Trustworthy Machine Learning

Course Description

Prepares students to understand the security and privacy problems in machine learning and educates students to propose different attack strategies to identify the vulnerabilities of a range of learning algorithms and understand different defense approaches towards trustworthy machine learning systems. Students will explore topics including basic machine learning foundations (e.g., linear regression and PCA), adversarial attacks against different learning algorithms, differential privacy, data valuation, and different categories of defenses. The lessons are reinforced via a series of topic-driven lectures, coding assignments, related paper readings, exams and in-class discussions. Students will learn to analyze current interactions between attackers and defenders on machine learning and therefore develop an understanding of the principles on trustworthy machine learning which is an emerging and important topic. Students will be required to finish three related homework projects, including 1) developing a machine learning classifier, 2) designing adversarial attacks against the built classifier, and 3) developing defenses to improve the robustness of the trained classifier against designed attacks. Students registered for 4 credit hours will also finish a final project based on the class topics, demonstrating their ability to propose related new algorithms based on the class subjects. 3 undergraduate hours. 4 graduate hours. Prerequisite: CS 225; one of CS 440, ECE 448, CS 441, CS 446 or ECE 449; one of MATH 225, MATH 257, MATH 415, MATH 416, ASRM 406 or BIOE 210.

Code

CS 442

Credits

3-4

Subject

Computer Science

Programs That Include This Course

• Bachelor's Bachelor of Science in Computer Science
• Bachelor's Bachelor of Science in Mechanical Engineering