Adversarial Attacks and Defences for In-Vehicle Networks
About this project
A lot of recent automobiles have adopted AI techniques to provide autonomous and safe driving. However, there are cybersecurity attacks on intelligent vehicles and their infrastructure and they have become a big risk to the automobile industry and customers.
The main goal of this research project is to develop adversarial attacks and defences for in-vehicle networks. Machine Learning (ML) techniques (including deep learning techniques) have been applied to intrusion detection for in-vehicle networks. The main function of ML-based intrusion detection is to monitor, identify cyber attacks to in-vehicle networks, in particular, CAN bus.
Adversarial attacks are critical security threats against deployed ML- based intrusion detection for in-vehicle networks. In this research, existing adversarial attacks will be tested against offline and deployed ML-based intrusion detection for in-vehicle networks and also novel adversarial attacks will be developed to evaluate the robustness of ML-based intrusion detection. In addition, defences against adversarial attacks will be developed against adversarial attacks and their effectiveness will be assessed via multiple metrics.
Literature survey on cyber attacks and defences on ML-based intrusion detection for in-vehicle networks.
Novel adversarial attacks to ML-based intrusion detection for in-vehicle networks.
Defences against adversarial attacks and their evaluation in terms of robustness and interpretation.
Information for applicants
Strong background in computer science/IT/software engineering/other related discipline.