University of Guelph

Purpose

Machine Learning (ML) technologies have been widely adopted in many mission-critical fields to support intelligent decision-making with superior performance. With the success of these new technologies, the application of ML introduces novel and significant threats to AI-powered systems. Policymakers around the world have made a number of ongoing efforts on regulation enactment to enforce and normalize AI cybersecurity and privacy. It is essential to ensure that ML systems can achieve regulatory compliance and satisfy the standard requirements. This project will focus on developing a taxonomy of state-of-the-art ML offensive/defensive technologies based on a comprehensive literature review, including a collection of open-source adversarial challenges and defense utilities; devising efficient security and privacy defense mechanisms against the threats faced in the ML model training and prediction phase; and developing an empirical framework consisting of a toolset of best practices that can be leveraged to enable robust ML application development and deployment.