Understanding the black-box: towards interpretable and reliable deep learning models.

Deep learning (DL) has revolutionized the field of artificial intelligence by providing sophisticated models across a diverse range of applications, from image and speech recognition to natural language processing and autonomous driving. However, deep learning models are typically black-box models where the reason for predictions is unknown. Consequently, the reliability of the model becomes questionable in many circumstances. Explainable AI (XAI) plays an important role in improving the transparency and interpretability of the model thereby making it more reliable for real-time deployment. To investigate the reliability and truthfulness of DL models, this research develops image classification models using transfer learning mechanism and validates the results using XAI technique. Thus, the contribution of this research is twofold, we employ three pre-trained models VGG16, MobileNetV2 and ResNet50 using multiple transfer learning techniques for a fruit classification task consisting of 131 classes. Next, we inspect the reliability of models, based on these pre-trained networks, by utilizing Local Interpretable Model-Agnostic Explanations, the LIME, a popular XAI technique that generates explanations for the predictions. Experimental results reveal that transfer learning provides optimized results of around 98% accuracy. The classification of the models is validated on different instances using LIME and it was observed that each model predictions are interpretable and understandable as they are based on pertinent image features that are relevant to particular classes. We believe that this research gives an insight for determining how an interpretation can be drawn from a complex AI model such that its accountability and trustworthiness can be increased.

Network intrusion detection using oversampling technique and machine learning algorithms.

The expeditious growth of the World Wide Web and the rampant flow of network traffic have resulted in a continuous increase of network security threats. Cyber attackers seek to exploit vulnerabilities in network architecture to steal valuable information or disrupt computer resources. Network Intrusion Detection System (NIDS) is used to effectively detect various attacks, thus providing timely protection to network resources from these attacks. To implement NIDS, a stream of supervised and unsupervised machine learning approaches is applied to detect irregularities in network traffic and to address network security issues. Such NIDSs are trained using various datasets that include attack traces. However, due to the advancement in modern-day attacks, these systems are unable to detect the emerging threats. Therefore, NIDS needs to be trained and developed with a modern comprehensive dataset which contains contemporary common and attack activities. This paper presents a framework in which different machine learning classification schemes are employed to detect various types of network attack categories. Five machine learning algorithms: Random Forest, Decision Tree, Logistic Regression, K-Nearest Neighbors and Artificial Neural Networks, are used for attack detection. This study uses a dataset published by the University of New South Wales (UNSW-NB15), a relatively new dataset that contains a large amount of network traffic data with nine categories of network attacks. The results show that the classification models achieved the highest accuracy of 89.29% by applying the Random Forest algorithm. Further improvement in the accuracy of classification models is observed when Synthetic Minority Oversampling Technique (SMOTE) is applied to address the class imbalance problem. After applying the SMOTE, the Random Forest classifier showed an accuracy of 95.1% with 24 selected features from the Principal Component Analysis method.

Secure and failure hybrid delay enabled a lightweight RPC and SHDS schemes in Industry 4.0 aware IIoHT enabled fog computing.

These days, the Industrial Internet of Healthcare Things (IIT) enabled applications have been growing progressively in practice. These applications are ubiquitous and run onto the different computing nodes for healthcare goals. The applications have these tasks such as online healthcare monitoring, live heartbeat streaming, and blood pressure monitoring and need a lot of resources for execution. In IIoHT, remote procedure call (RPC) mechanism-based applications have been widely designed with the network and computational delay constraints to run healthcare applications. However, there are many requirements of IIoHT applications such as security, network and computation, and failure efficient RPC with optimizing the quality of services of applications. In this study, the work devised the lightweight RPC mechanism for IIoHT applications and considered the hybrid constraints in the system. The study suggests the secure hybrid delay scheme (SHDS), which schedules all healthcare workloads under their deadlines. For the scheduling problem, the study formulated this problem based on linear integer programming, where all constraints are integer, as shown in the mathematical model. Simulation results show that the proposed SHDS scheme and lightweight RPC outperformed the hybrid for IIoHT applications and minimized 50% delays compared to existing RPC and their schemes.