Research Area:  Machine Learning
An enormous amount of data processing occurs at the cloud and edge to carry out different kinds of analytics in Industrial Internet of Things (IIoT). To deal with the analytics using such a huge amount of IIoT data, different Deep Learning (DL)-based analytical models are deployed. The learning process in an analytical model needs to follow a reliable and trustworthy life cycle for crucial analysis and decision making. Similarly, considering the vulnerabilities in different levels of an IIoT network, the learning process should be reliable and trustworthy. In this paper, we propose a reliable DL-based routing attack detection scheme for IIoT. We consider adversarial training of the model for detecting intended attacks in Routing Protocol in Low-Power and Lossy Networks (RPL). This helps in achieving a reliable learning model. A Generative Adversarial Network-Classifier (GAN-C) method has been developed for attack detection events which is a two stage combination of GAN and SVM models. We evaluate the performance improvement in GAN-C with respect to a standalone Support Vector Machine (SVM) classifier. The proposed method adopts parallel learning and detection model which supports DL on computationally constrained IIoT devices. The results of the parallel model are evaluated in an IIoT network to draw a performance comparison between distributed and centralized attack detection in an RPL network. The use of the parallel GAN-C model also shows a significant reduction in training time.
Author(s) Name:  Sharmistha Nayak,Nurzaman Ahmed,Sudip Misra
Journal name:  Ad Hoc Networks
Publisher name:  Elsevier
Volume Information:   Volume 123, 1 December 2021, 102661
Paper Link:   https://www.sciencedirect.com/science/article/abs/pii/S1570870521001748