Welcome to 15-780, Graduate Artificial Intelligence, for the Spring 2024 semester. The material covered in this course will be quite different from that taught in previous semesters (see e.g., here for a previous class page), which traditionally focuses on presenting a broad overview of many topics in AI. But in recent years this content has diverged fairly substantially from the colloquial definition of AI, which these days centers around large language models, generative AI, and related topics.
With that background, this semester is going to be an experiment. We are going to cover, assuming very little background knowledge, the material needed to understand how modern AI methods work. This is not a value judgement on the content of "traditional" AI courses (well, maybe a little), but mostly a trial of something new. Because as much as I really do like classical AI material, I think the content here is likely to be even more important for most graduate students in CS (and beyond). The material we cover will include (though not be limited to):
The main content of the course will be covered in lectures. Attendance at lectures is required (there will be no remote option, and if you need to miss a class this should be coordinated with the instructor). Course notes from lectures will be made available over the course of the semester, but there is no guarantee they will be posted e.g., prior to homeworks being due. There is no official textbook for the course other than the course notes, but we will post optional additional references as needed.
The course will have an in-person written midterm on Feburary 28. The midterm will cover all material from class lectures up until this point. The midterm is open notes, but you cannot use a laptop or any internet-connected device.
Students will complete a final project for the course in groups of 2-3 students. The final project is quite open-ended, and should consist of building an implementation, application, or analysis of some modern model in AI. There is no final exam in the course.
Important information:
Grades for the course will be assigned according to: 25% homework, 30% midterm, 35% final project, 10% class participation.
This is a course on AI. So yes, you can use whatever AI tools you want to help with the assignments or coding. This policy applies to this class only shouldn't be interpreted as implying anything for any other course. You are allowed and encouraged to collaborate with non-AI (i.e. human) colleagues on all the homeworks as well.
Homeworks will be assigned weekly in class, and will be due the following week in class. These will include written assignments and coding, and will be checked lightly for completeness. Your homework grade will mainly consist of whether you consistently attempt all the assignments, rather than a thorough grading of each problem.
Day | Time | Location | TA |
---|---|---|---|
Monday | 1:00pm - 2:00pm | Baker Hall 136E | Lingjing |
Tuesday | 5:00pm - 6:00pm | Wean Hall 5316 | Mingjie |
Wednesday | 1:00pm - 1:50pm | Porter Hall 125C | Lingjing |
Thursday | 2:00pm - 3:00pm | NSH 4201 | Mingjie |
A list of lectures and associated notes will be posted here some indeterminate time after each class. I'm experimenting with a system that generates the notes semi-automatically from lecture videos using 1) a whisper-based transcription of the lecture audio, combined with 2) a GPT4 (Vision) prompt that takes the transcript, a sequence of snapshots of the vidoes corresponding to that time, and generates a set of notes for the course. I then lightly edit the output. The result is about as you'd expect: rather disjointed and written in somewhat poor prose, but at the same time absolutely absurd to imagine as of a few years ago.