Machine learning (ML) has revolutionized numerous fields by providing sophisticated tools for data analysis and predictive modeling. Among these tools, Support Vector Machines (SVMs) stand out due to their strong theoretical foundations and practical efficacy. SVMs are particularly well-suited for tasks involving classification and regression, where they excel at finding the optimal hyperplane that separates data points of different classes with maximum margin.This paper aims to provide an in-depth review of SVM applications in various research areas, supported by an examination of relevant literature. The integration of SVMs with other advanced methodologies, such as deep learning and fuzzy systems, will also be discussed to demonstrate their versatility and potential for further innovations.

Public-key Encryption with Keyword Search (PEKS) is a promising cryptographic mechanism that enables a semi-trusted cloud server to perform (on-demand) keyword searches over encrypted data for data users. Existing PEKS schemes are limited to precise or fuzzy keyword searches, creating a gap given the widespread use of wildcards for rapid searches in real-world applications. To address this issue, several wildcard keyword search schemes have been proposed to support wildcard searches in the public key setting. However, these schemes suffer from inefficiency and/or inflexibility. Worse yet, they are all vulnerable to (insider) keyword guessing attacks (KGA), which is highly effective when the keyword space is polynomial in size. To address these vulnerabilities, this paper first proposes a new wildcard keyword search scheme called Public-key Encryption with Wildcard Search (PEWS), which is built based on the standard Decisional Diffie-Hellman (DDH) assumption. The complexity of all algorithms in PEWS increases linearly with the number of keywords, while remaining almost constant or even decreasing linearly with the number of wildcards. To resist against (insider) KGA, we further extend PEWS into the first Public-key Authenticated Encryption with Wildcard Search (PAEWS) scheme. Our PEWS and PAEWS schemes are highly flexible, supporting searches not only for sets of keywords but also for any number of wildcards positioned anywhere within the keyword set. We conduct a comprehensive performance evaluation of our PEWS and PAEWS, while also comparing PEWS with the state-of-the-art scheme in the public key setting. The experimental results demonstrate that both PEWS and PAEWS are efficient and practical in computation and communication, and the experimental comparisons illustrate that PEWS achieves significant improvements in computational efficiency.

Facial expression recognition (FER) plays an important role in human-computer interaction, psychology, and various applications. It requires significantly large annotated data with practical variations. Existing FER datasets in the uncontrolled environment are small and with a limited number of expressions. In addition, they are lacking with facial ambiguity and crosscultural diversity. To address this need, we have proposed a more comprehensive FER dataset (called FER20E) and an emotion theory, exploring a remarkable expansion to include 20 discrete facial expressions. FER20E contains more than 100,000 facial images with existing eight and twelve unexplored emotions. It takes into account cultural, environmental, and demographic variations like illumination, pose, rotation, scale, age, gender, and occlusion. To reduce the manual intervention for data annotation, we have additionally designed an automatic annotation tool by utilizing a lightweight deep convolutional neural network (DCNN). To evaluate its efficacy, various state-of-the-art models are applied to it. This research not only advances FER but also has broader implications, including enhancing security measures, improving health monitoring, and potentially aiding in the early diagnosis of mental health conditions, thereby contributing to overall well-being and mental health awareness. The code and dataset will be available at https://github.com/akstheme/FER20E.

Dementia is a medical syndrome resulting in substantial memory loss or deterioration and other cognitive capabilities, beyond the normal aging process [2]. Alzheimer's disease (AD) is the leading cause of dementia in aged adults, affecting up to 70% of the dementia patients, and posing a serious public health hazard in the twenty-first century. With the growing lifespan, the number of AD patients is also increasing, and estimated that by the year 2050, 135 million people will be affected [1]. With age being the predominant dementia factor, the dominance ranges from 1-2 percent in the age group of 65 to 30 percent at 85. AD is a progressive, irreversible and neurodegenerative disease with a long pre-clinical period, affecting brain cells leading to memory loss, misperception, learning problems, and improper decisions. Given its significance, presently no treatment options are available, although disease advancement can be retarded through medication. Unfortunately, AD is diagnosed at a very later stage, after irreversible damages to the brain cells have occurred, when there is no scope to prevent further cognitive decline. Individual diagnoses of AD are now based mostly on neuropsychological testing and clinical examination, but only post-mortem brain study may confirm the final diagnosis. The use of non-invasive neuroimaging procedures capable of detecting AD at preliminary stages is crucial for providing treatment retarding disease progression, and has stood as a promising area of research. We conducted a comprehensive assessment of papers employing machine learning to predict AD using neuroimaging data. Most of the studies employed brain images from Alzheimer's disease neuroimaging initiative (ADNI) dataset, consisting of magnetic resonance image (MRI) and positron emission tomography (PET) images. The most widely used method, the support vector machine (SVM), has a mean accuracy of 75.4 percent, whereas convolutional neural networks (CNN) have a mean accuracy of 78.5 percent. Better classification accuracy has been achieved by combining MRI and PET, rather using single neuroimaging technique. Overall, more complicated models, like deep learning, paired with multimodal and multidimensional data (neuroimaging, cognitive, clinical, behavioural and genetic) produced superlative results. However, promising results have been achieved, still there is a room for performance improvement of the proposed methods, providing assistance to healthcare professionals and clinicians.

Over the past few years, medical image processing has increased in use of deep learning algorithms, especially in the analysis of magnetic resonance (MR) scans. MRI is a crucial diagnostic tool for Alzheimer's disease (AD), a prevalent type of dementia that ranks seventh among fatal illnesses globally. As there is no known cure for Alzheimer's disease, early detection and intervention are vital to prevent its irreversible progression. This study proposes a comprehensive framework for detecting Alzheimer's disease that employs convolutional neural networks (CNNs) and deep learning approaches. We applied transfer learning to pretrained deep learning models rather than training them from scratch. Three distinct pretrained CNN models (VGG-19, ResNet-50, and Inception V3) with a fine-tuned transfer learning approach were used for five-way classification of AD. We employed the ADNI dataset, which includes MRI scans from 608 patients across five classes: Alzheimer's disease (AD), early mild cognitive impairment (EMCI), mild cognitive impairment (MCI), late mild cognitive impairment (LMCI), and normal control (NC). The models' performance was evaluated based on eight metrics: accuracy, precision, sensitivity/recall, specificity, error rate, false positive rate, F1-score, and kappa. Our findings indicate that the ResNet-50 architecture outperformed other pretrained models, achieving the highest overall accuracy of 98.7% for multiclass AD classification. Additionally, the ResNet-50 model excelled in classifying the EMCI category with an accuracy of 99.25%, indicating its effectiveness in detecting early signs of memory impairment. The proposed framework surpasses the performance of previous studies in terms of overall accuracy, sensitivity, and specificity, setting a new benchmark for five-way AD classification. The outcomes of this study will contribute significantly to early prevention efforts by enabling Alzheimer's disease to be detected before it progresses irreversibly. In conclusion, this research represents a promising approach for improving the early detection and classification of Alzheimer's disease using deep learning methods with MRI data.

Alzheimer's disease (AD) is a brain disorder that impairs one's ability to function on a daily basis. Machine learning (ML) has significantly helped clinicians in quickly and accurately identifying AD through imaging-based methods. Advanced neuroimaging techniques have made it possible to detect structural and functional abnormalities in the brain associated with AD. However, applying ML to clinical settings poses challenges due to differences in data distribution between training and test datasets, which can significantly impact model performance. To address this challenge, the domain adaptation (DA) approach is necessary, as data from different distribution domains present a performance gap. This review focuses on current research involving ML and DA techniques for early AD detection using neuroimaging data, and aims to encourage further research efforts to improve the clinical workflow for diagnosing and treating AD. The review also includes a discussion of the limitations of current studies and possible research directions.

Artificial intelligence ("AI") is deployed in various applications, from noise cancellation to image recognition, but AI-based products often come with high hardware and electricity costs; this makes them inaccessible for consumer devices and small-scale edge electronics.  Inspired by biological brains, deep neural networks ("DNNs") are modeled using mathematical formulae, yet general-purpose processors treat otherwise-parallelizable AI algorithms as step-by-step sequential logic.  In contrast, programmable logic devices ("PLDs") can be customized to the specific parameters of a trained DNN, thereby ensuring data-tailored computation and algorithmic parallelism at the register-transfer level.  Furthermore, a subgroup of PLDs, field-programmable gate arrays ("FPGAs"), are dynamically reconfigurable.  So, to improve AI runtime performance, I designed and open-sourced my hardware compiler: Innervator.  Written entirely in VHDL-2008, Innervator takes any DNN's metadata and parameters (e.g., number of layers, neurons per layer, and their weights/biases), generating its synthesizable FPGA hardware description with the appropriate pipelining and batch processing.  Innervator is entirely portable and vendor-independent.  As a proof of concept, I used Innervator to implement a sample 8x8-pixel handwritten digit-recognizing neural network in a low-cost AMD Xilinx Artix-7(TM) FPGA @ 100 MHz.  With 3 pipeline stages and 2 batches at about 67% LUT utilization, the Network achieved ~7.12 GOP/s, predicting the output in 630 ns and under 0.25 W of power.  In comparison, an Intel(R) Core(TM) i7-12700H CPU @ 4.70 GHz would take 40,000-60,000 ns at 45 to 115 W.  Ultimately, Innervator's hardware-accelerated approach bridges the inherent mismatch between current AI algorithms and the general-purpose digital hardware they run on.

The frequency response of many antennas includes an angle-independent factor that rises in proportion to frequency. Two passes through the radar antenna and once for the ground patch as an antenna causes a rising third-order frequency dependence to be impressed upon the received data. After a review of antenna theory and types, a simulation for polar and rectangular formatted spotlight synthetic aperture radar data over two bandwidth cases, 9-10 GHz and ultra wideband 1-10 GHz, is reported. Without correction, range resolution for the first case is minimally affected but for the latter case, range resolution is seriously degraded, rectangular data being worse. Response equalizing correction restores the resolution.

In this paper, a new conformal array antenna conical design has been proposed. It can electronically steer the main beam in azimuth and elevation to cover a hemispherical mesh surrounding the given antennas configuration. To achieve the desired beam configuration, it is absolutely essential to select the appropriate antenna control parameters. To this aim, the optimization of the neural technique is adapted to the optimal set of matching parameters. For more details, this work discusses the use of an efficiently designed artificial neural network (ANN) to model and synthesize the radiation pattern of an adaptive uniform cone lattice beam. Let's assume, then, that the learning database of the network has a finite number of target selections at certain feasible angles. Utilized a feedforward artificial neural network model called a multilayer perceptron (MLP), generates a set of appropriate outputs based on sets of input data. The obtained systems are performed in Matlab and the suggested conforming conical array shows a significant enhancement versus the existent structures in terms of 3-D scanning, size, directivity, HPBW and SLL reducing. The employed neural technique revealed its effectiveness in improving performance using the conical isotropic antenna array. To approve this work, several examples are shown. The proposed method can also be utilized for other kinds of conformal phased arrays such as spherical or cylindrical arrangements.

Large language models have dramatically transformed various fields by enhancing the capabilities of artificial intelligence in understanding and generating human language. The novel concept of predicting critical mixture ratios for continual pre-training introduces a significant advancement in optimizing the performance and efficiency of such models. Through leveraging advanced machine learning techniques, the study develops a predictive framework that systematically varies mixture ratios and evaluates their impact on model performance. The results highlight substantial improvements in accuracy, precision, recall, and F1 score, demonstrating the efficacy of optimized mixture ratios. Furthermore, the study shows that models trained with these optimized ratios exhibit enhanced generalization capabilities, maintaining high performance across unseen datasets. The research also addresses the reduction in training time, translating to operational efficiency gains and cost savings. Despite the promising findings, the study acknowledges several limitations, including the need for more diverse datasets and a deeper exploration of hyperparameter interactions. Future research directions include expanding the scope of data, integrating reinforcement learning techniques, and developing sophisticated validation methods. The study's insights and methodologies provide a valuable foundation for future research and practical implementations, aiming to create more effective and adaptable language models for various applications.

Interoperability is one of the enabling factors of real-time communications and data exchange between asynchronous data actors. Interoperability can be attained by introducing the notion of events to systems that extract data from consumed ground-truth event streams that utilize application-specific structures. Events are specific occurrences happening at a particular time and place. Event-data are observations of phenomena, or actions, as seen by different systems in Internet of Things (IoT) deployments, independent from data structures and platforms. Avena, an opensource software framework, facilitates communication among systems ranging from IoT devices, agricultural machines, and cloud-based data processing services and fosters effortless interoperability. Avena addresses the challenges of intermittent connectivity and incompatibility among precision agricultural systems, facilitating resilient IoT communications in complex agricultural networking environments. Constructed on modern distributed system architectures and leveraging overlay networks based on WireGuard tunnels and the NATS messaging system, Avena simplifies connections among software and other systems, making them inherently tolerant to communication outages and seamlessly adaptable to the dynamic nature of agriculture. Device discovery mechanisms and peer-to-peer communications between devices in precision agriculture ecosystems are integral components of Avena with eventlevel data abstraction and data interoperability. In this work, we integrated agricultural operational data which is mostly textual and manually recorded by farm employees. We have integrated Meta Ag, an open-source Android application featuring a geofence-responsive information bot that makes the in-field record-keeping process automatic and less error prone. Meta Ag, integrated into the Avena platform and powered by NATS, can give useful insights into machine and sensor data providing complete in-field operational records. With machine and sensor data now interoperable with contextual data, data-driven decision-making, including integration with biophysical models, will be easier, secure, and reliable.

Wireless Underground Sensor Networks (WUSNs) play a crucial role in precision agriculture by providing information about moisture levels, temperature, nutrient availability, and other relevant factors. However, the use of radio-frequency identification (RFID) devices for WUSNs has been relatively unexplored despite their benefits such as low power consumption. In this work, we develop a hardware platform, called OATSMobile, that enables radio-frequency identification (RFID) communications in WUSNs. OATSMobile is a mobile platform that carries a sensor reader for scanning underground RFID tags and is equipped with communication modules to transport collected data. In addition to collecting RFID data, our edge computer collects machine location and speed from a Real Time Kinematics (RTK) GPS device, soil moisture from externally installed LoRaWAN sensors, and ground-to-antenna height from a Wi-Fi enabled LiDAR sensor. Data is automatically transferred over the cell network to an Avena-based data pipeline for analysis and processing. With OATSMobile, we investigated the feasibility of RFID communications in the Ultra High Frequency (UHF) band for WUSNs. We evaluated the impact of soil moisture and temperature on RFID scanning performance by randomly placing corn seed sized RFID tags in soil blocks of different drainage classes at 2.5 cm depths and measuring the Received Signal Strength Index (RSSI) and scanning success rates at different stages of the corn growing season. The nominal antenna height was 32 cm above ground, and the machine traveled at 0.44 m/s. As a result, we confirmed that 152 out of 282 tags were detected, for an overall success rate of 53.9%. Other results, such as the impact of soil moisture, antenna height, and machine speed impact will be discussed in this paper. Finally, we outline pathways for developing communications frameworks for WUSNs applied to precision agriculture.

In this study, a Vernier time-to-digital converter (TDC) was investigated with a time resolution less than the signal transition in the circuit. A Xilinx FPGA was employed to implement the TDC. In this design, the timing difference between the two paths were generated by interconnects rather than LUT devices. Since the nature of uncertainty, a statistical model was proposed to analyze the output. Three methods were proposed to determine the output thermometer code. The time resolution can be around 5 ps in our measurement. As the resolution can be determined below the signal transition, it is promising to push the resolution down into sub-ps for FPGA-based TDCs by suitable interconnections.

This study presents a comprehensive Design-Technology Co-Optimization (DTCO) approach to evaluate and enhance power and performance in Gate-All-Around Nanosheet (GAA-Nsh) and Forksheet (Fsh) architectures. The analysis focuses on the impact of active widths and power delivery methods on the effective resistance (Reff) and capacitance (Ceff) of these devices. The research employs simulations of five-stage INVD1 ring oscillators (RO) at various metal pitches (Mx) to extract frequency and power data. Notably, a novel Gate-All-Around Forksheet (GAA-Fsh) structure is introduced, offering enhanced gate control while retaining the advantages of Fsh. The study also explores asymmetric N/PFETs within the Fsh technology, and innovative contacting approaches such as Buried Power Rail (BPR) and Backside Power Rail (BS-PR) with Backside Contact (BSC) to reduce access resistance. Results indicate that GAA-Fsh outperforms traditional GAA-Nsh and Fsh due to reduced Reff and Ceff, although it is process feasible only at larger Mx. At smaller Mx, GAA-Nsh demonstrates higher performance than Fsh at a given sheet width (WNS), but Fsh, with the advantage of additional WNS, can match GAA-Nsh performance at larger WNS. Furthermore, the BPR and BS-PR contacting schemes are found to provide similar performance. This research provides valuable insights into future semiconductor device designs, emphasizing higher performance and efficient scaling.

Progressions in technologies for renewable energy have rendered feasible fresh avenues for energy production that is sustainable. Nonetheless, melding renewable energy sources into pre-existing power grids encounters hurdles, owing to their sporadic nature. Machine learning methodologies present an auspicious remedy to enhance the assimilation of renewable energy sources and augment the efficiency and dependability of power infrastructures. Through the application of data-centric models, machine learning algorithms possess the capability to scrutinize extensive datasets to anticipate renewable energy production, streamline power generation timetables, and forecast prospective system malfunctions. This inquiry endeavors to probe the potential of utilizing machine learning to amplify the sustainability of renewable energy mechanisms. By exploiting artificial intelligence's capabilities, we can craft pioneering resolutions to tackle the intricacies linked with incorporating renewable energy sources into the grid, ultimately setting the stage for a future where energy is more sustainable.

This study presents the design and performance evaluation of a solar photovoltaic (PV)-based microgrid system for the Ola Fathia community in Mowe, Ogun State, Nigeria. The aim is to address the electricity access challenges faced by this peri-urban community through a reliable and sustainable energy solution. Using advanced modeling tools like PVsyst, the research involved a detailed analysis of the community's energy demand, solar resource assessment, and optimal system design. The proposed microgrid system, with a capacity of 92.65 kW comprising 371 solar panels and battery storage, was evaluated for its technical and economic performance. Results indicate that the solar PV microgrid can reliably meet the community's energy needs, offering significant cost savings and financial viability. The technical evaluation confirmed the system's efficiency and reliability, while the economic analysis highlighted its affordability. Social acceptance and the potential socioeconomic benefits, including job creation and improved quality of life, were also examined. Key challenges such as dependency on sunlight, energy storage limitations, and initial capital costs were identified and addressed. This research underscores the feasibility and effectiveness of solar PV-based microgrids in providing clean, reliable, and affordable electricity to underserved communities. It offers valuable insights into the technical, economic, and social aspects of deploying renewable energy systems in peri-urban areas of Nigeria. The findings support the broader goal of achieving universal access to sustainable electricity, contributing to sustainable development and environmental sustainability.

Within Industry 4.0, efficient fault diagnosis plays a pivotal role in predictive maintenance of industrial machinery. However, the challenge lies in the significant domain shift between the source (training) and target (testing) domains, which hampers the application of machine learning in engineering practice. Several approaches based on transfer learning have been proposed to cope with the lack of training data in the target domain and the related domain adaptation challenges. Those approaches leverage the knowledge from similar source domains, including related real-world applications or lab machines. Unfortunately, access to sufficient faulty data from such source domains is often restricted due to insufficient history of faults in real machines, as well as difficulties to get labeled datasets from lab machines, which is time-consuming and sometimes unfeasible. To tackle those issues, this paper proposes a novel diagnostic framework integrating digital twins and transfer learning to mitigate the limitations posed by insufficient training datasets and domain discrepancies. By leveraging digital twins, training datasets are generated as the source domain, while introducing a model update strategy based on parameter sensitivity analysis to enhance adaptability. In addition, the partial transfer diagnostic model, incorporating a double-layer attention mechanism, enables to cope with data distribution discrepancies between digital twins and real machines, as well as inconsistencies in label spaces across domains. The diagnostic framework is validated on an industrial rotating machine case study, where faulty behaviors originated by defects on the inner race, outer race, and ball of the bearing are considered. Real data from two publicly available datasets are leveraged. The results of the experimental analysis have been compared with state-of-the-art methodologies: the proposed approach is able to improve the diagnostic accuracy by over 11% in the specific case study. Therefore, the approach can effectively increase equipment reliability, optimize maintenance, and enhance operational efficiency.

Endoscopic Lumbar Discectomy (ELD) is a minimally invasive surgical procedure to remove the herniated material of a disc in the lumbar spine [1]. In the traditional approach, the surgeon holds the endoscope and inserts the instruments into its lumen to reach the surgical window and perform the discectomy. Some steps, such as opening of the ligamentum flavum, require the help of an assistant to hold the endoscope while the surgeon is occupied handling other tools: this operation results in an inevitable waste of time. To overcome this and other drawbacks of the traditional procedure, a robotic approach has been investigated, which involves the use of a collaborative robot to hold the endoscope throughout the procedure. In this context, the gripper and its actuation system must be safe, reliable and lightweight. The following paper describes a novel mechanism for manual actuation of the gripper responsible for holding the endoscope, specifically designed to allow the surgeon to have full control over the gripping force applied.