Search

A fundamental task in natural language processing (NLP) is part of speech (PoS) tagging. PoS tagging is crucial to many NLP applications, including question answering, machine translation, syntactic parsing, speech recognition, and semantic parsing. PoS tagging is a task for labeling sequences in which a tagger/system tags each word with its appropriate part of speech label. In NLP, PoS tagging is often considered as a language-specific task. Similarly, Pashto is a language that has not been explored regarding PoS tagging. Therefore, this research focuses on the PoS tagging considering the Pashto language and provides a baseline accuracy. The research has twofold benefits. First, it introduces a Pashto tag set that contains 2,81,205 words of the Pashto language. All these words are tagged with 17 unique PoS tags. Second, it proposes a deep learning-based model by examining classic Recursive Neural Networks (RNN) and Bidirectional Long Short Term Memory Networks (BLSTM). The results show promising performances when used with the word embedding technique. The proposed approach achieved 98.82% accuracy as a baseline on the test dataset by using the BLSTM model along with word embedding.

Because cardiovascular disease (CVD) is still one of the world’s leading causes of death, sophisticated predictive models are required for early detection and prevention. This study examined how to make and compare different CVD prediction models using a large dataset that included biochemical, clinical, and demographic information about each person. During the preprocessing stage, we took great care to ensure the data’s accuracy and quality. We have utilized a variety of machine learning algorithms such as random forest, logistic regression, support vector machines, and deep learning neural networks. We assessed the performance of these models using the accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC-ROC). Our findings show that while more sophisticated algorithms—especially deep learning models—perform better at spotting possible instances of CVD, more conventional models—such as regression—offer significant predictive power. We also investigated the role that feature selection has in improving the interpretability and efficiency of the model. This study highlights the potential of machine learning to transform CVD prediction and emphasizes the importance of using many forms of data to provide a thorough risk evaluation. Our research adds to the continuing efforts in personalized medicine by providing information on creating more precise and effective predictive tools for cardiovascular health management.

Online learning has accelerated with the development of the Internet and communication technology. The widely accessible open online courses are delivered using digital environments that allow students to participate at speed and location. Virtual learning environments (VLEs) have developed quickly in recent years, giving students access to high-quality digital resources. Online learning environments have numerous benefits but drawbacks, including poor engagement, high dropout rates, low engagement, and self-regulated behavior, making students define their aims. Forecasting failed students in a VLE can help organizations and teachers improve their pedagogical practices and make data-driven decisions. This work proposes a Hybrid Deep Learning (HDL) approach to predict students’ performance utilizing ECNN (Enhanced Convolution Neural Networks) Resnet model-based classification algorithms. The HDL approach is evaluated using the OULAD (Open University Learning Analytics Dataset), which provides a comprehensive and reliable assessment of the model’s performance. The hybrid DLT approaches, demonstrating superiority, exhibited greater prediction accuracy than the existing classifiers. Additionally, the models’ accuracy increases by about 95.67%, higher than other approaches are DFFNN model (93.9%) and MLP model (71.41%).

Existing fully-supervised semantic segmentation methods have achieved good performance. However, they all rely on high-quality pixel-level labels. To minimize the annotation costs, weakly-supervised methods or semi-supervised methods are proposed. When such methods are applied to the infrared ship image segmentation, inaccurate object localization occurs, leading to poor segmentation results. In this paper, we propose an infrared ship segmentation (ISS) method based on weakly-supervised and semi-supervised learning, aiming to improve the performance of ISS by combining the advantages of two learning methods. It uses only image-level labels and a minimal number of pixel-level labels to segment different classes of infrared ships. Our proposed method includes three steps. First, we designed a dual-branch localization network based on ResNet50 to generate ship localization maps. Second, we trained a saliency network with minimal pixel-level labels and many localization maps to obtain ship saliency maps. Then, we optimized the saliency maps with conditional random fields and combined them with image-level labels to generate pixel-level pseudo-labels. Finally, we trained the segmentation network with these pixel-level pseudo-labels to obtain the final segmentation results. Experimental results on the infrared ship dataset collected on real sites indicate that the proposed method achieves 71.18% mean intersection over union, which is at most 56.72% and 8.75% higher than the state-of-the-art weakly-supervised and semi-supervised methods, respectively.

Localization of motor unit (MU) innervation zones (IZs) is an important step in several clinical and non-clinical applications, such as 1) acquisition of surface electromyogram (sEMG) signal for accurate estimation of its amplitude and other parameters by avoiding placing electrodes on IZs, 2) accurate estimation of the EMG-Force relationship, 3) effective injection of Botulinum Toxin in Post-stroke Spasticity near the IZs, and 4) guiding obstetricians to perform episiotomy during child delivery by avoiding cutting near the IZs of External Anal Sphincter (EAS) muscle. The most minimally invasive way to identify the location of motor unit innervation zones (IZs) in any muscle, including the External Anal Sphincter (EAS) muscle, is to use multi-channel surface electromyography (sEMG) signals. In this manuscript, we propose a novel approach for automatic muscle motor unit innervation zone localization using multi-channel electromyography (EMG) signals. Our method is based on a linearity measure derived from eigenvalues of the Hessian matrix. The motor unit action potential (MUAP) propagation pattern is first detected in the spatio-temporal sEMG images using a Linearity measure based on eigenvalues of the Hessian matrix. The corresponding MU IZs is then identified as the starting point of propagation of the MUAP. A software is also developed which can be used to record and visualize the signals acquired from EAS and other muscles, detect, and display the IZs and more importantly compute and display the histogram of the IZs and generate reports which will help the obstetrician while performing episiotomy during child delivery to avoid cutting vulnerable regions that may lead to fecal incontinence at later age. The evaluations,on both simulated and experimental EMG signals, demonstrate the effectiveness and robustness of our proposed approach. Compared to existing methods, our approach achieves higher accuracy in innervation zone localization on experimental and simulated signals with mean absolute error of 0.53 and 0.53 inter electrode distance (IED).

A metaheuristic approach based on the nature-inspired and well-known Grey Wolf Optimization algorithm (GWO) was employed in this study to design an approach for retrieving strong designs of $8\times 8$ substitution boxes (S-boxes). The GWO was developed as a novel metaheuristic based on inspiration from grey wolves and how they hunt. The ability of the GWO to quickly explore the search space for the near/optimal feature subsets that maximize any given fitness function (in consideration of its distinctive hierarchical structure) aids in the construction of strong S-boxes that can satisfy the required criteria. However, when tackling optimization problems, GWO may experience the problem of premature convergence. Therefore, a variant of GWO called Crossover Grey Wolf Optimizer (XGWO) has been proposed in this study. The performance of the proposed novel approach was evaluated using numerous cryptographic performance metrics, including bijective property, bit independence, strict avalanche, linear probability, and I/O XOR distribution and the result was contrasted with a couple of existing S-box creation techniques. Overall, the results of the experiment showed that the suggested S-box design had adequate cryptographic features.

The Social Internet of Things (SIoT) has revolutionized user experience through various applications and networking services like Social Health Monitoring, Social Assistance, Emergency Alert Systems, and Collaborative Learning Platforms. However, transferring different types of data between the interconnected objects in the SIoT environment, including sensor data, user-generated data, and social interaction data, poses challenges due to their high dimensionality. This paper presents an alternative SIoT method that improves resource efficiency, system performance, and decision-making using the Barnacles Mating Optimizer (BMO). The BMO incorporates Triangular mutation and dynamic Opposition-based learning to enhance search space exploration and prevent getting stuck in local optima. Two experiments were conducted using UCI datasets from different applications and SIoT-related datasets. The results demonstrate that the developed method, DBMT, outperforms other algorithms in predicting social-related datasets in the IoT environment.

Fog computing (FC) has recently emerged as a promising new paradigm that provides resource-intensive Internet of Things (IoT) applications with low-latency services at the network edge. However, the limited capacity of computing resources in fog colonies poses great challenges for scheduling and allocating application tasks. In this paper, we propose an intelligent scheduling strategy algorithm in an FC system based on multi-objective deep reinforcement learning (MODRL) to select nodes for task processing (fog nodes or cloud nodes) based on three objectives: node current load, node distance, and task priority. The proposed model addresses two main problems: task allocation and task scheduling. We employ three deep reinforcement learning (DRL) agents based on a deep Q network (DQN), one for each objective. However, this is a more challenging scenario because there is a trade-off among these objectives, and eventually, each algorithm may select different processing nodes according to its own objective, which brings us to a Pareto front problem. To solve this problem, we propose using multi-objective optimization, a multi-objective evolutionary algorithm based on decomposition (MOEA/D), and a non-dominated sorting genetic algorithm (NSGA2), which are multi-objective optimization algorithms that can choose the optimal node by considering three objectives. The simulation results show that our proposed intelligent scheduling strategy could achieve better outcomes for the various employed performance, efficiency, and adaptability metrics: Task Completion Time, Makespan, Queueing Delay, Propagation Delay, Transmission Delay, Processing Delay, Computational Delay, Latency, CPU Load, and Storage Utilization, with an average value of 2.02 ms, 10 ms, 2 ms, 9.9 ms, 25 ms, 1.0 ms, 3.5 ms, 10 %, and 99 %, respectively, compared with the existing related research studies.

Features Selection (FS) techniques have been applied to several real-world applications which contain high dimension data. These FS techniques have main objectives that aim to achieve them, such as removing irrelevant features and increasing classification accuracy. This is considered a bi-objectives optimization problem that requires a suitable technique that can balance between the objectives. So, different sets of FS techniques have been developed, and those techniques that depend on meta-heuristic (MH) established their performance overall traditional FS techniques. However, these MH approaches still require more enhancement to neutralize their exploration and exploitation abilities during the searching process. Enhancing the meta-heuristic optimization algorithm using the perspective of fractional calculus (FC) is an attractive and novel approach. In this paper, the slime mould algorithm (SMA) is modified using the FC for handling the optimizer drawback of the inefficient diversification phase. As a result, a fractional-order SMA is proposed to avoid the local solutions and discover the search landscape efficiently via considering a historic memorize of agents’ positions. The proposed FOSMA is applied to extract features from a set of real-world data and increase classification accuracy. For boosting the optimizer performance while processing with these datasets, the rough set (RS) is used as the fitness function to handle the uncertainty inside the real-world data. Finally, the proposed FOSMA’s results are compared with a set of well-known FS techniques to investigate its performance. The comparison illustrates the superiority of FOSMA in providing high accuracy.

K-Nearest Neighbor (kNN) is an efficient algorithm used in many applications, e.g., text categorization, data mining, and predictive analysis. Despite having a high computational complexity, kNN is a candidate for hardware acceleration since it is a parallelizable algorithm. This paper presents an efficient novel architecture and implementation for a kNN hardware accelerator targeting modern System-on-Chips (SoCs). The architecture adopts a selection-based sorter dedicated for kNN that outperforms traditional sorters in terms of hardware resources, time latency, and energy efficiency. The kNN architecture has been designed using High-Level Synthesis (HLS) and implemented on the Xilinx Zynqberry platform. Compared to similar state-of-the-art implementations, the proposed kNN provides speedups between $1.4\times $ and $875\times $ with 41% to 94% reductions in energy consumption. To further enhance the proposed architecture, algorithmic-level Approximate Computing Techniques (ACTs) have been applied. The proposed approximate kNN implementation accelerates the classification process by $2.3\times $ with an average reduced area size of 56% for a real-time tactile data processing case study. The approximate kNN consumes 69% less energy with an accuracy loss of less than 3% when compared to the proposed Exact kNN.

In order to accelerate the performance of various Natural Language Processing tasks for Roman Urdu, this article for the very first time provides 3 neural word embeddings prepared using most widely used approaches namely Word2vec, FastText, and Glove. The integrity of generated neural word embeddings is evaluated using intrinsic and extrinsic evaluation approaches. Considering the lack of publicly available benchmark datasets, it provides a first-ever Roman Urdu public dataset which consists of 3241 sentiments annotated against positive, negative, and neutral classes. To provide benchmark baseline performance over the presented dataset for Roman Urdu sentiment analysis, we adapt diverse machine learning (Support Vector Machine, Logistic Regression, Naive Bayes), deep learning (convolutional neural network, recurrent neural network), and hybrid deep learning approaches. Performance impact of generated neural word embeddings based representation is compared with other most widely used bag of words based feature representation approaches using diverse machine and deep learning classifiers. In order to improve the performance of Roman Urdu sentiment analysis, it proposes a novel precisely extreme multi-channel hybrid methodology which makes use of convolutional and recurrent neural networks along with pre-trained neural word embeddings. The proposed hybrid approach outperforms adapted machine learning approaches by the significant figure of 9% and deep learning approaches by the figure of 4% in terms of F1-score.

Trust management becomes an urgent requirement in the cloud environment and a trust relationship between service user and service provider is required. Trust is the estimation of the ability of cloud resources in completing a task based on some criteria such as availability, reliability, and resource processing power. In this paper, an enhanced QoS-based model for evaluating the trustworthiness of the cloud provider is introduced. The proposed model calculates the accumulative trust value which is updated dynamically at each transaction and reflects the current or latest transaction of the provider in the cloud. The trustworthiness of a cloud resource is evaluated based on its provider reputation history from user feedback ratings based on the covariance mathematical technique to evaluate the credibility of the user's feedback. The trustworthiness of a cloud resource is also evaluated by calculating the computing power of resources at run-time. Experimental results confirm the effect of user opinion and resources processing speed on trust value calculation, which in turn assesses the trustworthiness of the cloud provider. The simulation has been performed using the CloudSim with the platform Eclipse for developing the proposed model.

Several optimization problems from various types of applications have been efficiently resolved using available meta-heuristic algorithms such as Particle Swarm Optimization and Genetic Algorithm. Recently, many meta-heuristic optimization techniques have been extensively reported in the literature. Nevertheless, there is still room for new optimization techniques and strategies since, according to the literature, there is no meta-heuristic optimization algorithm that may be considered as the best choice to cope with all modern optimization problems. This paper introduces a novel meta-heuristic optimization algorithm named Dynamic Group-based Optimization Algorithm (DGCO). The proposed algorithm is inspired by the cooperative behavior adopted by swarm individuals to achieve their global goals. DGCO has been validated and tested against twenty-three mathematical optimization problems, and the results have been verified by a comparative study with respect to state-of-the-art optimization algorithms that are already available. The results have shown the high exploration capabilities of DGCO as well as its ability to avoid local optima. Moreover, the performance of DGCO has also been verified against five constrained engineering design problems. The results demonstrate the competitive performance and capabilities of DGCO with respect to well-known state-of-the-art meta-heuristic optimization algorithms. Finally, a sensitivity analysis is performed to study the effect of different parameters on the performance of the DGCO algorithm.

Big data is an important factor almost in all nowadays technologies, such as, social media, smart cities, and internet of things. Most of standard classifiers tends to be trapped in local optima problem when dealing with such massive datasets. Hence, investigating new techniques for dealing with such massive data sets is required. This paper presents a novel imbalanced big data mining framework for improving optimization algorithms performance by eliminating the local optima problem consists of three main stages. Firstly, the preprocessing stage, which uses the LSH-SMOTE algorithm for solving the class imbalance problem, then it uses the LSH algorithm for hashing the data set instances into buckets. Secondly, the bucket search stage, which uses the GWO for training bidirectional recurrent neural network BRNN and searching for the global optimum in each bucket. Lastly, the final tournament winner stage, which uses the GWO+BRNN for finding the global optimum of the whole data set among all global optimums from all buckets. Our proposed framework LSHGWOBRNN has been tested over 9 data sets one of them is big data set in terms of AUC, MSE, against seven well-known machine-learning algorithms (Naive Bayes, Random Tree, Decision Table, and AdaBoostM1, WOA+MLP, GWO+MLP, and WOA+BRNN), then, we tested our algorithm over four well-known data sets against GWO+MLP, ACO+MLP, GA+MLP, PSO+MLP, PBIL+MLP, and ES+MLP in terms of classification accuracy and MSE. Our experimental results have proved that our proposed framework LSHGWOBRNN has provided high local optima avoidance, and higher accuracy, less complexity and overhead.