The proliferation of Internet of Things (IoT) devices and the advent of 6G technologies have introduced computationally intensive tasks that often surpass the processing capabilities of user devices. Efficient and secure resource allocation in serverless multi-cloud edge computing environments is essential for supporting these demands and advancing distributed computing. However, existing solutions frequently struggle with the complexity of multi-cloud infrastructures, robust security integration, and effective application of traditional deep reinforcement learning (DRL) techniques under system constraints. To address these challenges, we present SARMTO, a novel framework that integrates an action-constrained DRL model. SARMTO dynamically balances resource allocation, task offloading, security, and performance by utilizing a Markov decision process formulation, an adaptive security mechanism, and sophisticated optimization techniques. Extensive simulations across varying scenarios-including different task loads, data sizes, and MEC capacities-show that SARMTO consistently outperforms five baseline approaches, achieving up to a 40% reduction in system costs and a 41.5% improvement in energy efficiency over state-of-the-art methods. These enhancements highlight SARMTO's potential to revolutionize resource management in intricate distributed computing environments, opening the door to more efficient and secure IoT and edge computing applications.

Intelligent Transportation systems require sensor data to function safely. Cameras present increased resolution and dynamic range, demanding significant datarates which existing vehicular wired and wireless communication networks cannot support. We here propose to use video compression to reduce the required datarate. This aspect must be carefully analysed, as any manipulation of sensor data can have implications on vehicle safety. Most of the widely adopted compression schemes consume demosaiced RGB or YCbCr (Luma, blue difference, red difference) data, which increases the data amount without adding information. RGB data has been traditionally generated for human perception and is potentially not essential for automotive. To leverage existing and highly efficient compression schemes, this paper focuses on studying and proposing novel Bayer adaptation techniques to make Bayer directly consumable by traditional codecs. Two novel and computationally light Colour Space Transform (CST) Bayer adaptation techniques, namely Average CST and De-Interlacing CST, are proposed and compared to state-of-the-art techniques. Adapted Bayer data are deployed into widely usedcompression standards, such as H.264 and JPEG codecs, and then evaluated with IQA (Image Quality Assessment) metrics and deep neural network-based perception, particularly object detection. Through experiments, these novel techniques provide the best perception performance at around 700-1300 kb/fr for H.264 and at over 1000 kb/fr for JPEG, with a negligible reduction of the colour accuracy. This work sets a foundation for further research to optimise Bayer compression for automotive applications. The outcomes of this research will be beneficial to optimise data transmissions and communications for intelligent and cooperative transportation systems.

At any given time, a battery unit can only be in either of two states: It is either charging or discharging. For a microgrid with a rolling time horizon partitioned into a fixed number of specific time intervals, an optimal battery power scheduling is therefore inevitably discrete in nature, and one may have to resort to very complex and time-consuming mixedinteger programming techniques. In this paper, we show that this can in many cases be avoided by introducing small perturbing cost elements in an alternative convex formulation where the solution can be forced to take unique battery charge/discharge states (or configurations) without the need to employ integer variables. A formulation is then obtained which can be solved by using very efficient disciplined convex programming techniques. The new approach is demonstrated by formulating an explicit theory comprising of eight propositions with proofs which are associated with a small generic case consisting of one battery unit, two renewable energy sources, a general-purpose power grid, and one consumer. The general procedure can be straightforwardly extended to higher dimensions. Numerical examples are included to illustrate the theory based on a 24-hour spot price scenario. In particular, it is also demonstrated how a suitably chosen convex battery cost can provide an efficient threshold to avoid unnecessary battery use, and thereby promote a longer lifetime.

Abnormal joint rigidity is a common symptom of many neurological disorders like Parkinson's disease and stroke. Joint rigidity is clinically described as increased resistance during passive joint movement and is quantitatively related to the static passive component of joint impedance, i.e., passive stiffness. Here, we introduce a novel approach to estimate the passive stiffness of the wrist in humans across two coupled Degrees of Freedom (DoFs): Flexo-Extension (FE) and Radio-Ulnar Deviation (RUD). This method employs a subject-specific kinematic model of the human wrist, treated as a universal joint with two skew-oblique axes (FE and RUD), not intersecting and not necessarily perpendicular one to each other. We tested this methodology on ten healthy volunteers using infrared cameras and a hand-held device equipped with a 6-axis load cell for manual wrist perturbations. We used motion and force/torque data to determine angles and torques at the wrist DoFs, and applied multiple linear regression to calculate the 2-by-2 stiffness matrix. Our findings align with existing literature in terms of stiffness behavior, showing stiffness anisotropy with the highest value predominantly along the RUD direction. However, compared to a simplified wrist model with intersecting, orthogonal axes and to prior studies assuming alignment between human joints and robotic ones, our method demonstrates significant differences, particularly in the magnitude of the stiffness ellipse. By providing a more anatomically plausible representation of wrist kinematics through relaxing the constraints of simplified methods, our results show, for the first time, an accurate subject-specific estimation of wrist stiffness.

Digital leisure activities—online games, short videos, and extracurricular reading—are popular among elementary school students. These activities can extend learning beyond the classroom but may also compete for time and energy, potentially impacting academic performance, especially for students with less self-control. This study investigates the relationship between these activities and exam scores, using advanced analytical methods. The findings reveal a significant correlation between learning outcomes and the frequency, duration, and fatigue associated with these leisure activities. Specifically, the frequency of extracurricular reading negatively correlates with science performance (effect coefficient -0.242), while the duration positively correlates (effect coefficient 0.440). An intriguing ”hourglass effect” was observed, where exam scores are heavily influenced by the time and energy students with low self-control spend on recreational activities. These results provide valuable insights into balancing digital leisure with academic responsibilities, offering practical guidance for educators in planning school content and developing educational strategies. By understanding the impact of digital leisure on learning, this research supports the development of more effective learning technologies to enhance academic outcomes in elementary education.

Maize research has grown an interest in exploring the capability to produce a uniform orientation of maize leaves in a field to potentially improve the sunlight capture and increase yield. Previous research tracked leaf orientations to further studies of exploring the potential of a relationship between seed and leaf orientations. Tracking ear orientation may offer a higher throughput insight and the ear tends to follow the leaf orientation. We proposed vision-based machine learning (ML) models to detect ears and estimate the orientation passively at a high rate. To evaluate the proposed method as a proof-of-concept, we had to overcome both seasonal limitations for real-world data generation and manual labor hours for producing a large dataset. The solution was the development of a synthetic maize field and an automated synthetic data generation and labeling pipeline, enabling year-round maize orientation research. The maize field three-dimensional scanning method circumvented millimeter-thin structure challenges to generate low-fidelity leaf structures at scale. The synthetic data pipeline was scaled-up from simpler proof-of-concept objects to the complex maize ear. The synthetic pipeline determined the ear angle could be passively estimated in real-time. The synthetic maize ML models were able to identify the best to worst performing angle estimation model types for the real world. Synthetic ear object detectors were unable to transfer between synthetic and real-world domains. A ResNet18 classifier trained on synthetic ear data tested at 66% accuracy in the real world.

Traffic flow prediction is a critical area of research within Intelligent Transportation Systems (ITS), aiding in traffic planning, control, and more. Real-world traffic flow exhibits inherent non-stationarity, presenting a significant challenge for deep forecasting models. Current non-stationary traffic flow prediction models often decompose the sequence into multiple frequency components, process them in groups, and then recombine to get the output. This approach can lead to a loss of information from the original sequence. In this paper, we propose a novel framework for non-stationary traffic flow prediction called Trend-Variation Separated Learning Framework (TVSLF). TVSLF addresses these limitations by decomposing the raw traffic flow input into trend and variation terms. These terms are processed separately through NET1 and NET2, respectively, before being combined to generate the final output. TVSLF effectively mitigates two key challenges in traffic flow forecasting: underfitting the detailed variations of true values in multi-horizon predictions and lagging in single-horizon predictions. Our method's effectiveness and efficiency are validated against existing methods using four real-world datasets.

In this paper, we propose Zero-Voice, a novel zeroshot text-to-speech (TTS) method, which consists of an Acoustic Feature Encoder, an Acoustic Feature Refiner, and a Waveform Vocoder, specifically optimized for low-resource scenarios with innovations in two key aspects: (i) To enhance Zero-Voice's zero-shot capability, we propose a novel Source Filter Network for unsupervised decoupling of the prosodic components of a reference speaker's voice; (ii) To enhance the quality of synthesized audio, we train the Acoustic Feature Refiner and the Waveform Vocoder concurrently using two diffusion models respectively. This method enables the Acoustic Feature Refiner to generate mel-spectrograms and the Waveform Vocoder to simultaneously produce high-fidelity audios conditioned on these mel-spectrograms. We conduct objective experiments under lowresource settings to compare our model with recent strong zeroshot TTS baseline methods under high-resource settings (e.g., StyleTTS 2 and HierSpeech++). Experimental results demonstrate that Zero-Voice achieves comparable performance to these high-resource methods. Notably, Zero-Voice demonstrates strong generalization and robustness even when trained on a very small number of speakers and small datasets (e.g., 5-8 hours of transcribed data). Moreover, we collect and label 27 hours Te Reo Māori speech data (i.e., an official and endangered language of New Zealand). We train the Zero-Voice model on this dataset, and use it to synthesize Te Reo Māori speech data to enhance speech recognition models for the language. This approach yields state-of-the-art results for the Māori (language code: nz mi) test set of Google Fleurs dataset. Project demo page is at https://github.com/zwan074/zero-voice/.

Voltage source behind an impedance is a term widely used in grid-codes requirements for grid forming (GFM) power electronic converters. However, the exact meaning of this requirement is hard to quantify and verify, and its necessity is not adequately justified. This paper identifies key characteristics that GFM controls should possess and evaluates how GFM controls, with and without a current controller, meet these characteristics. The study utilizes the dq-frame admittance and the frequency-domain power flow Jacobian to analyze an ideal voltage source behind an RL impedance, a synchronous generator, and various GFM and grid-following controls implemented in modular multilevel converters (MMCs). The research identifies several essential voltage source characteristics, which can be mathematically calculated and used to define the frequency ranges where the requirements are met. The paper demonstrates that the GFM structures meet the essential voltage source characteristics in a limited frequency range. The GFM with the current controller is particularly affected due to the nonpassivities of the current controller and virtual impedance. The requirements calculated from the essential voltage source characteristics can serve as a screening method to assess the effectiveness of converter controls, and the insights gained can be utilized to enhance the design thereof.

In this work, we investigate the underlying physical mechanism for electric-field induced resistive switching in Au/MoS 2 /Au based memristive devices by combining reactive Molecular Dynamics (MD) and ab-initio quantum transport calculations. Using MD with Au/Mo/S ReaxFF potential, we observe the formation of realistic conductive filament consisting of gold atoms through monolayer MoS 2 layer when sufficient electric field is applied. We furthermore instigate the rupture of the gold atom filament when a sufficiently large electric field is applied in the opposite direction. To calculate the conductance of the obtained structures and identify the High Resistance (HR) and Low Resistance (LR) states, we employ the ab-initio electron transport calculations by importing the atomic structures from MD calculations. For single-defect MoS 2 memristors, the obtained LRS, HRS current densities are in order of 10 7 A/cm 2 which agrees reasonably well with the reported experiments.

To improve the cognitive autonomy of humanoid robots, this research proposes a multi-scenario reasoning architecture to solve the technical shortcomings of multi-modal understanding in this field. It draws on simulation based experimental design that adopts multi-modal synthesis (visual, auditory, tactile) and builds a simulator "Mahā" to perform the experiment. The findings demonstrate the feasibility of this architecture in multimodal data.

Big Data analytics has become an essential tool for industries to make informed decisions, enhance operational efficiency, and gain deep insights from diversified vast amounts of data. This paper explores the transformative power of Big Data analysis, focusing on four core types: Exploratory, Descriptive, Sentiment, and Predictive analysis. These methods are critical in sectors such as healthcare, retail, finance, and transportation, enabling businesses to identify trends, improve customer satisfaction, and forecast future outcomes. The paper discusses the methodology, required resources, and enabling technologies for effective Big Data utilization, offering a comprehensive framework for implementation. A discussion on challenges and strategies for overcoming them is included, alongside real-world applications that demonstrate Big Data's tangible benefits. Finally, a forward-looking conclusion emphasizes the importance of continuous adaptation and innovation to leverage Big Data's full potential.

Handwritten document recognition (HDR) is one of the most challenging tasks in the field of computer vision, due to the various writing styles and complex layouts inherent in handwritten texts. Traditionally, this problem has been approached as two separate tasks, handwritten text recognition and layout analysis, and struggled to integrate the two processes effectively. This paper introduces HAND (Hierarchical Attention Network for Multi-Scale Document), a novel end-to-end and segmentation-free architecture for simultaneous text recognition and layout analysis tasks. Our model's key components include an advanced convolutional encoder integrating Gated Depth-wise Separable and Octave Convolutions for robust feature extraction, a Multi-Scale Adaptive Processing (MSAP) framework that dynamically adjusts to document complexity and a hierarchical attention decoder with memoryaugmented and sparse attention mechanisms. These components enable our model to scale effectively from single-line to triple-column pages while maintaining computational efficiency. Additionally, HAND adopts curriculum learning across five complexity levels. To improve the recognition accuracy of complex ancient manuscripts, we fine-tune and integrate a Domain-Adaptive Pre-trained mT5 model for postprocessing refinement. Extensive evaluations on the READ 2016 dataset demonstrate the superior performance of HAND, achieving up to 59. 8% reduction in CER for line-level recognition and 31. 2% for page-level recognition compared to stateof-the-art methods. The model also maintains a compact size of 5.60M parameters while establishing new benchmarks in both text recognition and layout analysis. Source code and pre-trained models are available at https://github.com/MHHamdan/HAND.

In-Vehicle Networks (IVNs) have become increasingly complex due to the need to enhance driving experience and provide passengers with a broader range of services. The Controller Area Network (CAN) bus is widely used in modern vehicles, enabling communication between various electronic control units. However, its cybersecurity vulnerabilities require the use of protection mechanisms, such as an Intrusion Detection System (IDS) that can efficiently identify and mitigate threats in CAN networks. In this paper, we perform an extensive and thorough analysis of a CAN dataset, namely CANini. The dataset comes from the augmentation of the CAN-MIRGU dataset, which has been collected from a real vehicle under various driving scenarios. Our analysis offers a comprehensive and extensive characterization of CAN network traffic, examining specifically inter-frame arrival times and patterns in data bytes. We proceeded to test these features by designing a simple anomaly-based IDS that employs a multiclass Decision Tree classifier. Then, a robustness analysis on the IDS has been performed by manipulating the properties of the CAN traffic to find the weaknesses of our selected architecture. Our novel analysis provides critical insights into the behavior and structure of CAN in a typical IVN, which is crucial for the development and validation of a reliable IDS.

Graph neural networks (GNN) have become a powerful framework for analyzing structured data in the form of graphs, with applications spanning diverse fields such as social networks, biology, recommender systems, etc. This survey explores methodology, development, and advances in GNN architectures. We methodically survey the major classes of GNNs, including Convolutional GNNs (ConvGNNs), Spatial-Temporal Graph Neural Networks (STGNNs), Recurrent-based GNNs (RecGNNs), and Graph Autoencoders (GAEs). Every model is discussed in terms of underlying mathematical formulations, design principles, and practical applications. This survey goals to supply a comprehensive conception of GNNs for practitioners and researchers alike, highlighting their versatility and potential for future innovations in graph neural networks. We will furthermore discuss applications of graph neural networks across different fields and epitomize open-source codes, benchmark datasets, and model valuation for graph neural networks. In the end, this survey specifies existing challenges in interpretability, generalization, and scalability and proposes possible future research trends to further promote the performance of GNNs across various graph-based learning missions.

The work considers the N-server distributed computing scenario with K users requesting functions that are linearly-decomposable over an arbitrary basis of L real (potentially non-linear) subfunctions, and the aim is to receive the function outputs with zero error, reduced computing cost (γ; the fraction of subfunctions each server must compute), and reduced communication cost (δ; the fraction of users each server must connect to). For a matrix F ∈ R K×L representing the linearly-decomposable form of the K requested functions, our problem is made equivalent to the open problem of zeroerror sparse matrix factorization that seeks F = DE over a special subset of γ-sparse and δ-sparse E ∈ R N ×L and D ∈ R K×N matrices that respectively define which servers compute each subfunction, and which users connect to each server. We here design an achievable scheme designing E, D by utilizing a fixed-support SVD-based matrix factorization method that first splits F into properly sized and carefully positioned submatrices, and then decomposes these into properly designed submatrices of D and E. For the zero-error case and under basic dimensionality and single-shot assumptions, this work reveals that the optimal number of servers can be upper-bounded as Nopt ≤ min(∆, Γ)⌊K/∆⌋⌊L/Γ⌋ + min(mod(K, ∆), Γ)⌊L/Γ⌋ + min(mod(L, Γ), ∆)⌊K/∆⌋ + min(mod(K, ∆), mod(L, Γ)). In the special case, it reveals that the maximum possible ratio K/N or the zero-error capacity of this system for any feasible single-shot scheme satisfying δ −1 , γ −1 ∈ N is C = max(K/Lδ, γ).

Assessing urban environments using intelligent tools helps decision-makers improve citizen quality of life. Our research introduces a Geospatial Augmented Reality (AR) System for Aesthetic and Environmental Assessment of the University. This framework integrates user presence from geospatial analysis, greenery evaluated through image processing, user sentiments derived from biosensor data using deep learning, and contextual information. These factors are synthesized via Multi-Criteria Decision Making (MCDM) to generate scores. Subsequently, the system utilizes voice recognition to identify user-selected areas, overlaying scores, and relevant contextual data through AR technology. This approach empowers decision-makers to enhance urban spaces effectively.

The interactions between geometrical Doppler, beam pattern, and normalized radar cross section (NRCS) result in leakage of geometrical Doppler into the perceived geophysical Doppler. Starting from high-resolution synthetic aperture radar (SAR) data we model this leakage for synthesized real aperture radar (RAR) observations of ocean motion. The uncertainty introduced by leakage is ∼1 m s −1. Corrections proposed in this work, using only the synthesized lower-resolution RAR NRCS, decrease this to O(0.1 m s −1). Further reduction of instantaneous leakage may be achieved through temporal averaging, since the leakage-inducing NRCS gradients appear mostly atmosphere induced and decorrelate rapidly. The azimuth resolution and number of independent samples determine a system's sensitivity to leakage. C-band systems are inherently prone to suffer from greater leakage and worse corrections than their Ku-and Ka-band counterparts. With DopSCA, the propagated effect of leakage is only secondary compared to the pulse-pair uncertainty. In similar systems where pulse-pair uncertainty is suppressed, leakage will dominate instead.This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.