Search

Pre-evaluating scientific paper quality aids in alleviating peer review pressure and fostering scientific advancement. Although prior studies have identified numerous quality-related features, their effectiveness and representativeness of paper content remain to be comprehensively investigated. Addressing this issue, we propose a content-based interpretable method for pre-evaluating the quality of scientific papers. Firstly, we define quality attributes of computer science (CS) papers as integrity , clarity , novelty , and significance , based on peer review criteria from 11 top-tier CS conferences. We formulate the problem as two classification tasks: Accepted/Disputed/Rejected (ADR) and Accepted/Rejected (AR). Subsequently, we construct fine-grained features from metadata and knowledge entity networks, including text structure, readability, references, citations, semantic novelty, and network structure. We empirically evaluate our method using the ICLR paper dataset, achieving optimal performance with the Random Forest model, yielding F1 scores of 0.715 and 0.762 for the two tasks, respectively. Through feature analysis and case studies employing SHAP interpretable methods, we demonstrate that the proposed features enhance the performance of machine learning models in scientific paper quality evaluation, offering interpretable evidence for model decisions. • Define four criteria for quality evaluation of scientific papers: integrity, clarity, novelty, and significance. • Propose a framework for quality evaluation of scientific papers based on coarse features and knowledge entity network. • An effective algorithm for measuring the novelty and significance of scientific papers based on knowledge entity networks. • Create and release a rigorous dataset, which could serve as the gold standard for quality evaluation of scientific papers. • Conduct extensive experiments to validate the effectiveness of the proposed framework. [ABSTRACT FROM AUTHOR]

Zero-knowledge proofs introduce a mechanism to prove that certain computations were performed without revealing any underlying information and are used commonly in blockchain-based decentralized apps (dapps). This cryptographic technique addresses trust issues prevalent in blockchain applications, and has now been adapted for machine learning (ML) services, known as Zero-Knowledge Machine Learning (ZKML). By leveraging the distributed nature of blockchains, this approach enhances the trustworthiness of ML deployments, and opens up new possibilities for privacy-preserving and robust ML applications within dapps. This paper provides a comprehensive overview of the ZKML process and its critical components for verifying ML services on-chain. Furthermore, this paper explores how blockchain technology and smart contracts can offer verifiable, trustless proof that a specific ML model has been used correctly to perform inference, all without relying on a single trusted entity. Additionally, the paper compares and reviews existing frameworks for implementing ZKML in dapps, serving as a reference point for researchers interested in this emerging field. • A review of core on-chain ZKML concepts, establishing an understanding of the field. • Focus on frameworks EZKL and Orion, highlighting their approach differences. • Results from lab experiments, comparative analysis, and use case execution comparison. [ABSTRACT FROM AUTHOR]

Structured Query Language injection (SQLi) and Cross-Site Scripting (XSS) are the most renowned kinds of input validation vulnerabilities. Of late, vulnerability prediction models based on machine learning have been gaining acceptance in the domain of Web security. Such models offer an easy and effective way of dealing with web application security concerns. However, most of them, in particular, rely on complex graphs generated from source code or regex patterns based on expert knowledge. This paper proposed a method for extracting features from source code and predicting input validation vulnerabilities using machine learning algorithms. The proposed method can extract all features related to the flow of vulnerabilities among the programs and remove the features that are irrelevant to the vulnerability flow. In addition, each vulnerability's context has been assigned, providing additional data for our model to use in learning about the vulnerability context. Compared to other related methods, the feature extraction method proposed in this paper has been found to have high reusability and better performance. The best model related to the LSTM classifier had a 98.1% recall rate, a 97.9% precision, an accuracy of 98.67%, and a 99.03% area under the curve (AUC) in the test dataset. [ABSTRACT FROM AUTHOR]

Structured Query Language injection (SQLi) and Cross-Site Scripting (XSS) are the most renowned kinds of input validation vulnerabilities. Of late, vulnerability prediction models based on machine learning have been gaining acceptance in the domain of Web security. Such models offer an easy and effective way of dealing with web application security concerns. However, most of them, in particular, rely on complex graphs generated from source code or regex patterns based on expert knowledge. This paper proposed a method for extracting features from source code and predicting input validation vulnerabilities using machine learning algorithms. The proposed method can extract all features related to the flow of vulnerabilities among the programs and remove the features that are irrelevant to the vulnerability flow. In addition, each vulnerability's context has been assigned, providing additional data for our model to use in learning about the vulnerability context. Compared to other related methods, the feature extraction method proposed in this paper has been found to have high reusability and better performance. The best model related to the LSTM classifier had a 98.1% recall rate, a 97.9% precision, an accuracy of 98.67%, and a 99.03% area under the curve (AUC) in the test dataset. [ABSTRACT FROM AUTHOR]

In this paper, we proposed a machine learning and computer vision-based automated apple disease detection and recognition system based on leaf symptoms. The proposed method is composed of three parts: diseased region segmentation, feature extraction, and classification. We have segmented the infected portion of the leaf using L*a*b* space-based color segmentation method. Here, average color markers in a*b* space and the nearest neighbor method have been used for classifying each pixel into either healthy, infected, or background regions. We have extracted two types of features: one is the proposed DWT feature and another is L*a*b* space-based color histogram features. Horizontal feature fusion is performed to create the final feature vector. The feature vectors have been classified using several classifiers keeping Random Forrest as the base classifier. In this paper, the experiment is made on Plant Village dataset, where image of Apple Scab, Black Rot, and Cedar Apple Rust disease are taken for both training and testing our model. The fusion of proposed DWT and color histogram features is a novel approach in detecting and recognizing apple leaf disease, which got an accuracy of 98.63%. [ABSTRACT FROM AUTHOR]

Security issues in mobile apps are increasingly relevant as this software have become part of the daily life of billions of people. As the dominant OS, Android is a primary target for ill-intentioned programmers willing to exploit its vulnerabilities by spreading malwares. Significant research has been devoted to the identification of these malwares. The current paper is an extension of our previous effort to contribute to said research with a new benchmark of Android vulnerabilities. We proposed AndroVul, a repository for Android security vulnerabilities, that builds on AndroZoo – a well-known Android app dataset – and contains data on vulnerabilities for a representative sample of about 16,000 Android apps. The present paper adds confirmed malwares from the VirusShare dataset and explores more thoroughly the effectiveness of different machine learning techniques, with respect to the classification of malicious apps. We investigated different classifiers and feature selection techniques as well as different combinations for our input data. Our results suggest that the classifier MPL is the leading classifier, with competitive results that favorably compare to recent malware detection work. Additionally, we investigate how to classify (as benign or malicious) AndroZoo apps based on the number of antivirus flags they are tagged with. We found that different thresholds only marginally affect the machine learning classifier results and that the strictest choice (i.e. one flag) performs best on the confirmed malwares from VirusShare. [ABSTRACT FROM AUTHOR]

Today, the rapid growth of the internet and advancements in mobile technology and increased internet connectivity have brought us to a data-driven economy where an enormous amount of data is being used to train machine learning models to make strategic decisions. However, in the aftermath of a data breach by Facebook in 2018, there are some serious concerns over user data privacy and security being used to train the Machine Learning models. In this context, a new approach, Federated Machine Learning is now one of the most talked-about and recent approaches. Current research primarily focuses on Federated Learning's advantages over the traditional methods and/or its classification. However, being in a nascent stage of development as a method, certain challenges need to be addressed. This paper intends to address the totality of federated learning with a complete vulnerability assessment. During the study of the literature, it is found that security being promised as one of the key advantages of federated learning can still not be guaranteed because of some issues inherently present, and this can lead to poisoning, inference attacks and insertion of backdoors, etc. This paper intends to provide a complete picture by giving an in-depth and comprehensive analysis of Federated Learning and its taxonomy. It also provides a detailed vulnerability assessment and highlights the challenges faced in the current setting and future research directions to make federated learning a more functional, robust and secure method to train machine learning models. [ABSTRACT FROM AUTHOR]

In spite of the insight commonality may have concerning agrarian practice, fact is that nowadays agricultural science diligence is accurate, precise, data-driven, and vigorous than ever. The emanation of the technologies based on Internet of Things (IoT) has reformed nearly each industry like smart city, smart health, smart grid, smart home, including "smart agriculture or precision agriculture". Applying machine learning using the IoT data analytics in agricultural sector will rise new benefits to increase the quantity and quality of production from the crop fields to meet the increasing food demand. Such world-shattering advancements are rocking the current agrarian approaches and generating novel and best chances besides a number of limitations. This paper climaxes the power and capability of computing techniques including internet of things, wireless sensor networks, data analytics and machine learning in agriculture. The paper proposed the prediction model of Apple disease in the apple orchards of Kashmir valley using data analytics and Machine learning in IoT system. Furthermore, a local survey was conducted to know from the farmers about the trending technologies and their effect in precision agriculture. Finally, the paper discusses the challenges faced when incorporating these technologies in the traditional farming approaches. [ABSTRACT FROM AUTHOR]

This paper presents a new contribution to the recommendation system following E-commerce reviews. Here, it implements a dictionary-based sentiment analysis based on four main steps (i) Pre-processing (ii) Semantic word extraction (iii) Feature Extraction (iv) Classification. Initially, the tweets or reviews extracted from the database are subjected to pre-processing, which involves three processes: stop word removal, stemming, and blank space removal. In the semantic word extraction process, the semantic words from the dictionary will be extracted by matching with the extracted keywords. The next process is the feature extraction, which extracts joint holoentropy and cross holoentropy of all keywords, and further proceeds with conditional holoentropy-based feature selection process. Finally, the selected features are subjected to a classification process using Deep Belief Network (DBN), which is a well-performing deep learning algorithm. Moreover, the main contribution of the paper relies on the improvement of DBN architecture, where the number of hidden neurons and activation function is optimally tuned by a hybrid optimization algorithm termed as Grey linked Chick Update-based Chicken Swarm Optimization (GCU-CSO). At last, the comparative analysis of proposed and conventional algorithms in terms of various performance metrics proves the efficacy of the proposed E-commerce recommendation system. [ABSTRACT FROM AUTHOR]

Emotion recognition based on electroencephalography (EEG) signal features is now one of the booming big data research areas. As the number of commercial EEG devices in the current market increases, there is a need to understand current trends and provide researchers and young practitioners with insights into future investigations of emotion recognition systems. This paper aims to evaluate popular consumer-grade EEG devices' status and review relevant studies that examined the reliability of these low-cost devices for emotion recognition over the last five years. Additionally, a comparison with research-grade devices is conducted. This paper also highlights EEG-based emotion recognition research's key areas, including different feature extraction capabilities, characteristics, and machine learning algorithms. Finally, the main challenges for building an EEG-based emotion recognition system, focusing on the data collection process with commercial EEG devices and machine learning algorithms' performance, are presented. [ABSTRACT FROM AUTHOR]

Software-as-a-service (SaaS) is a license for access to a specific cloud application through the Internet. However, these services are delayed and sometimes totally disrupted because of the Internet's unavailability that provides opportunities for a lot of attacks. Research into the security of the cloud focuses mainly on prohibiting malicious users from using the cloud to launch attacks, such as those currently done by botnets, including launching a DDoS attack, sending spam, etc. Because SaaS uses computational power from both servers of cloud computing providers and machines of customers, we argue that SaaS may be elaborately exploited in an unprecedented way as an attack vector for botnets. The previous research work has developed a novel framework for detecting the attacks with the aid of a deep learning approach (Doriguzzi-Corin et al., 2020). Although the attack node in the network has been identified, they were not diminished. An enhanced and powerful adversary model is to be provided as a solution to this problem. Hence, this paper intends to develop a novel framework for attack node mitigation using a secure SaaS Framework. The main contribution of this paper provides an attack detection process that takes place in Deep Belief Network (DBN), in which the weight, as well as activation function, are fine-tuned with Median Fitness oriented Sea Lion Optimization algorithm (MFSLnO). If DBN detects an attack node, the control is transferred to a lightweight bait approach that reliably mitigates the most common attack nodes without disrupting regular connections will be deployed. The performance of the proposed work yielded the best results over the traditional models with a packet loss ratio of 16% and throughput of 89%. [ABSTRACT FROM AUTHOR]

Affect detection from text has captured the attention of researchers recently. This is due to the rapid use of social media sites (e.g. Twitter, Facebook), which allows users to express their feelings, emotions, and thoughts in textual format. Analyzing emotion-rich textual data of social networks has many real-life applications. The context of an emotional text can be measured by analyzing certain features of this rich source of emotional information. Classifying text into emotional labels/intensities is considered a difficult problem. This paper resolves one of the state-of-the-art NLP research emotion and intensity detection tasks using Deep Learning and ensemble implementations. In this paper, we developed several innovative approaches; (a) bidirectional GRU_CNN (BiGRU_CNN), (b) conventional neural networks (CNN), and (c) XGBoost regressor (XGB). The ensemble of BiGRU_CNN, CNN, and XGB is used to solve an emotion intensity (EI-reg) task of the SemEval-2018 Task1 (Affect in Tweets). Our proposed ensemble approach was evaluated using a reference dataset of the SemEval-2018 Task1. Results show that our approach is well above the baseline for this task. It also achieved a Pearson of (69.2%), with an enhancement of 0.7% in comparison with previous best performing models. [ABSTRACT FROM AUTHOR]

The development of stereo matching algorithm is still one of the challenging problems, especially in ill-posed regions. Hence, this article presents a survey on the algorithm frameworks related to the stereo matching algorithm. Based on the early survey that had been conducted, two major frameworks available in current stereo matching algorithm development, they are traditional and artificial intelligence (AI) frameworks. Most of the traditional methods are very low accuracy compared to the AI-based approach. This can be observed in the standard benchmarking dataset, such as from the KITTI and the Middlebury, where AI methods rank at the top of the accuracy list. Additionally, the trend for solving computer vision problems uses AI or machine learning tools that become more apparent in recent years. Thus, this paper is focusing on the survey between the deep learning frameworks, which is one of the machine learning tools related to the convolutional neural network (CNN). Several mixed approaches between CNN based method and traditional handcraft method, as well as the end to end CNN method also discussed in this paper. [ABSTRACT FROM AUTHOR]

To enhance the accuracy of object recognition, various combination of recognition algorithms are used in recent literature. In this paper coherence of Discrete Wavelet Transform (DWT) is combined with four different algorithms: error vector of principal component analysis (PCA), eigen vector of PCA, eigen vector of Linear Discriminant Analysis (LDA) and Convolutional Neural Network (CNN) then combination of four results are done using entropy of detection probability and Fuzzy system. From this research the accuracy of recognition is found dependent on image and diversity of database. The combined method of the paper provides recognition rate of 89.56% for the worst case and 93.34% for the best case both can be said better in comparison with the previous works where individual method has been implemented on a specific set of images. [ABSTRACT FROM AUTHOR]

The Internet of Things (IoT) provides a major contribution to the innovation of smart manufacturing and industrial automation. Due to IoT, network devices and intelligent machines exchange information through different types of Internet connection and processes are predominantly automated. This reduces significantly the need for more human intervention and supports high performance. Nevertheless, the utilization of IoT in industrial automation called Industrial IoT (IIoT) has several issues, including the management of applications and IIoT devices. Moreover, heterogeneous networks and tremendous devices deployed in the IIoT environment require flexible configuration and reconfiguration according to the change for ensuring dynamic performance. We argue that Software-Defined Networking (SDN) is one of the technologies that can be used to solve some of the previously mentioned issues. In this paper, we propose a survey for the implementation of SDN solutions in IIoT and discuss the pros and cons brought about by this synergy named "SDN-IIoT". We explore the current articles on SDN-IIoT by considering different crucial domains such as flow installation techniques, fault tolerance, traffic routing optimization, resource management, energy efficiency, real-time, and network security. Furthermore, we analyze Artificial Intelligence (AI)/Machine Learning (ML) tasks to improve the performance of SDN-IIoT and the deployment of different technologies like Network Function Virtualization (NFV) and Time-Sensitive Networking (TSN) in SDN-IIoT. After observing the limitations of existing SDN-IIoT architectures, we propose an improved candidate architecture for SDN-IIoT based on a hierarchical distributed control plane. The new SDN-IIoT architecture contains AI, Industrial Backhaul Network (IBN), Dynamic Hash Table (DHT), AdaptFlow protocol, and edge/cloud storages. This paper selects the five most used SDN controllers by the literature review and identifies the features of each SDN controller. In the end, we provide open challenges and future research orientations in SDN-IIoT. We hope that this paper will be helpful for engineers, organizations, and researchers on the innovation of IIoT and SDN technologies. [ABSTRACT FROM AUTHOR]

In the business world, large companies that can achieve continuity in innovation gain a significant competitive advantage. The sensitivity of these companies to follow and monitor news sources in e-commerce, social media, and forums provides important information to businesses in the decision-making process. With the large amount of data shared in these resources, sentiment analysis can be made from people's comments about services and products, users' emotions can be extracted and important feedback can be obtained. All of this is of course possible with accurate sentiment analysis. In this study, new data sets were created for Turkish, English, and Arabic, and for the first time, comparative sentiment analysis was performed from texts in three different languages. In addition, a very comprehensive study was presented to the researchers by comparing the performances of both the pre-trained language models for Turkish, Arabic, and English, as well as the deep learning and machine learning models. Our paper will guide researchers working on sentiment analysis about which methods will be more successful in texts written in different languages, which contain different types and spelling mistakes, which factors will affect the success, and how much these factors will affect the performance. [ABSTRACT FROM AUTHOR]

Log-based anomaly detection is important for improving the reliability and availability of software systems, especially those evolving using DevOps, owing to the huge number of logs generated during continuous practices. However, DevOps practitioners typically inspect and interpret the generated data and logs manually on specific occasions as part of the troubleshooting process. This process of manual inspection is time consuming and challenging because these data and logs have grown to an unmanageable size owing to the increasing size and complexity of the systems. In this research paper, we introduce the D evOps A nomaly D etection F ramework (DADF). DADF is composed of two components that rely on Machine Learning (ML) and Artificial Intelligence (AI) to analyze the data and logs generated during DevOps practices to automatically detect anomalies. The first component is A nomaly D etection B efore P roduction (ADBP), which intends to detect anomalies in the prospective release before its operation in the production environment by adopting the Local Outlier Factor (LOF) technique on the data collected during implementation, building, testing, and deployment. The second component is A nomaly D etection A fter S taging (ADAS), which intends to detect anomalies after the operation of the released system by adopting the Vector Auto-Regression (VAR) technique on the data collected during monitoring from the system log, application log, and performance metrics (CPU and memory). We experimentally evaluated ADBP and ADAS in two different real-world industrial projects. The experimental results demonstrated that the accuracy, precision, recall, and F1-score of the ADBP component were 96%, 87.5%, 100%, and 93.3%, respectively, and the normalized Root Mean Squared Error (nRMSE) of the ADAS component was 2–19%. Hence, the results demonstrate the effectiveness of DADF in helping DevOps practitioners and researchers automatically detect anomalies throughout the lifecycle of DevOps by monitoring all DevOps' practices. [ABSTRACT FROM AUTHOR]

Cybersecurity attacks that target software have become profitable and popular targets for cybercriminals who consciously take advantage of web-based vulnerabilities and execute attacks that might jeopardize essential industry 5.0 features. Several machine learning-based techniques have been developed in the literature to identify these types of assaults. In contrast to single classifiers, ensemble methods have not been evaluated empirically. To the best of our knowledge, this work is the first empirical evaluation of both homogeneous and heterogeneous ensemble approaches compared to single classifiers for web-based attack detection in industry 5.0, utilizing two of the most realistic public web-based attack datasets. The authors divided the experiment into three main phases: In the first phase, they evaluated the performance of five well-established supervised machine learning (ML) classifiers. In the second phase, they constructed a heterogeneous ensemble of the three best-performing ML algorithms using max voting and stacking methods. In the third phase, they used four well-known homogeneous ensembles to evaluate the performance of the bagging and boosting method. The results based on the ECML/PKDD 2007 and CSIC HTTP 2010 datasets revealed that bagging, particularly Random Forest, outperformed single classifiers in terms of accuracy, precision, F-value, FPR, and area of the ROC curve with values of 99.597%, 98.274%, 99.129%, 0.523%, 100 and 99.867%, 99.867%, 99.867%, 0.267%, 100, respectively. In contrast, single classifiers performed better than boosting and stacking. However, in terms of FPR, the boosting exceeded single classifiers. Max voting is appropriate when accuracy, precision, and FPR are the primary concerns, whereas single classifiers can be employed when recall, FNR, training, and prediction times are critical elements. In terms of training time, ensemble approaches are more likely to be affected by data volume than single classifiers. The paper's findings will help security researchers and practitioners identify the most efficient learning techniques for securing web applications. [ABSTRACT FROM AUTHOR]

In machine learning, two approaches outperform traditional algorithms: ensemble learning and deep learning. The former refers to methods that integrate multiple base models in the same framework to obtain a stronger model that outperforms them. The success of an ensemble method depends on several factors, including how the baseline models are trained and how they are combined. In the literature, there are common approaches to building an ensemble model successfully applied in several domains. On the other hand, deep learning-based models have improved the predictive accuracy of machine learning across a wide range of domains. Despite the diversity of deep learning architectures and their ability to deal with complex problems and the ability to extract features automatically, the main challenge in deep learning is that it requires a lot of expertise and experience to tune the optimal hyper-parameters, which makes it a tedious and time-consuming task. Numerous recent research efforts have been made to approach ensemble learning to deep learning to overcome this challenge. Most of these efforts focus on simple ensemble methods that have some limitations. Hence, this review paper provides comprehensive reviews of the various strategies for ensemble learning, especially in the case of deep learning. Also, it explains in detail the various features or factors that influence the success of ensemble methods. In addition, it presents and accurately categorized several research efforts that used ensemble learning in a wide range of domains. [ABSTRACT FROM AUTHOR]

Phishing is a fraud attempt in which an attacker acts as a trusted person or entity to obtain sensitive information from an internet user. In this Systematic Literature Survey (SLR), different phishing detection approaches, namely Lists Based, Visual Similarity, Heuristic, Machine Learning, and Deep Learning based techniques, are studied and compared. For this purpose, several algorithms, data sets, and techniques for phishing website detection are revealed with the proposed research questions. A systematic Literature survey was conducted on 80 scientific papers published in the last five years in research journals, conferences, leading workshops, the thesis of researchers, book chapters, and from high-rank websites. The work carried out in this study is an update in the previous systematic literature surveys with more focus on the latest trends in phishing detection techniques. This study enhances readers' understanding of different types of phishing website detection techniques, the data sets used, and the comparative performance of algorithms used. Machine Learning techniques have been applied the most, i.e., 57 as per studies, according to the SLR. In addition, the survey revealed that while gathering the data sets, researchers primarily accessed two sources: 53 studies accessed the PhishTank website (53 for the phishing data set) and 29 studies used Alexa's website for downloading legitimate data sets. Also, as per the literature survey, most studies used Machine Learning techniques; 31 used Random Forest Classifier. Finally, as per different studies, Convolution Neural Network (CNN) achieved the highest Accuracy, 99.98%, for detecting phishing websites. [ABSTRACT FROM AUTHOR]

Digital forensic aims to provide an assistance for making decisions about a crime by looking at a file content which usually involves image files such as GIF, BMP, JPEG and etc. JPEG is a very popular image file format. It has less structured contents than other images which makes its recovery possible in the absence of some file system metadata. However, an essential problem of which is fragmented JPEG file intertwined with non-JPEG files and/or Bifragmented in the scan area. This paper proposes RX_myKarve as a new file carving framework for solving a number of forensic recovery problems including fragmentation. The RX_myKarve basic design includes a structure-based and content-based carving approaches. It adopts machine learning and evolutionary algorithms in its main components of identification validation and reassembling. The identification and validation techniques encompass an Extreme Learning Machine (ELM) for identifying and filtering the image data in the scan area. The reassembling technique encompasses a genetic algorithm to reconstruct the data from fragmented pieces to a complete image. The main contribution of the paper lies on the reassembling of fragmented image file clusters in the scan area. The RX_myKarve is tested and evaluated by using the Digital Forensic Research Workshop (DFRWS) 2006 and 2007 forensic challenge datasets. The results show that the RX_myKarve is able to carve and fully recover all the giving cases of the DFRWS-2006 dataset, which are 19 images, and all the relevant cases of the DFRWS-2007 dataset, which are 18 images. This improvement in file carving is mostly attributed to the novel identification and reassembling techniques. [ABSTRACT FROM AUTHOR]

In recent years, recognizing the visual focus of attention (VFoA) has attracted much attention among computer vision experts due to its various Human–Computer Interaction (HCI) or Human-Robot Interaction (HRI) applications. Although eye gaze is a potential cue to determine someone's focus of attention (FOA), it is challenging to determine FOA alone when the interacting partners are far away or the camera cannot capture high-resolution images from long-distance. Therefore, the head pose can be used as an approximation to recognize the focus of someone's attention. This paper proposes a vision-based framework to detect the FOA of humans using nine head poses consisting of four main modules: face detection and facial key-point selection (FDKPSM), head pose classification (HPCM), object localization and classification (OLCM), and focus of attention estimation (FoAEM). The FDKPSM uses the Multi-task Cascaded Neural Network (MTCNN) framework to detect head poses, and the HPCM classifies them into nine classes using the ResNet18. To estimate the FoA, the FoAEM uses a mapping Algorithm (EFoA) which integrates head poses on the focused object. Experimental results show that the proposed model outperformed other deep learning models by achieving the highest accuracy on three datasets: BIWI-M (96.97 %), Pointing'04-M (96.04 %) and HPoD 9 (98.99 %). The visual focus of the attention model gained an accuracy of 94.12 % in the multi-object scenario. [ABSTRACT FROM AUTHOR]

Label distribution learning is a novel machine learning paradigm to deal with label ambiguity, which is the generalization of the traditional single-label learning and multi-label learning paradigms. Though label distribution learning has attracted a lot of attentions recently, data sets with label distributions rather than logical labels have always been scarce and most of the existing models put emphasis mainly on the supervised learning, neglecting the utilization of unlabeled samples. In this paper, we propose a semi-supervised Label Distribution Learning model with label Correlations and data Manifold exploration (sLDLCM). On one hand, sLDLCM is a semi-supervised extension of existing label distribution learning models, which can effectively make use of both labeled and unlabeled data for better capturing the underlying data properties; on the other hand, sLDLCM jointly estimates the label distributions of unlabeled samples and the other model variables. Besides, both label correlations and local data manifold are explored in sLDLCM. Extensive experiments are conducted on six real-world data sets including two facial expression, one movie rating, two bioinformatics and one visual sentiment analysis data sets. Comparative studies demonstrate that the proposed sLDLCM model achieves better performance the state-of-the-arts in terms of five evaluation metrics. [ABSTRACT FROM AUTHOR]

High performance computing is now a major area where business and computing technologies need resilient high performance to meet business continuity and real-time needs. However, many top-level business and technology organizations are still in the process of improving high performance and traffic resiliency to ensure the availability of the system at all times. Machine learning is an important advancement of computer technology that helps in decision making by prediction and classification mechanism based on historical data. In this paper, we propose and integrate the concept of high-performance computing with artificial intelligence machine learning techniques in cloud platforms. The networking and computing performance data are used to validate, predict and classify the traffic and performance patterns and ensure system performance and continuous traffic flow resiliency decisions. The proposed integrated design approach has been analyzed on different step actions and decisions based on machine learning regression and classification models, which auto-correct the performance of the system at real run time instances. Our machine learning integrated design simulated results show its traffic resilience performs proactively 38.15% faster with respect to the failure point recovery along with 7.5% business cost savings as compared to today's existing non-machine learning based design models. [ABSTRACT FROM AUTHOR]

A user profile can be considered as an interface of one's characteristics, interests, thoughts, and preferences. This profile is dynamic as it is changed over time. Designing a user profile model that can track and accommodate these changes is important for providing personalized services on the Internet. This paper introduces a model for building a dynamic user profile based on the daily activities of the user and other users' activities who are followed by this user. The proposed model tracks the user, learns from his activities, and finally reformulates the user profile. Using two different supervised learning classification algorithms, the user's activities (tweets, retweets, etc.) are classified into a predefined list of topic categories of interest that are used to form and model the user's dynamic profile. The primary results show that the proposed model can be used to accurately describe the user, and thus, could be the default user interface for others. [ABSTRACT FROM AUTHOR]

Nowadays, machine learning and deep learning algorithms are used in recent studies as active security techniques instead of traditional ones to secure the cloud environment based on pre-trained data. In this paper, a literature review on machine and deep learning based defences against attacks and security issues in cloud computing is provided. A taxonomy of all different types of attacks and threats as per cloud security alliance (CSA) layers; and the general defences against cloud attacks is shown in this review as well as the reasons which let the traditional security techniques fail to satisfy the desired security level are discussed. Forty-two case studies are selected based on seven quality assessment standards and then, analyzed to answer seven research questions which help to protect cloud environments from various attacks, issues, and challenges. The analysis of case studies shows a description of the most common security issues in cloud; machine learning and deep learning models that are applied, datasets models, performance metrics, machine learning and deep learning based countermeasures and defences that are developed to prevent security issues. Finally, the future scope and open challenges in cloud computing security based on machine and deep learning are discussed as well. [ABSTRACT FROM AUTHOR]

Suicidal ideation detection from social media is an evolving research with great challenges. Many of the people who have the tendency to suicide share their thoughts and opinions through social media platforms. As part of many researches it is observed that the publicly available posts from social media contain valuable criteria to effectively detect individuals with suicidal thoughts. The most difficult part to prevent suicide is to detect and understand the complex risk factors and warning signs that may lead to suicide. This can be achieved by identifying the sudden changes in a user's behavior automatically. Natural language processing techniques can be used to collect behavioral and textual features from social media interactions and these features can be passed to a specially designed framework to detect anomalies in human interactions that are indicators of suicidal intentions. We can achieve fast detection of suicidal ideation using deep learning and/or machine learning based classification approaches. For such a purpose, we can employ the combination of LSTM and CNN models to detect such emotions from posts of the users. In order to improve the accuracy, some approaches like using more data for training, using attention model to improve the efficiency of existing models etc. could be done. This paper proposes a LSTM-Attention-CNN combined model to analyze social media submissions to detect any underlying suicidal intentions. During evaluations, the proposed model demonstrated an accuracy of 90.3% and an F1-score of 92.6%, which is greater than the baseline models. [ABSTRACT FROM AUTHOR]

In recent years, the availability of time series streaming information has been growing enormously. Learning from real-time data has been receiving increasingly more attention since the last decade. Online learning encounters the change in the distribution of data while extracting considerable information from data streams. Hidden data contexts, which are not known to the learning algorithms, are known as concept drift. Classifier classifies incoming instances using past training instances of the data stream. The accuracy of the classifier deteriorates because of the concept drift. The traditional classifiers are not expected to learn the patterns in a non-stationary distribution of data. For any real-time use, the classifier needs to detect the concept drift and adapts over time. In the real-time scenario, we have to deal with semi-supervised and unsupervised data, which provide no or fewer labeled data. The motivation behind this paper is to introduce a survey identified with a broad categorization of concept drift detectors with their key points, limitations, and advantages. Eventually, the article suggests research trends, research challenges, and future work. The adaptive mechanisms are also incorporated in this survey. [ABSTRACT FROM AUTHOR]

With the rise in popularity and its open system architecture, Android has become vulnerable to malicious attacks. There are several malware detection approaches available to fortify the Android operating system from such attacks. These malware detectors classify target applications based on the patterns found in the features present in the Android applications. As the analytics data continues to grow, it negatively impacts the Android defense mechanism. A large number of irrelevant features has become the performance bottleneck of the detection mechanism. This paper presents a multi-tiered feature selection model, which can discover relevant and significant features for improving the accuracy of malware detection approaches. The proposed method applies five machine learning classification techniques to the selected feature set. This work presents the Optimal Static Feature Set (OSFS) and, Most Important Features (MIFs) discovered with each machine learning approach. Rigorous testing and analysis show that Random Forest classification achieves the highest Accuracy rate of 96.28%. [ABSTRACT FROM AUTHOR]

In recent years, recognizing the visual focus of attention (VFoA) has attracted much attention among computer vision experts due to its various Human–Computer Interaction (HCI) or Human-Robot Interaction (HRI) applications. Although eye gaze is a potential cue to determine someone's focus of attention (FOA), it is challenging to determine FOA alone when the interacting partners are far away or the camera cannot capture high-resolution images from long-distance. Therefore, the head pose can be used as an approximation to recognize the focus of someone's attention. This paper proposes a vision-based framework to detect the FOA of humans using nine head poses consisting of four main modules: face detection and facial key-point selection (FDKPSM), head pose classification (HPCM), object localization and classification (OLCM), and focus of attention estimation (FoAEM). The FDKPSM uses the Multi-task Cascaded Neural Network (MTCNN) framework to detect head poses, and the HPCM classifies them into nine classes using the ResNet18. To estimate the FoA, the FoAEM uses a mapping Algorithm (EFoA) which integrates head poses on the focused object. Experimental results show that the proposed model outperformed other deep learning models by achieving the highest accuracy on three datasets: BIWI-M (96.97 %), Pointing'04-M (96.04 %) and HPoD 9 (98.99 %). The visual focus of the attention model gained an accuracy of 94.12 % in the multi-object scenario. [ABSTRACT FROM AUTHOR]

Label distribution learning is a novel machine learning paradigm to deal with label ambiguity, which is the generalization of the traditional single-label learning and multi-label learning paradigms. Though label distribution learning has attracted a lot of attentions recently, data sets with label distributions rather than logical labels have always been scarce and most of the existing models put emphasis mainly on the supervised learning, neglecting the utilization of unlabeled samples. In this paper, we propose a semi-supervised Label Distribution Learning model with label Correlations and data Manifold exploration (sLDLCM). On one hand, sLDLCM is a semi-supervised extension of existing label distribution learning models, which can effectively make use of both labeled and unlabeled data for better capturing the underlying data properties; on the other hand, sLDLCM jointly estimates the label distributions of unlabeled samples and the other model variables. Besides, both label correlations and local data manifold are explored in sLDLCM. Extensive experiments are conducted on six real-world data sets including two facial expression, one movie rating, two bioinformatics and one visual sentiment analysis data sets. Comparative studies demonstrate that the proposed sLDLCM model achieves better performance the state-of-the-arts in terms of five evaluation metrics. [ABSTRACT FROM AUTHOR]

High performance computing is now a major area where business and computing technologies need resilient high performance to meet business continuity and real-time needs. However, many top-level business and technology organizations are still in the process of improving high performance and traffic resiliency to ensure the availability of the system at all times. Machine learning is an important advancement of computer technology that helps in decision making by prediction and classification mechanism based on historical data. In this paper, we propose and integrate the concept of high-performance computing with artificial intelligence machine learning techniques in cloud platforms. The networking and computing performance data are used to validate, predict and classify the traffic and performance patterns and ensure system performance and continuous traffic flow resiliency decisions. The proposed integrated design approach has been analyzed on different step actions and decisions based on machine learning regression and classification models, which auto-correct the performance of the system at real run time instances. Our machine learning integrated design simulated results show its traffic resilience performs proactively 38.15% faster with respect to the failure point recovery along with 7.5% business cost savings as compared to today's existing non-machine learning based design models. [ABSTRACT FROM AUTHOR]

A user profile can be considered as an interface of one's characteristics, interests, thoughts, and preferences. This profile is dynamic as it is changed over time. Designing a user profile model that can track and accommodate these changes is important for providing personalized services on the Internet. This paper introduces a model for building a dynamic user profile based on the daily activities of the user and other users' activities who are followed by this user. The proposed model tracks the user, learns from his activities, and finally reformulates the user profile. Using two different supervised learning classification algorithms, the user's activities (tweets, retweets, etc.) are classified into a predefined list of topic categories of interest that are used to form and model the user's dynamic profile. The primary results show that the proposed model can be used to accurately describe the user, and thus, could be the default user interface for others. [ABSTRACT FROM AUTHOR]

Nowadays, machine learning and deep learning algorithms are used in recent studies as active security techniques instead of traditional ones to secure the cloud environment based on pre-trained data. In this paper, a literature review on machine and deep learning based defences against attacks and security issues in cloud computing is provided. A taxonomy of all different types of attacks and threats as per cloud security alliance (CSA) layers; and the general defences against cloud attacks is shown in this review as well as the reasons which let the traditional security techniques fail to satisfy the desired security level are discussed. Forty-two case studies are selected based on seven quality assessment standards and then, analyzed to answer seven research questions which help to protect cloud environments from various attacks, issues, and challenges. The analysis of case studies shows a description of the most common security issues in cloud; machine learning and deep learning models that are applied, datasets models, performance metrics, machine learning and deep learning based countermeasures and defences that are developed to prevent security issues. Finally, the future scope and open challenges in cloud computing security based on machine and deep learning are discussed as well. [ABSTRACT FROM AUTHOR]

Suicidal ideation detection from social media is an evolving research with great challenges. Many of the people who have the tendency to suicide share their thoughts and opinions through social media platforms. As part of many researches it is observed that the publicly available posts from social media contain valuable criteria to effectively detect individuals with suicidal thoughts. The most difficult part to prevent suicide is to detect and understand the complex risk factors and warning signs that may lead to suicide. This can be achieved by identifying the sudden changes in a user's behavior automatically. Natural language processing techniques can be used to collect behavioral and textual features from social media interactions and these features can be passed to a specially designed framework to detect anomalies in human interactions that are indicators of suicidal intentions. We can achieve fast detection of suicidal ideation using deep learning and/or machine learning based classification approaches. For such a purpose, we can employ the combination of LSTM and CNN models to detect such emotions from posts of the users. In order to improve the accuracy, some approaches like using more data for training, using attention model to improve the efficiency of existing models etc. could be done. This paper proposes a LSTM-Attention-CNN combined model to analyze social media submissions to detect any underlying suicidal intentions. During evaluations, the proposed model demonstrated an accuracy of 90.3% and an F1-score of 92.6%, which is greater than the baseline models. [ABSTRACT FROM AUTHOR]

In recent years, the availability of time series streaming information has been growing enormously. Learning from real-time data has been receiving increasingly more attention since the last decade. Online learning encounters the change in the distribution of data while extracting considerable information from data streams. Hidden data contexts, which are not known to the learning algorithms, are known as concept drift. Classifier classifies incoming instances using past training instances of the data stream. The accuracy of the classifier deteriorates because of the concept drift. The traditional classifiers are not expected to learn the patterns in a non-stationary distribution of data. For any real-time use, the classifier needs to detect the concept drift and adapts over time. In the real-time scenario, we have to deal with semi-supervised and unsupervised data, which provide no or fewer labeled data. The motivation behind this paper is to introduce a survey identified with a broad categorization of concept drift detectors with their key points, limitations, and advantages. Eventually, the article suggests research trends, research challenges, and future work. The adaptive mechanisms are also incorporated in this survey. [ABSTRACT FROM AUTHOR]

With the rise in popularity and its open system architecture, Android has become vulnerable to malicious attacks. There are several malware detection approaches available to fortify the Android operating system from such attacks. These malware detectors classify target applications based on the patterns found in the features present in the Android applications. As the analytics data continues to grow, it negatively impacts the Android defense mechanism. A large number of irrelevant features has become the performance bottleneck of the detection mechanism. This paper presents a multi-tiered feature selection model, which can discover relevant and significant features for improving the accuracy of malware detection approaches. The proposed method applies five machine learning classification techniques to the selected feature set. This work presents the Optimal Static Feature Set (OSFS) and, Most Important Features (MIFs) discovered with each machine learning approach. Rigorous testing and analysis show that Random Forest classification achieves the highest Accuracy rate of 96.28%. [ABSTRACT FROM AUTHOR]

Stock price prediction is a significant index which helps to achieve maximum benefit with minimum risk by increasing the decision making capability of financial investigators and investors. However, the problem of short term stock price prediction is a complex task due to its uncertainty, discontinuity, and random nonlinear nature. In this paper, a prediction model is proposed to predict random nonlinear stock market price using Intrinsic Time-Scale Decomposition (ITD), Cluster based Modified Crow Search Algorithm (CMCSA), and Optimized Extreme Learning Machine (OELM). ITD is adopted to decompose the non-stationary stock price data into some Proper-Rotation-Components (PRCs) and a residual component. ITD is used to convert non-stationary stock price data to stationary data which are simpler and steady for analysis. The CMCSA is proposed by modifying Crow Search Algorithm (CSA) with better capability to select optimal weight and biases of ELM. Thereafter, the optimized ELM is used to predict the PRCs and residual component individually which are then incorporated to predict closing price of short term stocks. The effectiveness of proposed CMCSA is tested and validated by solving benchmark problems. The experimental study indicates that the proposed ITD-CMCSA-OELM model outperforms the CMCSA-OELM, CSA-OELM, DE-OELM (Differential Evolution optimized ELM) and ANN models. [ABSTRACT FROM AUTHOR]

Over the last decade Deep learning-based models surpasses classical machine learning models in a variety of text classification tasks. The primary challenge with text classification is determining the most appropriate deep learning classifier. Numerous research initiatives incorporated ensemble learning to boost the performance, minimize errors and avoid overfitting. However, the performance of the ensemble-methods is limited by the baseline classifiers and the fusion method. The current study makes the following contributions: First, it proposes a new meta-learning ensemble method that fuses baseline deep learning models using 2-tiers of meta-classifiers. Second, it conducts several experiments on six public benchmark datasets to evaluate the performance of the proposed ensemble. For each benchmark dataset, committees of different deep baseline classifiers are trained, and their best performance is compared with the performance of the proposed ensemble. Furthermore, the paper extends the results by comparing the performance of the proposed ensemble method to other state-of-the-art ensemble methods. The findings indicate that the proposed ensemble method significantly improve the classification accuracy of the baseline deep models. Furthermore, the proposed method outperforms the state-of-art ensemble methods. Finally, using the probability distributions for each class label of the deep baseline models improves the performance of the proposed ensemble method. [ABSTRACT FROM AUTHOR]

Nearest neighbor (NN) is a simple machine learning algorithm, which is often used in classification problems. In this paper, a concept of modified nearest neighbor (MNN) is proposed to strengthen the optimization capability of artificial bee colony (ABC) algorithm. The new approach is called ABC based on modified nearest neighbor sequence (NNSABC). Firstly, MNN is used to construct solution sequences. Unlike the original roulette selection, NNSABC randomly chooses a solution from the corresponding nearest neighbor sequence to generate offspring. Then, two novel search strategies based on the modified nearest neighbor sequence are employed to build a strategy pool. In the optimization process, different search strategies are dynamically chosen from the strategy pool according to the current search status. In order to study the optimization capability of NNSABC, two benchmark sets including 22 classical problems and 28 CEC 2013 complex problems are tested. Experimental results show NNSABC obtains competitive performance when compared with twenty-three other ABCs and evolutionary algorithms. [ABSTRACT FROM AUTHOR]

Aspect-based Opinion Mining is a form of fine-grained Sentiment Analysis and it models the semasiological relationship between aspect terms and context words in a sentence. The presence of a variety of context words has a significant impact on a sentence's sentiment polarity. As a result, while designing a model, it is necessary to consider the interaction of aspects and context words. Although existing approaches have taken into account an aspect's position in a sentence, much of the research works have not explored the use of Sentiment Lexicons with the Deep Learning algorithms. In this paper, we propose a framework for an Attention-based position-aware Bidirectional Long Short-Term Memory network for Aspect-based Opinion Mining that incorporates a Sentiment Intensity Lexicon. The aspect word's pre-trained vector is adjusted to be closer to semantically and sentimentally similar nearest neighbors and further away from sentimentally dissimilar neighbors. The proposed framework calculates aspect weights by concatenating the external knowledge in the form of lexicon sentiment intensity scores with word embeddings and position information. The framework experiments on the SemEval 2014 dataset. The results of the experiments illustrate that injecting external knowledge into the Bidirectional Long Short-Term Memory network can improve classification accuracy significantly. [ABSTRACT FROM AUTHOR]

Classification of Alzheimer's disease (AD) is one of the most challenging issues for neurologists. Manual methods are time consuming and may not be accurate all the time. Since, brain is the most affected region in AD, a proper classification framework using brain images may provide more accurate results. Deep Learning (DL) is a popular representation of machine learning techniques, that emulate the functionalities of a human brain to process information and creates patterns that help in making complex decisions. The ability to absorb information, even from the unstructured and unlabeled data, makes DL one of the first choices by researchers. In this paper, some of the most popular DL models are discussed along with their implementation results for AD classification. All brain Magnetic Resonance (MR) images are acquired from the online data-set, "Alzheimer's Disease Neuroimaging Initiative (ADNI)". From the performance comparison amongst all the discussed models, it is observed that the DenseNet-121 model achieves a convincing result with an average performance rate of 88.78%. But one limitation of the DenseNet model is that it uses lots of convolutional operations that make the model computationally slower than many of the discussed models. Depth-wise convolution is a popular way to make a convolutional operation faster and better. Hence, to improve the execution time, we have proposed replacing the convolution layers in the original DenseNet-121 architecture with depth-wise convolution layers. The new architecture also improved the performance of the model with an average rate of 90.22%. [ABSTRACT FROM AUTHOR]

Wireless Body Area Networks (WBAN) can provide continuous monitoring of patients' health. Such monitoring can be a decisive factor in health and death situations. Fault management in WBANs is a key reliability component to make it socially acceptable and to overcome pertained challenges such as unpredicted faults, massive data streaming, and detection accuracy. Failures in fault detection due to hardware, software, and network issues may put human lives at risk. This paper focuses on detecting and predicting faults in sensors in the context of a WBAN. A framework is proposed to manage AI-based prediction models and fault detection using thresholds where four Machine learning techniques: Artificial Neural Networks (ANN), Deep Neural Networks (DNN), Support Vector Machines (SVM), and Decision Trees (DT), are used. The framework also provides alarm notifications, prediction model deployment, version control, and sensing node profiling. As a proof of concept, a fault management prototype is implemented and validated. The prototype classifies faults, manages automation of sensing node profiling, training, and validation of new models. The obtained experimental results show an accuracy greater than 96% for detecting faults with an inferior false alarm rate. [ABSTRACT FROM AUTHOR]

Resilience and flexibility are the means by which a network stands for an efficient service provider platform. Cognitive packet network (CPN) is such a paradigm that solely depends on the behavioral and intelligence capacity of the packets that carry valuable information within the network. Unlike conventional networks, CPN relies on the decision-making ability of the packets. In CPN, different types of packets take part in finding the optimal route from the source to destination. Further, packets dynamically adapt environmental changes by using recurrent neural network-based learning algorithms. Mapping cognitive features along with enrichment of quality of service is the main task of CPN. Due to continuous updates in the packet information vulnerable elements are found to be less prone to attack the CPN, thus resulting in a robust networking solution. In this paper we firstly present a novel taxonomic structure of the CPN. Then, we discuss the stringent routing techniques normally used in CPN. We also investigate various adaptive aware features in CPN. Next, we present intuitive emergency service and sensor network design prospects of the CPN. The article also discusses various means of cognitive aware mitigation aligned with the packets in CPN. Finally, we deliver how CPN is secure and available security aspects. [ABSTRACT FROM AUTHOR]

• A multi-voter multi-commission nearest neighbor (MVMCNN) classifier is proposed. • It clusters each class into some commissions and utilizes LMPNN to make a decision. • Evaluation is performed on 30 datasets from UCI Repository using 10-fold cross-validation. • MVMCNN is more accurate and stable than KNN, LMPNN, and BM-FKNN. Many improved versions of k -nearest neighbor (KNN) have been proposed by minimizing total distances of multi k nearest neighbors (multi-voter) in each class instead of the majority voting, such as a local mean-based pseudo nearest neighbor (LMPNN) that give a better decision. In this paper, a new KNN variant called multi-voter multi-commission nearest neighbor (MVMCNN) is proposed to examine its benefits in enhancing the LMPNN. As the name suggests, MVMCNN uses some commissions: each calculates the total distance between the given query point (test pattern) and k pseudo nearest neighbors using the LMPNN scheme. The decision class is defined by minimizing those total distances. Hence, the decision in MVMCNN is obtained more locally than LMPNN. Examination based on 10-fold cross-validation shows that the proposed multi-commission scheme can enhance the original (single-commission) LMPNN. Compared with two single-voter models: KNN and Bonferroni Mean Fuzzy k -Nearest Neighbors (BM-FKNN), the proposed MVMCNN also gives lower mean error rates as well as higher Precision, Recall, and F1 Score, indicating that the multi-voter model provides a better decision than the single-voter ones. [ABSTRACT FROM AUTHOR]

Modern cities dynamically face several challenges including digitalization, sustainability, resilience and economic development. Urban planners and designers must develop urban forms that address these challenges. With the integration of new communication and information technologies (Smartphone, GIS, Drones, IoT, Sensors, etc.), urban activities have generated large volumes of urban data. The rapid growth in terms of collection and big data storage capacities combined with the ever-increasing computational power of modern machines have made possible their efficient treatment using machine (ML) and deep learning (DL) algorithms. The emergence of such groundbreaking methods has in turn helped to address the challenges of modern-day cities in several domains (health, security, mobility, etc). ML algorithms have been proposed to model the urban form's indicators for intelligent urban planning decision making. They have been proven to perform better than the traditional methods. However, the potential of ML has not yet been fully explored in research for urban planning decision support. This paper presents a comprehensive review of ML applications for mitigating the challenges of modern cities planning. First and foremost, an overview of the urban forms, sources of urban data, the ML and DL techniques as well as their potential in solving the aforementioned challenges. For each ML method, we will highlight it working principle, advantages, disadvantages and potential applications using comparative tables. Finally, we will discuss the issues and challenges of ML methods in urban form's modeling while ultimately advocating some future research directions. [ABSTRACT FROM AUTHOR]

Artificial Intelligence (AI) has played a substantial role in the response to the challenges posed by the current pandemic. The growing interest in using AI to handle Covid-19 issues has accelerated the pace of AI research and resulted in an exponential increase in articles and review studies within a very short period of time. Hence, it is becoming challenging to explore the large corpus of academic publications dedicated to the global health crisis. Even with the presence of systematic review studies, given their number and diversity, identifying trends and research avenues beyond the pandemic should be an arduous task. We conclude therefore that after the one-year mark of the declaration of Covid-19 as a pandemic, the accumulated scientific contribution lacks two fundamental aspects: Knowledge synthesis and Future projections. In contribution to fill this void, this paper is a (i) synthesis study and (ii) foresight exercise. The synthesis study aims to provide the scholars a consolidation of findings and a knowledge synthesis through a systematic review of the reviews (umbrella review) studying AI applications against Covid-19. Following the PRISMA guidelines, we systematically searched PubMed, Scopus, and other preprint sources from 1st December 2019 to 1st June 2021 for eligible reviews. The literature search and screening process resulted in 45 included reviews. Our findings reveal patterns, relationships, and trends in the AI research community response to the pandemic. We found that in the space of few months, the research objectives of the literature have developed rapidly from identifying potential AI applications to evaluating current uses of intelligent systems. Only few reviews have adopted the meta -analysis as a study design. Moreover, a clear dominance of the medical theme and the DNN methods has been observed in the reported AI applications. Based on its constructive systematic umbrella review, this work conducts a foresight exercise that tries to envision the post-Covid-19 research landscape of the AI field. We see seven key themes of research that may be an outcome of the present crisis and which advocate a more sustainable and responsible form of intelligent systems. We set accordingly a post-pandemic research agenda articulated around these seven drivers. The results of this study can be useful for the AI research community to obtain a holistic view of the current literature and to help prioritize research needs as we are heading toward the new normal. [ABSTRACT FROM AUTHOR]

The Coronavirus disease is quickly spreading all over the world and the emergency situation is still out of control. Latest achievements of deep learning algorithms suggest the use of deep Convolutional Neural Network to implement a computer-aided diagnostic system for automatic classification of COVID-19 CT images. In this paper, we propose to employ a feature-wise attention layer in order to enhance the discriminative features obtained by convolutional networks. Moreover, the original performance of the network has been improved using the mixup data augmentation technique. This work compares the proposed attention-based model against the stacked attention networks, and traditional versus mixup data augmentation approaches. We deduced that feature-wise attention extension, while outperforming the stacked attention variants, achieves remarkable improvements over the baseline convolutional neural networks. That is, ResNet50 architecture extended with a feature-wise attention layer obtained 95.57% accuracy score, which, to best of our knowledge, fixes the state-of-the-art in the challenging COVID-CT dataset. [ABSTRACT FROM AUTHOR]

Machine learning classifiers perform well on balanced datasets. Unfortunately, a lot of the real-world data sets are naturally imbalanced. So, imbalanced classification is a serious problem in machine learning. The imbalanced class distribution misleads classifiers from correctly classifying the minor class. This paper introduces Reduced Noise-SMOTE (RN-SMOTE) for pre-processing imbalanced data. RN-SMOTE firstly, oversamples the training data using SMOTE which introduces noisy oversampled synthetic instances in the minority class. Then, applying DBSCAN to detect and remove noise. Next, the clean artificial instances are combined with the original data. Finally, RN-SMOTE applies SMOTE again to rebalance the dataset before introducing it to the underlying classifier. RN-SMOTE is evaluated using 9 different classifiers and 9 different imbalanced datasets with different imbalance ratios and five of them are used for outlier detection. The results proved that the performance of the classifiers has been improved with RN-SMOTE and outperformed the performance with original data and SMOTE with percentage based on the classifier, dataset and evaluation metric. Also, performance of RN-SMOTE has been compared to the performance of the current state of art and resulted in an increase up to 37.41%, 23.28%, 13.95% and 9.07% in terms of Recall, F1, Precision and Accuracy for RN-SMOTE. [ABSTRACT FROM AUTHOR]

This paper investigates negative sentiment tweets with the presence of hyperboles for sarcasm detection. Six thousand and six hundred pre-processed negative sentiment tweets comprising #Chinesevirus, #Kungflu, #COVID19, #Hantavirus and #Coronavirus were gathered for sarcasm detection. Five hyperbole features, namely interjection, intensifier, capital letter, punctuation mark and elongated word were analysed using three renowned machine learning algorithms, that is, Support Vector Machine, Random Forest, and Random Forest with Bagging. With the presence of hyperbolic words in the tweets in an unbiased dataset, the proposed model with elongated word achieved an accuracy and F-score of 78.74% and 71%, respectively. Intensifier was found to be the most significant hyperbole (p <.0001). Experiments and analysis conducted in this study concluded that hyperboles exist in an unbiased dataset which helps enhance the sarcasm detection as well. [ABSTRACT FROM AUTHOR]