The rise of creative machines is attributed to generative AI which enabled machines to create new contents. Wherein the introduction of the advanced neural network architecture known as a transformer revolutionized the landscape of generative AI. A transformer transforms one sequence into another sequence, and is primarily used in natural language processing and computer vision tasks. Which determines the relationship between tokens or words in a sequence to understand the context, while processing these tokens or words simultaneously. Transformers were built to resolve various issues of its previous neural networks, such as a Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM); and are now the brains behind majority of generative AI models, for example, most Large Language Models (LLMs) and Large Multimodal Models (LMMs). This paper will explain about the transformer and its architectural components and working. Subsequently, it will illustrate the decoder-only transformer architecture and its components and working, including the reason why this type of transformer architecture is used in most generative AI models such as majority of LLMs and LMMs.

This paper presents a novel methodology for design space exploration and resource optimisation of three-dimensional Networks-on-Chip (3D NoC) architectures using hypergraph modelling and genetic algorithms. The proposed approach combines the mathematical rigour of hypergraph theory with the evolutionary search capabilities of genetic algorithms to efficiently explore the vast design space of 3D NoC configurations. Unlike existing vendor-specific tools, our framework provides a vendor-independent solution for NoC architects and designers. The key contribution is the development of Performance-Cost-Ratio (PCR) functions that enable quantitative evaluation of different topologies and routing algorithms. We validate our framework through two compute-intensive use cases: double SHA256 cryptographic operations and real-time facial recognition. Experimental results demonstrate significant improvements in both applications, with the optimised architectures achieving up to 33% reduction in latency, 40% increase in throughput, and 30% reduction in power consumption compared to baseline implementations. The insights and techniques presented have farreaching implications for developing efficient and scalable NoC solutions as processor designs advance towards kilo-core scales and beyond.

Transcranial Magnetic Stimulation (TMS) efficacy relies on accurate determination of the resting motor threshold (RMT), which is currently determined manually. While previous studies have identified various anatomical and physiological factors influencing RMT, their analyses remained primarily descriptive. This study introduces a machine learning approach to predict RMT levels using anatomical MRI, EEG, and simulated E-field features. We analyzed data from 17 subjects with mild-to-moderate traumatic brain injury, implementing a three-phase analytical approach: independent analysis of feature groups, evaluation of feature importance, and development of a final predictive model. Our results confirmed previously identified RMT descriptors (gray matter percentage, brain-scalp distance, maximum E-field intensity) and revealed novel anatomical factors, including volume percentages of skull, white matter, brain, and skin. EEG analysis showed that beta band activity and delta band connectivity at specific channels were predictive of RMT levels. The optimal model, an SVM with radial basis function kernel, achieved the highest stability in crossvalidation and 100% balanced accuracy score on the test set. This work represents the first machine learning model for RMT prediction, contributing to the development of more efficient, personalized TMS protocols.

Objective: Panoramic basket catheters offer a rapid alternative to conventional local activation time (LAT) mapping for visualizing conduction patterns in arrhythmia patients. However, their limited spatial resolution often makes producing highly interpretable visualizations of the underlying phenomena challenging. We propose a novel interpolation method that addresses uncertainties arising from limited spatial resolution and variable tissue contact associated with basket catheters. By leveraging high sampling rates in the time domain to impose spatiotemporal constraints on the reconstruction, we enable enhanced spatial resolution in visualizing wave propagation. Methods: We constructed overlapping triangles from adjacent electrodes. The local apparent conduction velocity (CV) was estimated from cross-correlations of signals, and the triangles were classified into linear conducting, singularity, and non-conducting types based on CV and goodness of fit. Within each triangle, a linear constraint was imposed on the reconstruction, depending on the triangle type, by projecting signals along the CV. A smoothness constraint was added to ensure consistency at clique boundaries. The hyperparameters were calibrated in an unsupervised manner. Results: The proposed method reduced error by up to 16% compared to traditional interpolation methods in simulations and showed qualitative consistency with LAT maps in clinical cases. Conclusion: The method enhances spatial resolution in panoramic mapping, mitigating aliasing and signal artifacts, especially in regions with complex wavefronts or large interelectrode distances. Significance: The proposed method enables fast and accurate visualizations of panoramic conduction patterns from a single atrial cycle, potentially reducing procedure times compared to sequential mapping.

The increasing integration of inverter-based components into the grid has necessitated the development of enhanced control strategies to maintain grid reliability. Beyond the traditional control provided by inverter-based resources, there is a growing interest in the potential of power electronics interfaced loads to participate in primary control, referred to as Grid Supporting Load (GSL). This paper introduces a novel control approach that equips inverter-based loads in low-voltage DC Microgrid with grid support capabilities, relying solely on local measurements. The proposed method offers a generalized solution for a variety of loads, including constant impedance (Z), constant current (I), and constant power (P) loads, collectively known as ZIP loads. The proposed GSL is experimentally tested and validated in a real-world DC Microgrid setup. Compared to conventional proportional-integral control methods, the proposed GSL method significantly improves voltage regulation, addressing bus voltage drops and enhancing the reliability of DC Microgrid operations.

Transformer networks have attracted significant scholarly attention for their superior ability to model long-range dependencies, making them effective in video object-tracking applications. However, existing Transformer-based trackers still show limitations in handling scale variation, complex background transformations, and occlusion, primarily due to insufficient utilization of the spatiotemporal feature information of the target. Specifically, these trackers struggle with maintaining target semantic integrity, adequately perceiving crucial areas, and incorporating historical information. To address these challenges, this paper introduces several enhancements. First, it replaces the pixel-level attention mechanism with a multi-scale cyclicshifted window attention mechanism, preserving target semantic integrity and improving the model's ability to capture scale variations. Subsequently, the convolution theorem is applied to transform the attention computation of cyclically shifted spatial samples into entry-wise multiplication in the frequency domain for non-cyclically shifted samples, thus enhancing computational efficiency. Additionally, a selective elimination module is proposed to focus on critical areas resembling those in the search frame, effectively minimizing background interference. Finally, a head network assesses the reliability of candidate targets, reintegrating reliable samples as dynamic templates into the tracking network to fully leverage spatiotemporal information from historical frames. Extensive comparative experiments demonstrate that the proposed model outperforms current state-of-the-art trackers across several public datasets.

This paper delves into the transformative impact of Artificial Intelligence (AI) on project management (PM) within the UK construction industry. Our research unfolds in two main dimensions: firstly, examining the current applications of AI in this sector, and secondly, evaluating the challenges that hinder its broader adoption, alongside the potential benefits it offers. Central to our study is the role of AI in enhancing productivity, which is crucial for understanding the factors that influence AI integration in the construction industry.

Generative Adversarial Networks (GANs) are increasingly becoming essential tools for improving the perception of autonomous vehicles in adverse environmental conditions, like low light, fog, rain, and snow. This survey discusses recent trends in applying GANs to tackle these problems and classifies the available work based on environmental conditions and GAN architectures. Studies showcase the impact of different GAN models such as CycleGAN, Super-Resolution GANs, and DeblurGAN in enhancing image quality, visibility, and detection accuracy of objects under challenging conditions. For instance, CycleGAN has been remarkably effective in nighttime image enhancement; DeblurGANs significantly reduce motion blur at capture. Furthermore, multi-modal fusion approaches have been applied very successfully with RGB and infrared (IR) data creating strong perception systems that work well in diverse operational conditions. However, practically deployed driving solutions must process information quickly while remaining computationally efficient. This survey emphasizes how GAN technology has made autonomous vehicle research more adaptable and safer; it also points out potential graduate study directions involving lighter-weight models of GANs or domain-adaptive architectures to make them more applicable. We hope our comprehensive review will help researchers utilize GAN technology to enhance perception for autonomous vehicles.

Deriving the basic reproduction number (R0) in compartmental infectious disease models often involves finding a closed-form formula using the next generation matrix, a process that can be complex and time-consuming. In this paper, I present a novel alternative method for estimating R0 that leverages numerical simulations and machine learning, specifically logistic regression. This approach not only simplifies the calculation of R0 but also integrates parameter sensitivity analysis into a unified framework, enhancing the efficiency of model analysis. The method's simplicity is demonstrated through accessible R code, making it a versatile template for applying machine learning techniques to R0 estimation across various infectious disease models.

The RAS pathway, known as the Ras/Raf/MAPK pathway, is a crucial signal transduction cascade that regulates essential cellular processes, including growth, differentiation, and survival. Central to this pathway is the RAS protein, a small GTPase that acts as a molecular switch, alternating between inactive GDP-bound and active GTP-bound states. Activation of RAS is initiated by extracellular signals, primarily through the binding of ligands to receptor tyrosine kinases (RTKs), which subsequently activate guanine nucleotide exchange factors (GEFs) and the conversion of GDP to GTP. Notably, mutations in the RAS gene, especially the KRAS G12C mutation, are implicated in various cancers, particularly non-small cell lung cancer (NSCLC) and colorectal cancer (CRC), where they lead to uncontrolled cell proliferation and tumorigenesis. This study examines the intricate roles of KRAS, including its domains, isoforms, and mutation profiles, and discusses emerging therapeutic approaches targeting KRAS mutations, such as specific inhibitors like Adagrasib. A comprehensive understanding of the RAS pathways complexities and its oncogenic potential is essential for advancing targeted therapies and enhancing cancer treatment outcomes.

Currently, the 5th generation mobile networks, known as 5G, have been standardized by the 3rd generation partnership project (3GPP) and are already being used commercially. The system is characterized by its flexibility and heterogeneity in mobile cellular communication and networking. Simultaneously, the issue of 5G security has emerged swiftly with the utilization of cloud and non-3GPP networks in the 5G communications network. Extensive research has been carried out on system vulnerabilities and attacks, but there is a relatively new and understudied area known as handover vulnerabilities and their security solutions. Handover refers to the process of transferring connections from one network to another or from a base station to another within the same network. It is important to ensure that any proposed solution for handover vulnerabilities in 5G networks maintains the key characteristics of low latency and network availability. Several security solutions have been suggested, but they involve a trade-off between the additional communication overhead they introduce to the system and the system's requirement for low latency. Our study proposes a new and comprehensive model leveraging the unique physical layer characteristics of the users to enhance the security of the Xn-based handover in 5G networks. We have validated our solution using Scyther tools, which demonstrate that our approach ensures secure communication during the handover process without introducing any additional overhead to the existing system. Furthermore, our solution does not introduce any complexity that could negatively impact the low latency of the system.

The adoption of big data techniques in the capturing, processing and storage of health data was necessitated by the need to improve healthcare quality and patient outcomes.Big Data applications have huge potentials, especially with the availability of huge and growing healthcare data to transform and improve patient monitoring operations, which in turn can help in early detection of diseases, mitigation and the cure of diseases based on historic data. This could also help identify the best treatment for a patient’s specific circumstance, thus improving personalized treatment in patients.Designing effective Big Data applications in healthcare presents some challenges such as real-time data analysis, data privacy, data integration and system interoperability. This paper examines the challenges and opportunities associated with designing Big Data applications for healthcare monitoring, providing insights into the requirements for data architecture, interoperability, security, and real-time analytics.

Spin-orbit torque (SOT) has been extensively studied as a key mechanism in spintronics applications. However, conventional SOT materials limit the spin polarization direction to the in-plane orientation, which is suboptimal for efficient magnetization switching. Recently, multi-directional spins generated by low-symmetry materials have been observed, offering a promising energy-efficient strategy for the field-free switching of magnetic materials with perpendicular magnetic anisotropy. However, the efficiency of this mechanism is highly dependent on the crystallographic texture of the SOT materials, a factor that has not been systematically and numerically studied. Here, we investigate the impact of in-plane crystallographic orientations of SOT materials on the unconventional SOT generated by Dresselhaus-like and out-of-plane spin polarizations. By employing a theoretical orientation distribution function, we calculate the effective unconventional SOT values for SOT materials with tunable in-plane crystallographic texture. This analysis provides insights into the synthesis and optimization of future low-symmetry SOT materials, which can enhance operational efficiency for spintronics applications in magnetoresistive random-access memory and spin logic devices.

The growing demand for green hydrogen highlights the potential of wind energy as a renewable energy source. However, water electrolyzers, crucial for green hydrogen production, suffer from frequent shutdowns that can degrade membranes and catalysts, thus reducing both efficiency and lifespan. The variability of wind power adds further challenges, necessitating advanced control strategies to maintain system stability. Traditional rule-based control (RBC) methods often fail to effectively manage real-time power fluctuations and the uncertainty of future wind conditions, which limits system adaptability. This study introduces a novel RBC method that incorporates a 'target battery level' to balance reliability and productivity. Utilizing a bi-level optimization framework, the design and target battery level are co-optimized, providing a more adaptable and efficient solution. Case studies in six regions reveal that the levelized cost of hydrogen is realistically 6-20% higher than in idealized scenarios that assume perfect wind foresight. Furthermore, the proposed RBC strategy, maintaining a target battery level above 90%, consistently outperforms conventional RBC approaches in terms of reliability and economic viability. This method presents a robust, economically feasible solution to support sustainable green hydrogen production under real-world conditions. Highlights  A bi-level optimization framework is developed to optimize design and operational policies under uncertainty.  A novel rule-based control strategy is introduced, featuring a tunable target battery level, which can be optimized to balance reliability and productivity for a specific case.  Realistic evaluations indicate the Levelized Costs of Hydrogen approximately 6-20% higher than those in ideal scenarios with complete foresight of future wind conditions.

The digital revolution in retail has transformed ecommerce into a sophisticated, data-centric industry, with supervised learning algorithms playing a crucial role in improving business performance. This review explores the effectiveness of various supervised learning methods in tackling the primary challenges encountered by online marketplaces, including customer segmentation, recommendation systems, sales forecasting, and fraud detection. The study assesses the effectiveness of models such as logistic regression, decision trees, and neural networks, considering their predictive precision, commercial influence, and ability to scale. Additionally, it explores limitations like data integrity issues, model overfitting, and difficulties in interpretation, while also highlighting emerging trends such as hybrid modeling approaches and techniques for real-time adaptation. By synthesizing current research, this review offers insights into both the achievements and constraints of supervised learning models in e-commerce, and proposes directions for future studies to enhance their application in a swiftly changing marketplace.

The release of the Citizens Broadband Radio Service (CBRS) spectrum has enabled private network deployments, but signal strength in this band is highly sensitive to weather conditions. While traditional signal propagation models are often constrained by their assumptions and lack adaptability to varying environments, Machine Learning (ML) models offer a flexible alternative for predicting signal strength under complex conditions. In this work, we develop a Neural Network (NN) based predictive model for CBRS signal strength, leveraging a comprehensive set of weather parameters, collected over a 24month period from a live CBRS network in Buffalo, NY. Our model demonstrates a Mean Absolute Error (MAE) of 1.83 dB in predicting hourly signal strength, and we further evaluate its generalization capability by applying a transfer learning approach to another CBRS network in Elmira, NY, achieving a MAE of 2.13 dB. This study highlights the potential of ML methods in modeling weather-induced signal variability and their applicability across different network environments.

Healthcare data warehousing test automation is becoming a success through the help of AI-based technology that results in accuracy, efficiency, and data integrity on the automated test. The one central to personalized patient data that forms the core of traditional data warehousing solutions frequently faces problems of complexity and dissatisfaction. Deep learning with test automation solutions makes data ingestion, processing, and testing conversant through machine learning algorithms [1]. These systems encompass separate acts such as testing for data integration, testing the quality of data as well as performance testing removing the factors of time and human error. In healthcare organizations, this approach delivers the message of enhancing quality, compliance and organizational efficiency in value-based care and precision medicine initiatives. As such, artificial intelligence tools can be used to address the essential issues regarding data warehousing, and this can in turn enhance decision-making, and thus, enhance the health of patient and the overall population.

Evaluating privacy in data-driven software is hindered by its subjective nature and the multitude of privacyenhancing technologies. To address this, we propose a framework for categorizing data based on its privacy implications. This proposal aims to establish a standardized approach for quantifying privacy across various data sources, aligned with DevPrivOps.