Theresia El Khoury1, Georges Badr2, Amir Hajjam El Hassani1 and Stéphane N’Guyen Van Ky3, 1Nanomedicine Lab, Université Bourgogne Franche-Comté, Rue Ernest Thierry Mieg, Belfort, France, 2TICKET Lab, Antonine University, Hadat-Baabda, Lebanon, 3Département de Recherche & Développement, STUDEC, Blagnac, France
Technical documentation written using S1000D norm in aeronautics and aerospace fields is based on ‘appicability’ tags that define the context in which a document or part of it is valid. Filtering data modules by applicability is challenging because tags can have the potential to accumulate infinitely, reaching a level of complexity that becomes challenging for human interpretation. On the other hand, many changes we made to the applicability tag's structure when upgraded from version 2.3 to version 4.1. This paper presents a general AND/OR tree-based model to filter documents based on their applicability. Transformations were made to both versions to get a unique one, thereby, evaluating the input values and selecting accessible content for each user. The model has shown accurate filtering for all tests made, on multiple datasets of different versions, having applicability of different complexity ranges.
Applicability, And-Or Trees, Knowledge-based Filtering, S1000D.
Sylivia Yangisiriza, Johnson Mwebaze and Godliver Owomugisha, Department of Networks, Makerere University, Kampala, Uganda
More and more people's daily lives are starting to incorporate elements of the Internet of Things (IoT). IoT is a network made up of devices with limited resources. Smart wearable technology is one of the IoT applications that is widely used nowadays. It is undeniable that research is being focused on IoT security due to its extensive and significant use. Because IoT networks are vulnerable to several types of attacks, it is important to identify attacks to improve IoT security. Machine learning can be a useful tool for training malware detection models. However, machine learning techniques sometimes require large amounts of memory and computing power, which motivates research on implementing energy efficient machine learning-based malware detection models on resource-constrained devices for IoT networks. In this paper, we model an energy efficient machine learning algorithm that can be used to do malware classification in IoT network. In this paper, we implement the following algorithms: Decision Tree (DT), Naïve Bayes (NB), Convolution Neuron Network (CNN) and K-Nearest Neighbour (KNN). The machine learning models were trained using the IoT-23 dataset. The IoT network was simulated using Cooja based on the contiki-2.7 operating system. Accuracy and time cost were the metrics used to determine the best model to use in energy optimization.
Internet of Things, malware detection, Machine Learning, Cooja, simulation, Energy efficiency.
Sonia Bouzidi and Imen Jdey, Faculty of Sciences and Techniques of Sidi Bouzid, Universityof Kairouan, Tunisia
With the growth of clothing recognition, recovery engines, and automatic product recommendation systems have become increasingly important for fashion-related companies. The images classification task is one of the most of these applications. In this paper, we propose a deep learning-based approach named 'CrossViTL2' by using the Vision Transformer (ViT) for clothing recognition on the Fashion-MNIST dataset. Whenever to address the overfitting issue of ViT, we incorporate L2 regularization technique into the vision architecture parameter. Additionally, we use k-fold cross-validation to ensure robustness of the model performance. We evaluate our proposed approach on a standard clothing benchmark that it achieves high accuracy in various tasks. Our experimental results demonstrate that our proposed approach achieves state-of-the-art performance on the Fashion-MNIST Benchmark for the task of image classification, with an accuracy of 93.47%, a precision value of 93.50% and 93.28% recall. We also analyse the impact of L2 regularization and k-fold cross-validation on the model architecture and performance. Our findings suggest that incorporating L2 regularization and k-fold cross-validation can significantly improve the generalization ability of our used ViT model. Overall, our work demonstrates the potential of deep learning-based approaches for fashion-related applications and highlights the importance of regularizing and evaluation techniques to ensure robustness of deep learning performance.
Image Classification, vision Transformers, Fashion-MNIST, L2 regularization &K-fold cross validation.