A Unified Hardware / Algorithm Approach to Secured Machine Learning for Cyber-Physical Systems: Applications in Autonomous Vehicles and Connected Autonomous Vehicles Networks

The advancement of artificial intelligence (AI) systems has enabled development systems such as autonomous vehicles (AVs). However, like any other technology, AI systems suffer from security vulnerabilities, and they can be easily fooled by a smart adversary. Malicious attacks on AI systems in safety-critical system such as AVs can be life-threatening or result in financial harms. Unfortunately, the research on defensive methods against adversarial attacks on AI systems is at its infancy, and there is a lack of proper understanding of the inherent security vulnerabilities in these systems. In this research, we use a unified hardware / software approach to develop secure AI systems for AVs and connected autonomous vehicles. We develop specialized hardware architectures and algorithms, so that mitigation algorithms can be performed faster and more efficiently. We believe that improving the security through a unified software / hardware approach is essential in enabling the use of machine learning in safety-critical systems.

Faculty Supervisor:

Deepa Kundur

Student:

Mohammadmehdi Ataei

Partner:

Cognitive Systems

Discipline:

Engineering - computer / electrical

Sector:

Professional, scientific and technical services

University:

University of Toronto