DAD: A Distributed Anomaly Detection framework for future In-vehicle network

Future in-vehicle (autonomous vehicles) network architectures will consider many aspects of modern network security by design. The general system contains many sub-systems related to different tasks with specific functional priorities and dedicated security mechanisms. In this work, we propose a Distributed Anomaly Detection (DAD) Intrusion Detection System (IDS) using a deep learning model that fits the in-vehicle network architecture. DAD aims to model the complex correlations among different views (sub-systems) by harnessing the joint distribution of the different sources of CAN (Controller Area Network) data. To this end, we propose DAD by jointly learning an anomaly detection model for critical applications such as security and maintenance while adopting the same isolation constraint on the sub-systems. On top of that, we introduce a new optimisation scheme that lowers both the computational inference time and the IDS's communication overhead.