Master’s Research Paper Presentation • Machine Learning — The Boy Who Cried Wolf: On Precision in CAN Bus Intrusion Detection

Friday, May 7, 2021 4:00 pm - 4:00 pm EDT (GMT -04:00)

Please note: This master’s research paper presentation will be given online. Please also note that the presentation time has been changed to 4:00 p.m.

Shikhar Sakhuja, Master’s candidate
David R. Cheriton School of Computer Science

Supervisor: Professor Sebastian Fischmeister

Modern connected vehicles host dozens of independent devices communicating on a shared CAN bus and controlling critical vehicle functions. With the advent of connectivity, cybersecurity of these devices has become a significant concern. Machine learning based approaches have found considerable traction for intrusion detection in CAN buses.

A key property overlooked by most of the work on intrusion detection is that vehicles are also safety-critical systems, meaning that they must be safe to use besides being secure. While most related work draws on ideas and concepts of intrusion detection from classical network intrusion detection, it neglects the reality that zero false positives are a must when wanting to act, and not just inform, on cybersecurity incidents.

False positives for cybersecurity have devastating consequences for safety-critical systems. In the best case, false positives cause alarm fatigue, which causes operators to ignore the alarms. In the worst case, false positives will lead to responses that increase the risk for the vehicle operator or road participants. In this work, we argue that intrusion detection for safety-critical systems such as automotive networks must focus on false positives instead of true positives. We demonstrate that this is achievable by creating ensembles of smaller independent networks instead of single complex networks. The ensembles further have the benefit of reduced resource consumption compared to alternative approaches.


To join this master’s research paper presentation on MS Teams, please go to https://tinyurl.com/26sc8ewp.