Hierarchical ontologies play a key role in organizing documents in a repository. While matching the ontologies, the relationships among the concepts are considered to be a major aspect. In hierarchical ontologies, the concepts are associated with one another only through the “is-a” relation. In this paper, we discuss an approach for matching heterogeneous hierarchical ontologies that are related to the same domain through the semantic interpretation and implicit context of the concepts. We have designed rules that can handle heterogeneities and inconsistencies that are found in hierarchical ontologies. These rules can be embedded to complement the existing matching systems, to resolve the matching complexities in the hierarchical ontologies
20 June 2022 in IEEE Transactions on Intelligent Transportation Systems
Internet of Things (IoTs) is envisaged to widely capture the realm of logistics and transportation services in future. The applications of ubiquitous IoTs have been extended to Maritime Transportation Systems (MTS) that spawned increasing security threats; posing serious fiscal concerns to stakeholders involved. Among these threats, Distributed Denial of Service Attack (DDoS) is ranked very high that can wreak havoc on IoT artifacts of the MTS networks. Timely and effective detection of such attacks is imperative for necessary mitigation. Conventional approaches exploit entropy of attributes in network traffic for detecting DDoS attacks. However, the majority of these approaches are static in nature and consider only a few network traffic parameters, limiting the number of DDoS attack detection to a few types and intensities. In current research, a novel framework named “Dual Stack Machine Learning (S2ML)” has been proposed to calculate distinct entropy-based varying 10-Tuple (T) features from network traffic features, three window sizes and associated Rate of Exponent Separation (RES). These features have been exploited for developing an intelligent model over MTS-IoT datasets to successfully detect multiple types of DDoS attacks in MTS. S2ML is an efficient framework that overcomes the shortcomings of prevalent DDoS detection approaches, as evident from the comparison with Multi-layer Perceptron (MLP), Alternating Decision Tree (ADT) and Simple Logistic Regression (SLR) over different evaluation metrics (Confusion metrics, ROCs). The proposed S2ML technique outperforms prevalent ones with 1.5% better results compared to asserted approaches on distribution of normal/attack traffic. We look forward to enhancing the model performance through dynamic windowing, measuring packet drop rates and infrastructure of Software Defined Networks (SDNs).
Multiple Input and Multiple Output is a technology through which data are transmitted over the channel through multiple antennas. However, during its deployment and implementation, some pragmatic issues arise such as interference, multipath fading and noise leading to potential packet losses and consume substantial energy. To address such issues, Hybrid ARQ transmissions provide effective means for error correction, especially in a noisy wireless channel. More often few bits in packets are found to be in error and it is unnecessary to use the entire MIMO channel for retransmission to correct these errors. Hence, this paper proposes a novel approach, i.e., Shared Hybrid ARQ (SHARQ IR) using the piggyback technique in an overloaded MIMO system where the transmitting antennas are more than the receiving antennas. Additionally, the concept of a simple retransmission method is used to transform an overloaded MIMO system into either the critically loaded or under loaded MIMO systems. Simulation results outperform the contemporary approaches through reduced BER and 20% throughput gain. Such improvements have the potential to support energy-efficient transmissions for encouraging green IoT applications.
16 January 2025 in The Asian Bulletin of Big Data Management
Edge-based applications are envisaged to have a paramount impact on the academic landscape. One of the key aspects of academic activities is to keep the learners engaged. The pertinent engagement activities require constant analysis of the learner’s emotions and expressions. Both aspects of comprehending the learner’s engagement greatly rely on facial coding/recognition systems. Facial recognition widely depends upon the component analysis of facial action units which are structured sets of semantic features. This research proposes a lightweight machine learning-based framework for edge-driven gadgets with five major modules and 68 recognized points of interest in each image of 360x360 dimensions. It exploits Support Vector Machine (SVM) and Convolutional Neural Networks (CNN) to recognize the audience, capture expressions and semantically label emotions. Hence, a real-time adaptive feedback-based experience can be enhanced through dynamic interaction for improvising the outcomes of learning activities. The Extended Cohn-Kanade (CK+) dataset has been employed through an extraction function that classifies several learner emotions. The expressions and emotions of the audience during interaction were taken as input to the proposed model and maintained in the dataset along with the output. This feedback identified the spectators’ requirement for better interaction and more focus on engaging the learners. An overall accuracy of 98% was achieved in correctly predicting the learner's emotions by the proposed approach, as highlighted through standard machine learning metrics. The potential future directions are deep-learning-based gesture analysis, improving the precision for spectator feedback and efficiently catering for the computational requirements.
22 December 2020 in Programming and Computer Software
Ontology based activity learning models play a vital role in diverse fields of Internet of Things (IoT) such as smart homes, smart hospitals or smart communities etc. The prevalent challenges with ontological models are their static nature and inability of self-evolution. The models cannot be completed at once and smart home inhabitants cannot be restricted to limit their activities. Also, inhabitants are not predictable in nature and may perform “Activities of Daily Life (ADL)” not listed in ontological model. This gives rise to the need of developing an integrated framework based on unified conceptual backbone (i.e. activity ontologies), addressing the lifecycle of activity recognition and producing behavioral models according to inhabitant’s routine. In this paper, an ontology evolution process has been proposed that learns particular activities from existing set of activities in daily life (ADL). It learns new activities that have not been identified by the recognition model, adds new properties with existing activities and learns inhabitant’s newest behavior of performing activities through Artificial Neural Network (ANN). The better degree of true positivity is evidence of activity recognition with detection of noisy sensor data. Effectiveness of proposed approach is evident from improved rate of activity learning, activity detection and ontology evolution.
Consumer Electronics (CE) devices (such as smartphones, tablets, laptops, and cameras) equip learners with educational ubiquity, eliminating barriers of cost, time, and space. The features of emotion recognition and analysis in CE devices have the potential to effectively engage learners through personalized and adaptive recommen- dations of learning contents. Engagement activities triggered by learner’s emotions and expressions rely on the semantic features of facial coding units. Current research integrates Deep-Emotion-AI for learner’s behavior analysis and semantic labeling of emotions. Hence, the system can dynamically gain deeper insights into learners’ emo- tional states and cognitive responses to maximize the learner’s engagement based on real-time streams of facial visuals. Vision-based CE gadgets were employed with five major modules and 68 recognized points of interest in each image of 360x360 dimen- sions. Deep-Emotion-AI utilizes a hybrid of Support Vector Machine (SVM) and Con- volutional Neural Networks (CNN) to recognize learners, capture their expressions, and semantically label learner’s emotions (annotated through Emotions Ontology) over Extended Cohn-Kanade (CK+), MMI and Google datasets. The comparison of CK+ dataset with Google-dataset and MMI highlighted better performance of the proposed model with CK+. An overall accuracy of 98 % was achieved in evaluating semantics and correctly labeling the learner’s emotions. Proposed research has established that the current application of CE with fusion of semantics, deep-emotion-AI and selected feature set can significantly captivate learner’s engagement. The potential future di- rections are deep-learning-based gesture analysis, reinforcing the content precision on learner’s feedback, and efficiently catering for the computational requirements of a larger audience.