Probing the limits of anomaly detectors for automobiles with a cyber attack framework
Adrian Taylor (Defence R&D Canada), Sylvain Leblanc (Royal Military College of Canada), and Nathalie Japkowicz (American University, USA)
Modern vehicles are increasingly governed and controlled by a network of computers. Automobile security requires that these networks are secure. Detecting malicious traffic on a vehicles communication network can provide an extra layer of security when all else has failed. Taylor et al test the use of recurrent neural networks to detect irregular traffic on the car network. These networks have been used to suggest the next word for language processing; they might be used to guess the most likely next letter or word while you type. In a car network they can predict the traffic that is expected to occur and consequently be used to highlight improbable network traffic as abnormal. Their testing shows that the longer and more unusual an event is, the more likely that an recurring neural network will identify the traffic as an issue. This kind of approach to providing greater vehicle security appears to have merit but there is more work to be done to better understand and optimize the technology.
Cite:
A. Taylor, S. P. Leblanc and N. Japkowicz, “Probing the Limits of Anomaly Detectors for Automobiles with a Cyber Attack Framework,” in IEEE Intelligent Systems, vol. PP, no. 99, pp. 1-1.
doi: 10.1109/MIS.2018.111145054
Source:
http://ieeexplore.ieee.org/document/8255780/