This research investigates the application of machine learning for diagnosing COVID-19 from chest X-rays. We analyze various popular architectures, including efficient neural networks (EfficientNet), multiscale vision transformers (MViT), efficient vision transformers (EfficientViT), and vision transformers (ViT), on a dataset categorized into COVID, lung opacity, normal, and viral pneumonia. While multiscale models demonstrate a tendency to overfit, our proposed fine-tuning ViT model achieves significant accuracy, reaching 95.79% in four-class classification, 99.57% in a clinically relevant three-class grouping, and similarly high performance in binary classification. Validation through quantitative metrics and visualization solidifies the model's effectiveness. Comparative analysis showcases the superiority of our approach. Overall, these findings showcase the potential of ViT for accurate COVID-19 diagnosis, contributing to the advancement of medical image analysis.

The convergence of Web3, Metaverse, and Digital Twins technologies is bringing a transformative revolution to Cyber-Physical-Social systems. Web3, which is driven by blockchain and decentralization, allows users to have control over their data and digital assets. Meanwhile, the Metaverse is creating a virtual space where people can interact, work, and live together. Digital twins offer a real-time digital representation of physical objects or spaces. When these three concepts intersect, they create a dynamic and interconnected digital ecosystem where the physical and virtual worlds blend seamlessly. This paper focuses on discussing the convergence of Metaverse, Web3 technologies, and Digital Twin. We will focus on the architecture of a Web3-enabled Metaverse, which leverages the decentralized nature of Web3 to offer a distributed Metaverse. The digital twin technology will realize a Cyber-Physical-Social data binding that enables the seamless mapping of physical and social activities into cyberspace. Finally, we will discuss the challenges of a Web3-enabled Metaverse, such as security, privacy, interoperability, and ethical concerns.

In this work, we attempt to find a suitable oxide resistive layer among four popular metal oxides, including HfOX, NiOX, TaOX, and TiOX for GE-based RRAM devices by investigating the device performance using a fully experimentally calibrated numerical simulation model. The results reveal that among four metal oxides, HfOX resistive layer with GE provides lower reset voltage (0.12 V), lower sneak current (81.07 nA) with better power efficiency (45.92%) for 2 8 bit capacity. On the other hand, NiOX-based RRAM shows a higher switching current (14.45 µA), more thermal stability, better uniform and distinguishable multilevel states, and higher readout margin for crossbar array size. Our study offers a comprehensive examination of the performance characteristics of distinct resistive layer materials that can provide an important guide to experimental efforts and trigger more efforts.

Approximate computing has been extensively employed in arithmetic circuits such as a ripple carry adder (RCA); in an approximate adder, errors are introduced by design. VLSI circuits are also prone to errors caused by external and physical phenomena (such as cosmic rays or a stuck-at); hereafter, these errors are referred to as functional. This paper investigates the combined effects of single functional error (SFE) in an approximate cell and an RCA. The exact and approximate cell designs are considered using a state transition diagram-based analysis to identify relationships between the types of functional error in a cell against the expected behavior for all possible cases. A probabilistic analysis for an exact RCA is proposed; it shows excellent agreement with the simulation results for different metrics such as the error rate (ER). Next, an error analysis is performed on the RCA by considering the number of approximate cells as well as the location of the erroneous cell; the simulation results and modeling analysis of the exact RCA show and prove that the ER and MED (Mean Error Distance) of odd-numbered cases is higher than for evennumbered cases. Moreover, the MED for an approximate RCA in the presence of an SFE is higher than for the exact RCA. Simulation and analytical results show that an approximate RCA in the presence of an SFE can have a serious degradation in accuracy performance if the approximate cell type is not properly selected. In addition, the binary tree-based mathematical analysis and pseudocode are provided to support the exhaustive simulation results for the ER of the RCA using different approximate cells.

When designing a power or CPU constrained device where a four-axis robotic arm is required and access to the Robot Operating System (ROS) is not an option, finding an efficient state space controller for a four-axis arm can be an obstacle. In this paper, I explore a method to optimize the computing power required for a computer algebra system (CAS) to compute linear quadratic regulator (LQR) matrices by precomputing the gain matrix for different states. Example C++ code is provided on Github, along with ideas for further exploration.

As businesses move toward cloud computing environments, security and privacy concerns become increasingly important. This research report systematically investigates various challenges and vulnerabilities in cloud computing, focusing on security and privacy issues. This study comprehensively examines potential threats, from data breaches to unauthorized access, and assesses the impact of these challenges on user trust and data integrity in cloud infrastructure. Additionally, effective strategies and solutions to reduce security risks and protect user privacy in cloud computing are proposed and discussed. This paper contributes to the ongoing debate on cloud security and privacy and provides both practitioners and researchers with a valuable resource for navigating the evolving landscape of cloud-based services.

The field of blockchain interoperability plays a pivotal role in blockchain adoption. Despite these advances, a notorious problem persists: the high number and success rate of attacks on blockchain bridges. We propose Harmonia, a framework for building robust, secure, efficient, and decentralized cross-chain applications. A main component of Harmonia is DendrETH, a decentralized and efficient zero-knowledge proof-based light client. DendrETH mitigates security problems by lowering the attack surface by relying on the properties of zero-knowledge proofs. The DendrETH instance of this paper is an improvement of Ethereum’s light client sync protocol that fixes critical security flaws. This light client protocol is implemented as a smart contract, allowing blockchains to read the state of the source blockchain in a trust-minimized way. Harmonia and DendrETH support several cross-chain use cases, such as secure cross-blockchain bridges (asset transfers) and smart contract migrations (data transfers), without a trusted operator. We implemented Harmonia in 9K lines of code. Our implementation is compatible with the Ethereum Virtual Machine (EVM) based chains and some non-EVM chains. Our experimental evaluation shows that Harmonia can generate light client updates with reasonable latency, costs (a dozen to a few thousand US dollars per year), and minimal storage requirements (around 4.5 MB per year). We also carried out experiments to evaluate the security of DendrETH. We provide an open-source implementation and reproducible environment for researchers and practitioners to replicate our results.

Federated Learning (FL) has established itself as a widely accepted distributed paradigm. Without sharing data, it may seem like a privacy-preserving paradigm, but recent studies have revealed vulnerabilities in weight sharing which results in information disclosure. Hence, privacy-preserving approaches must be incorporated during aggregation to avoid disclosures. In the literature of FL, not much focus has been given on generating generalized models which can be generated by multiple sets of datasets thus avoiding identity disclosure. Integrally private models are the models which recur frequently from different datasets. So, in this paper we focus on generating the integrally private global models proposing k-Anonymous Integrally Private Federated Averaging (k-IPfedAvg), a novel aggregation algorithm which clusters similar user weights to compute a global model which can be generated by multiple sets of users. Convergence analysis of k-IPfedAvg reveals a rate of O(1 T) over training epochs. Furthermore, the experimental analysis shows that k-IPfedAvg maintains a consistent level of utility across various privacy parameters in contrast to existing noise based privacypreserving mechanisms. We have compared k-IPfedAvg with classical fedAvg and its differentially private counterpart. Our results shows that k-IPfedAvg has comparable accuracy score with baseline fedAvg and outperforms DP-fedAvg on iid and non-iid distributions of MNIST, FashionMNIST and CIFAR10 datasets.

The vision of biocomputing is to develop computing paradigms using biological systems, ranging from micron-level components to collections of cells, such as organoids. This paradigm shift exploits hidden natural computing properties, developing miniaturized wet computing devices deployable in harsh environments, and exploring designing novel energyefficient systems. Parallelly, we witness the emergence of AI hardware including neuromorphic processors aiming to improve computational capacity. This study brings together the concept of bio-computing and neuromorphic systems by focusing on the Bacterial gene regulatory networks and their transformation into Gene Regulatory Neural Networks (GRNNs) that can be used for biocomputing. We explore the intrinsic properties of gene regulations, map this to a gene-perceptron function, and propose an application-specific sub-GRNN search algorithm that maps the network structure to match a problem. Focusing on the model organism Escherichia coli (E. coli), the base-GRNN is initially extracted and validated for accuracy. Subsequently, a comprehensive feasibility analysis of the derived GRNN confirms its computational prowess in classification and regression tasks. Furthermore, we discuss the possibility of performing a wellknown digit classification task as a use case. Our analysis and simulation experiments show promising results in offloading computation to GRNN in bacterial cells, advancing wet-neuromorphic computing using natural cells.

This paper presents the design of an open-API-based explainable AI (XAI) service to provide feature-contribution explanations for cloud AI services. Cloud AI services have broad usage in developing domain-specific applications with learning precision metrics. However, the underlying AI models remain opaque on how the model produces the prediction. We argue that XAI operations are accessible as open APIs to enable the consolidation of the XAI operations into the cloud AI services assessment. We propose a design using a microservice architecture that offers feature contribution explanations for cloud AI services without unfolding the network structure of the cloud AI services. We can also utilize this architecture to evaluate the performance and XAI consistency metrics showing cloud AI services' trustworthiness. We collect provenance data from XAI operations to enable traceability within the XAI service. Furthermore, we present the discovery scenarios for the experimental tests regarding performance and XAI consistency metrics for the leading cloud AI services for computer vision. The results confirm the open-API-based architecture cloud-agnostic. Additionally, data augmentation has a marked improvement in XAI consistency metrics for the cloud AI services.

This work highlights issues that were not deeply investigated in previous studies on clustering solutions, which have essential impacts on performance in long-term real-world applications that are challenging to detect instantly. Thus, we addressed these issues by proposing two novel techniques: first, we expand the idea of clustering based on centroids to multiple sub-centroids that assist assignment functions in finding the optimal solution. In contrast to recent studies, we extended the concept of gravitational force toward clustering solutions. Finally, the introduced gap generation concept has been associated with these techniques to support a superior clustering solution. Our model is termed semi-supervised gravity clustering (SSGC). To demonstrate the strength of SSGC, we consider multiple performance measurements besides the traditional ones to validate the clustering models in various scenarios. The experimental results show that SSGC outperforms baseline models and successfully obtains the best performance of 30 different domain datasets. Finally, our methodology code is already released.

Many businesses seeking new capabilities that blockchains may offer are deterred from fully embracing the technology due to fears of the classic "vendor lock-in" and platform-capture into one specific blockchain. From an asset-centric perspective, most business applications seek certain desirable functional guarantees with regard to the state of the tokenized asset on the blockchain. These new capabilities must be accessible through standardized service interfaces. The emerging tokenized asset networks based on decentralized ledger technology must integrate seamlessly into existing financial IT systems through similar standard interfaces. As such, if blockchains are to be a foundational technology in the future Web3 Internet of Value, then several classes and types of standardized APIs must be specified, published, and widely deployed by the nascent tokenized asset industry. These standard APIs must provide business applications with a single uniform interface to the many and varied blockchains today, thereby reducing business IT costs and preventing platform-capture.

The Internet of Things (IoT) has revolutionized the way we interact with technology and devices. Several IoT systems are being deployed across diverse domains, including but not limited to health, transportation, agriculture, and manufacturing. They fulfill critical tasks and, thus, must function correctly and securely and meet the users' expectations. However, testing IoT systems poses many challenges, primarily due to their distributed nature, dynamism, and heterogeneity as well as the multiple layers of which they are composed, i.e., device, edge, cloud, and application layers. The absence of testing guidance can hinder the quality of IoT systems. Testing guidelines, including taxonomy, are vital for proper IoT systems testing. In the context of software testing, taxonomy organizes and categorizes testing aspects, helping testers to understand what, how, and when to test. However, no IoT systems testing taxonomy exists, and traditional software testing taxonomy may not sufficiently meet IoT systems testing requirements. To address this, we introduce an IoT-specific testing taxonomy, informed by a review of 83 primary studies and validated through surveys with 16 IoT industry practitioners. We assess its effectiveness by conducting an empirical evaluation with 12 testers. The results show that our taxonomy can help IoT testers become more efficient by fostering their understanding of various aspects of testing IoT systems. This taxonomy can help the testers to increase test coverage, enhance the efficiency and effectiveness of testing efforts, and ensure thorough testing of important system aspects, thus ensuring functional correctness, improving the security of IoT systems, and better meeting users' expectations.

Rapid advancements have been made in artificial intelligence applications recently, and generative models have prominently emerged as effective tools for domain transfer, image enhancement, and simulation. However, when dealing with large-scale gigapixel images, the use of traditional patch-based image aggregation methods introduces checkerboard or blocking artifacts, which compromises image quality and fidelity. Here we propose a parametric kernel that is specifically designed to target the underlying grid structure to mitigate these artifacts. The proposed parametric kernels are validated using three medical imaging modalities for three different generative model tasks, demonstrating improved visual fidelity and quantitative quality evaluation of the generated patch-aggregated images. The proposed method is versatile and compatible with various generative models, offering a robust framework for artifact reduction that can be seamlessly adjusted by modifying kernel parameters, and they can be directly applied and extended to other imaging modalities that employ large-scale images, such as astronomy and satellite imaging. The findings of this study have significant implications for medical imaging applications: by mitigating aggregation artifacts, our approach enhances the overall quality of medical images synthesized with generative models, which is crucial for accurate clinical assessment and subsequent image analysis. Furthermore, the proposed kernels provide a general formulation that can be extended to unpaired tasks, semantic segmentation, classification networks, and other large field-of-view imaging applications.

In this paper, we present a low-cost 5G-SA system based on edge-computing hardware with open-source components and off-the-shelf hardware. Our system is suitable for nomadic and ad-hoc 5G-SA use cases that require low power consumption and small space. We show a complete development covering how to prepare the operating system for the installation of the 5G next generation node B (gNB) and how to install and deploy the 5G-SA software components. We evaluate the performance and power consumption of our system with measurements. We show that our system achieves acceptable data rates, is very space and energy efficient with an average power consumption of 24.81 W. We provide detailed and maintained documentation on the deployment of our system and discuss possible extensions for future work.

In a world where traditional authentication systems are constrained, the introduction of voice-based authentication provides a viable option. This cutting-edge biometric security method uses unique vocal traits including pitch, tone, and speech patterns to create a unique voiceprint. Its sophisticated and dependable verification procedure, when combined with voice-activated services and secure access systems, not only solves au-thentication issues but also improves user experience overall. Our suggested method provides a thorough blueprint for developing a lightweight voice authentication system that is based on text. This system leverages an effective encoder model to convert voice spectrograms, making deployment easy even on edge devices with limited resources. Various dimensionality reduction techniques are explored to obtain optimal voice embeddings that capture speaker uniqueness while minimizing model complexity. A key novelty is the application of model compression techniques including lightweight architectures and Siamese Networks to obtain highly condensed voice embeddings for each user, reducing storage and infrastructure costs compared to traditional voice authentication methods. The proposed lightweight spectrogram embeddings leverage language-agnostic acoustic features, enabling language-independent speaker verification. Additionally, dimensionality reduction applied during voice registration allows the capturing of discriminative voice characteristics in a low-dimensional compact feature space. This significantly cuts down the storage requirements and infrastructure costs per user compared to standard voice biometric approaches, while retaining competitive verification performance. The architecture is optimized for responsiveness by leveraging lightweight frameworks. The proposed system delivers competitive voice authentication capabilities while minimizing memory, computational, and energy footprints. This makes the system useful for integration into smart devices and paves the way for ubiquitous voice biometrics.

This paper explores the problems caused by spam messages on the internet. It talks about different types of spam, like email spam and fake advertisements on social media. The paper also looks at how spam can make things difficult for people, like draining resources and tricking them into clicking on harmful links. The paper suggests fighting against spam by enabling filters in emails, using third-party spam filtering, avoiding clicking suspicious links, reporting, and blocking and avoiding using personal or professional emails. Lastly, the paper discusses what might happen with spam in the future and the challenges we may face. Overall, the paper aims to help and understand then find ways to deal with it for a safer and smoother experience.

Emotions are considered to convey much meaning in communication. Hence, artificial methods for emotion categorization are being developed to meet the increasing demand to introduce intelligent systems, such as robots, into shared workspaces. Deep learning algorithms have demonstrated limited competency in categorizing images from posed datasets with the main features of the face being visible. However, the use of sunglasses and face masks are common in our daily lives, especially with the outbreak of communicable diseases such as the recent coronavirus. Anecdotally, partial coverings of the face reduce the effectiveness of human communication, so would this have hampering effects on computer vision, and if so, would different emotion categories be affected equally? Here, we analyze the performance of emotion classification systems when faces are partially covered with simulated sunglasses and face masks. Deep neural networks consider all pixels in an image as equally important unlike the neuroscientific findings on how humans recognize emotions. Hence, we propose a method that considers different constituent parts (e.g. mouth, eyes, and jaw) separately, giving more attention to relevant (uncovered) regions of the face. The method is compared with three standard, partial coverings-based and attention-based methods. We found that face coverings worsen emotion categorization by up to 74% for the state-of-the-art methods, whereby emotion categories are affected differently by different coverings, e.g. clear mouth coverings have little effect on categorizing happiness, but sadness is affected badly. The proposed method (on average 60.43%) has significantly improved the performance over the standard deep learning (< 46% on average), partial coverings-based (< 47% on average), as well as the attention-based (< 51% on average) methods for both the CK+, KDEF, and RAF-DB datasets when faces were partially covered with sunglasses or different face masks.

We compute the weight distribution of R(4, 9) by combining the approach described in D. V. Sarwate's Ph.D. thesis from 1973 with knowledge on the affine equivalence classification of Boolean functions. To solve this problem posed, e.g., in the MacWilliams and Sloane book [12, p. 447], we apply a refined approach based on the classification of Boolean quartic forms in eight variables due to Ph. Langevin and G. Leander, and recent results on the classification of the quotient space R(4, 7)/R(2, 7) due to V. Gillot and Ph. Langevin.

Unmanned Aerial Vehicle (UAV) detection in real-time is an emerging field of study that focuses on computer vision and deep learning algorithms. However, the increasing use of UAVs in numerous applications has generated worries about possible risks and misuse. The purpose of this research is to detect UAVs, under adverse weather conditions (such as rain) and image distortions (such as motion blur and noise). The goal is to examine how these adverse conditions affect UAV detection performance and to provide techniques to increase model robustness. To achieve this, a custom training dataset was constructed by combining multiple existing datasets, supplementing them with complex backgrounds. In addition, a custom testing dataset was generated containing UAV images affected by adverse conditions.  On the proposed dataset, the performance of well-known object detection algorithms including YOLOv5, YOLOv8, Faster-RCNN, RetinaNet, and YOLO-NAS was investigated. In comparison to clean images, the results demonstrated a considerable performance decrease under adverse conditions. However, training the models on the augmented dataset containing samples of distorted and weather-affected images significantly enhanced the models' performance under challenging settings. These findings highlight the importance of taking adverse weather conditions into account during model training and underscore the significance of data enrichment for improving model generalization. The work also accentuates the need for further research into advanced techniques and architectures to ensure reliable UAV detection under extreme weather conditions and image distortions. (Note: This is a pre-print of a paper submitted to IEEE for potential journal publication and final version may vary upon acceptance and publication)

The performance of machine learning models is closely linked to the quality of training data, underpinning the 'garbage in, garbage out' principle. Label noise in datasets is a key challenge in training and evaluation. This study introduces two innovative ChatGPT-based methods, ChatGPT-Predict and ChatGPT-Detect, for effective noise detection in labeled datasets. We assess the efficacy of these methods against conventional vote-based techniques, focusing on factors like noise characteristics , dataset complexity, and the impact of prompt-engineering. Comprehensive evaluations using both artificial and real-world datasets demonstrate the adaptability of our methods to different noise types and levels. Key findings emphasize the critical role of prompt design in language model performance and the distinct contrasts in handling artificial versus real-world noise. The research acknowledges potential limitations due to prompt design variability and suggests possible enhancements with more advanced models like GPT-4. Future research avenues include applying these methods with GPT-4, exploring diverse prompt templates, and extending the methodology to real-world datasets with high noise levels. This study contributes to the field by refining noise detection methodologies, thereby enhancing the robustness and reliability of machine learning models.

Masking is one of the most well-established methods to thwart side-channel attacks. Many masking schemes have been proposed in the literature, and code-based masking emerges and unifies several masking schemes in a coding-theoretic framework. In this work, we investigate the side-channel resistance of code-based masking from a non-profiling perspective by utilizing correlation-based side-channel attacks. We present a systematic evaluation of correlation attacks with various higher-order (centered) moments and then present the form of optimal correlation attacks. Interestingly, the Pearson correlation coefficient between the hypothetical leakage and the measured traces is connected to the signal-to-noise ratio in higher-order moments, and it turns out to be easy to evaluate rather than launch repeated attacks. We also identify some ineffective higher-order correlation attacks at certain orders when the device leaks under the Hamming weight leakage model. Our theoretical findings are verified through both simulated and real-world measurements.

Large language models are experiencing a significant surge of attention and rapid development. It is happening mainly due to the publication of OpenAI's ChatGPT models: GPT3.5-turbo and GPT-4. This article uses prompt engineering to present an innovative approach to synthetic data generation and knowledge distillation. Specifically, we focus on three methods: basic prompts, composite prompts, and similarity prompts. This research aims to investigate the potential of these techniques to address the problem of unbalanced datasets, a common issue in machine learning applications. Experimental results reveal that none of the prompt-based strategies achieve scores on par with the entire dataset. However, the similarity prompts method shows promising potential, outperforming other approaches. The study suggests a significant opportunity to develop these techniques further to generate more diverse synthetic data. Although the results are preliminary, they open up exciting possibilities for future research in this area, including integrating more advanced versions of Large Language Models and exploring other machine learning domains.

In conventional 3D printing, the layer-by-layer approach leads to mechanical weaknesses, particularly in the vertical tensile strength (Z-axis) and the shear resistance between layers. To address this, our study introduces a novel non-planar toolpath planning framework for 3-axis printers, named WeaveX, inspired by nature. This method involves moving the nozzle in the XY plane while periodically adjusting its height in the Z-axis, enhancing interlayer bonding and shear resistance. We developed two distinct toolpath designs: Scheme 1, which varies layer thickness, and Scheme 2, which maintains a consistent layer thickness. These designs were closely examined to understand their impact on toolpath width and layer thickness, considering various parameters. Both schemes resulted in "dumbbell"-shaped toolpath geometries, a characteristic that can be lessened by reducing print speed. Mechanical tests revealed that objects printed using these schemes significantly outperform traditional planar toolpath methods in terms of mechanical strength, showing improvements of 31.9% and 67.5% in interlayer shear resistance. Notably, these new strategies can be combined with each other or with conventional methods, broadening their potential applications.