Table of Contents
Course Description
Deep learning models can be considered as programs with semantics. In this course, we will study program analysis techniques and their applications in addressing deep learning problems, including model debugging, model attack and defense, model testing and verification, and automating model training using compiler based techniques.
Topics
Program analysis; Deep learning security, testing, debugging, and verification
Grading
- Project (3 small and 1 term projects): 70%
- 3 small: 30%
- Term project: 40%
- Presentation: 15%
- Paper presentation 10%
- Final presentation 5%
- Quiz: 5%
- Midterm: 10%
Deadlines
- Discussion team formation and topic selection (9/9, send to TA)
- Three small projects (10/16, send to TA)
- Term project proposal (10/7, send to both instructor and TA)
- Term project and project report (12/7, last day of the semester, send to both instructor and TA)