CS 680A - AI Security

Credits: 3

The objective of the course would be to develop an understanding of modern security problems and defensive solutions in emerging AI techniques. The first part of the course will give the student a comprehensive overview of the state-of-art attack surface and algorithms, including but not limited to adversarial input attacks, weight attacks, Trojan/backdoor attacks, model inversion attacks, model extraction attacks and remote side-channel attacks on ML models. In the second part, the material will cover the SOTA defensive measures, protective learning techniques, and attack mitigation strategies. Prerequisites: Introduction to Machine Learning (CS 436/CS 536), Introduction to Deep Learning (CS 480E/CS 580E) and Basic Math Courses: Linear Algebra, Calculus, and Statistics. Expected to be offered at least once every two years.