporates mechanisms for enhancing model interpretability, including feature attribution techniques and post-hoc analysis, which provide insight into the decision-making process of the AI models. This transparency is critical for satisfying regulatory scrutiny and ensuring that financial institutions can explain their automated fraud detection processes when required.This research also explores the scalability of generative AI models in fraud detection, particularly as financial systems continue to grow in complexity and volume. With millions of transactions occurring every second globally, fraud detection systems must scale efficiently to handle this massive influx of data. The paper presents a detailed analysis of the scalability of the proposed framework, discussing its adaptability to various transaction volumes, different types of financial services, and diverse user profiles. By deploying GAN-based models that can scale in parallel across distributed systems, financial institutions can ensure robust fraud detection without compromising on speed or accuracy.Moreover, the paper highlights the potential of adversarial training in detecting new types of fraud. Financial fraud is an ever-evolving challenge, with fraudsters continuously developing new tactics to bypass detection systems. Generative AI models, particularly GANs, offer a proactive approach to addressing this issue by simulating possible fraudulent strategies in a controlled environment, which can then be used to train the detection system. This ability to generate synthetic fraudulent data allows the detection models to remain ahead of emerging threats, improving the overall resilience of the financial system.

The tracheotomy procedure is crucial for situations involving intubation, airway blockages, or neck injuries, requiring precise incision placement to minimize risks and ensure effectiveness. Traditional methods involve palpating neck landmarks, but challenges arise in scenarios like teleoperation or critical care. Recent advancements in Augmented Reality (AR) and robotic-assisted surgery (RAS) offer promising solutions to enhance procedural safety and accuracy. In our study, we aim to explore the utilization of AR guidance in assisting tracheostomy incision localization. We employ a handheld ultrasound (US) probe to acquire preoperative anatomical data from a larynx phantom and convert ultrasound data into visual feedback, thus employing a multimodal approach that integrates US information into the visual domain. By marking the region of interest (ROI) on the laryngeal phantom model and importing it into the Hololens 2 device, we achieve visual guidance for precise incision insertion within the ROI. Additionally, we conducted laser localization comparison experiments, comparing procedures performed with and without AR glasses. AR-guided incision localization exhibited impressive performance metrics with a high precision of 0.932 for the cricothyrotomy scene and 0.938 for standard tracheotomy while demonstrating minimal mean central positioning errors of 0.301 mm and 0.236 mm, respectively. The results demonstrate that AR guidance enables surgeons to locate the corresponding laryngeal area touchlessly, efficiently, and accurately, thereby facilitating the progress of robotic-assisted tracheotomy.

The rapid evolution of neural architectures has led to significant advancements in understanding and generating human language; however, existing models often struggle with the limitations imposed by static memory retention mechanisms that cannot adequately manage the complexities of context over extended sequences. The introduction of a dynamic memory approach, which continually adjusts based on token relevance, signifies a substantial breakthrough in enhancing how models process and retain information, ultimately allowing for a more nuanced understanding of context and improved generation of coherent responses. This novel method not only optimizes memory usage but also demonstrates impressive reductions in perplexity and memory consumption, showcasing its capacity to improve both performance and computational efficiency. Experimental findings reveal that models employing this dynamic contextual memory embedding technique outperform traditional architectures in various language tasks, particularly those requiring long-term dependencies, indicating a transformative potential for future research in the field of language models. Moreover, this approach opens new avenues for investigating memory mechanisms in neural networks, paving the way for more adaptive and contextaware systems capable of better handling intricate language structures.

Substations and transmission lines are inspected by unmanned aerial vehicle (UAV) on a large scale, and the captured images are usually delivered to the cloud to be detected by manual or convolutional neural network (CNN) intelligent algorithms for detecting foreign objects or defects, as these cannot be detected locally online by UAVs at this stage. In this paper, we propose a lightweight model based on improved wavelet scattering deep network, which can realize local online recognition of foreign objects on edge devices. The lightweight model contains an improved wavelet scattering network and a 3- layer small deep network. Among them, the wavelet scattering network is constructed considering the extraction of foreign object features and the computational complexity, using biorthogonal (bior) wavelets basis including bior1.1, bior2.2, and bior1.3 wavelets to construct a 3-layer scattering network and compute the modulus and scattering values of the nodes in each layer, replacing the functions of convolutional and pooling layers in the CNN. The small deep network contains 3 fully connected layers and 2 ReLU activation functions to recognize the extracted features. The experimental results show that the accuracy of this lightweight model is higher than 90% for all kinds of foreign object recognition; the accuracy of recognizing 1280*720 foreign object images on the Orange Pi 5B edge device is 94.9%, which is better than other models, and the processing speed is 149.3 frames per second (FPS), which enables local frame-by-frame online detection when the frame rate of the UAV captured video equal to or less than 60 FPS. The algorithm can also efficiently detect high-resolution images up to 1920*1080 with recognition speed of 67 FPS.

This paper presents practical realization of a two-turn meander microstrip line with reduced size. For the first time, two approaches were applied for this purpose: the use of radio absorbing material (RAM) and additional folding of each meander line turn into non-core turns. It was revealed that the sequences of arrival of decomposed pulses in structures with and without RAM differ. In this regard, new conditions for decomposition of USP were formulated. It was revealed that folding of each turn of meander MSL into minor turns leads to additional attenuation of the USP. According to the experimental results, the attenuation of a USP in such a structure was 15.3 dB. The width and length of the fabricated prototype were 46 and 57 mm. By contrast, without using RAM and folding, the dimensions were 10 and 575 mm, and the attenuation of a USP was only 4.08 dB. The analysis of the N-norms of the studied structure showed, that N1, N2, N3, and N5 decreased by 5.75, 13.16, 1.04, and 2.92 times, while N4 increased by 1.54 times. The structure can be used in DC power circuits with voltages and currents up to 530 V and 850 mA.

Machine learning efficiency can often be hindered by static resource allocation and recognition limitations. To tackle these concerns, we propose a dynamic resource allocation strategy coupled with robust recognition mechanisms that adapt to system demands in real-time. Our method continuously optimizes resource utilization by adjusting allocations based on the current workload, ensuring that critical tasks are prioritized without risking system overload. Utilizing a feedback loop, we assess performance metrics to refine resource distribution dynamically. Furthermore, the robust recognition framework incorporates adaptive techniques that significantly enhance model accuracy by effectively managing data input variations. Through extensive experimentation, we validate that our approach surpasses traditional static resource allocation methods, achieving improved operational efficiency and performance across various applications, thereby elevating machine learning resource management to a new standard.

Parametric quantile regression is illustrated for the one parameter new unit Rayleigh distribution called Median Based Unit Rayleigh distribution (MBUR) distribution. The estimation process using reparameterized maximum likelihood function is highlighted with real dataset example. The inference and goodness of fit is also explored.