ECE 692 - Adversarial Learning
Adversarial learning is a new research area at the intersection of machine learning,
artificial intelligence, security, and digital forensics. The aim of this special topics
class is to introduce graduate students to the selection of fundamental adversarial
topics through focused lectures, reviews of state-of-the-art topics, and hands-on
projects. The topics covered in this class will cover general adversarial perturbations,
data, poisoning, bias, and misinformation.
The class will spend about 30% of the time on lectures and paper reviews, and 70%
on the group research projects, where each group will focus on one research area, and
will be expected to complete a research project and a research report/paper during a
semester.
Topics
- General principles of adversarial AI
- Adversarial patches – physical and digital
- Adversarial perturbations
- Black box vs. white box attacks
- Data poisoning
- Extraction attacks
- Data inference attacks
- Remediation and robustness measures
Assignments
- Quizzes - 20% of the grade
- Projects - 80% of the grade
Course Logistics
All the reading materials, assignments and discussions will be hosted on the course Canvas web page. We have access to Google Cloud Resources through the Google Credits for Education grant.