Frequently Asked Questions

Q: I just found this website, what should I do next?

A: Please read through this FAQ and the Syllabus page. If you are registered (or waitlisted) for the course, the course staff will enroll you in the technologies we will use for communication (Piazza) and homework assignment submission (Gradescope). If it is after the first day of class, you have been registered for more than two days, and you still don’t have access to one of these, then go ahead and enroll yourself in Piazza using your Andrew Email and send a “Private Note” to the instructors that includes your Andrew ID.

Q: Why is this course listed as an “In Person Expectation Rotation” modality?

So that we can accommodate as many students who want to take this course as possible, subject only to the limitations of our course staff (i.e. how many TAs we have relative to the number of students).

The course is in “In Person Expectation Rotation” modality, meaning that each student is expected to attend N lectures in person and then is expected to watch 1 lecture online. This means that the course has an effective capacity of twice the room size. For example, GHC 4401 seats 244 students, so we can enroll up to 488 students at the cost of less frequent in-person attendance of lectures.

Q: Can I attend every lecture in-person?

A: That depends on how many students are enrolled and how many seats we have. There are two cases:

• Case 1: If our enrollment exceeds the number of seats, then you will sometimes attend in-person and sometimes attend online. For example, in F25 we have \(244 = 61 \times 4\) seats in GHC 4401, so if we have \(305 = 61 \times 5\) students enrolled, then we will divide into 5 groups of 61 students. Each group will rotate through attending 4 lectures in-person, followed by 1 lecture watched online. The rotations will be offset, so we have enough seats for the in-person groups.
• Case 2: If our enrollment drops to fewer than the number of seats (e.g. less than 244 in F25), then everyone can attend every lecture in-person every time.

Q: Will I be able to get off the waitlist?

A: No one should be on the waitlist. The course is in “In Person Expectation Rotation” modality, meaning that each student is expected to attend N lectures in person and then is expected to watch 1 lecture online. This means that the course has an effective capacity of twice the room size (e.g. GHC 4401 seats 244 students, so we can enroll up to 488 students). We predict that this is far more capacity than we will have students wanting to take the course. So everyone who wants to take the course will get in.

Note that there is a bug in the registration system that occasionally causes a short pileup on the waitlist for each section. However, someone will manually add you or invite you to register within a few business days.

Q: How will we all attend in-class quizzes in-person?

A: Simple: The group that is expected to watch the lecture online will take the quiz 2:00 - 2:15pm. Then the groups that are expected to attend lecture in-person will take the quiz 2:20 - 2:35pm.

Q: Does this course fulfill an academic requirement for me?

A: As of the start of Spring 2024, this course does fulfill an academic requirement for the Master’s in Machine Learning (MSML), the AI major, the ML Minor, and the ML Concentration. It may fulfill an academic requirement for your program, and the best way to find out is to ask your program director.

Q: Why was this course created?

A: Students interested in generative AI can already access most of the important methods driving the recent growth in the field. However, to do so they might need to take four to five courses in MLD (e.g. an undergrad could take 10-417, 10-403, 10-414, 10-405, 10-425) or a variety from LTI and RI. The purpose of this course is to provide a single course that brings all of these topics together under one roof. In doing so, we will also be able to draw ties between the different methods and how they interact.

Q: Can I learn about generative AI in other courses?

A: Certainly! Here’s a very incomplete list…

• 11-667 Large Language Models: Methods and Applications: Explores the models, optimization methods, and training regimes that are driving the current advancements in LLMs.
• 16-726 Learning-Based Image Synthesis: Covers a variety of ML techniques for image synthesis (aka. generation) and includes many of the modern models and techniques driving the field.
• 16-824 Visual Learning and Recognition: Considers an array of computer vision applications besides just generation and state-of-the-art models for them.
• 10-414/714 Deep Learning Systems: Covers aspects of how to build efficient scalable systems, such as those for LLMs.
• 10-403 Deep Reinforcement Learning: Covers reinforcement learning techniques that are important for understanding RLHF used to fine-tune LLMs for chat.
• 10-417/617 Intermediate Deep Learning: Covers the basics architectures and methods used by generative AI, such as Transformers.
• 10-405/605 Machine Learning for Large Datasets: Covers aspects of how to scale up machine learning to massive datasets and distributed learning across many machines.
• 10-425/625 Introduction to Convex Optimization: Covers modern optimization techniques used to train foundation models.