Assignments

Homeworks

These are intended to build your intuition and experience in implementing real inference algorithms on toy problems.

• HW1: Gaussian Processes
• HW2: Bayesian Neural Nets + Hamiltonian Monte Carlo
• HW3: Bayesian Neural Nets + Variational Inference
• HW4: Variational Autoencoders

In-Class Discussion

At twice-a-week meetings, we will discuss a specific paper or set of papers, as defined by the course schedule. All students should expect to be active participants in the discussion and have read relevant papers before attending class.

For the first month, the instructor will present material in each class, with some hands-on activities and class discussion.

Starting in mid-October, the format will transition to one where students lead most classwork. Each student will be a discussion leader for exactly one class.

** Template for student-run classes **

We intend these meetings to have the following template from the start of class until we release 75 min later:

• 00-05 min: Announcements & short overview of the paper by your instructor
• • What is the big idea? Why is this exciting?
• • How does this connect to the broader ideas of the course?
• 05-25 min: Short presentation by that day's student discussion leaders
• • Presenters can use any format (whiteboard, slides, etc)
• • The presentation should cover at least one of each:
• • • key technical contribution (model, algorithm, etc.)
• • • key experimental result
• • • limitation or weakness of the method
• • • limitation or weakness of the experimental design
• 25-75: Group work time
• • Student teams can work on their projects
• • Instructor will be available for detailed help / debugging

Discussion leader responsibilities:

• • Work cooperatively with co-leaders as a team
• • Read the paper well in advance of the assigned date
• • Meet with instructor as a team during office hours (or by appointment scheduled via email) at least 1 day before the scheduled class to discuss strategy
• • Prepare ~20 min presentation of key ideas
• • Prepare questions for students

Critical Comments on Readings

Students are expected to regularly engage in detailed critical responses to readings.

For 5 readings in Units 1-2, and for 4 readings in Units 3-4, each student is expected to post comments on Piazza before class starts.

• When posting your comments, please reply to the thread created by the instructor for that particular reading, usually named something like "Thu 09/26 Reading response"
• You should AVOID reading other comments before posting your own. Do your own work!
• After you post, feel free to engage in responding to other students

Please show that you've read the paper critically by providing the following:

• Pros: 2 or more full sentences. What is the most exciting or interesting idea or method described here? Why is it important? Don't just copy the abstract - what do you think?
• Cons: 2 or more full sentences. What is the biggest weakness of this method, model, or algorithm? Problems may include weak empirical validation, missing theory, unacknowledged assumptions, applicability to a narrow range of problems, or more.
• Questions 1 or more full sentences. What questions do you have after reading this paper?

These comments will be evaluated on a "sufficient"/"insufficient" binary grading basis. If your comments show low-effort or poor understanding, it will not count toward your required number of comments.

Open-Ended Project

You will spend about 2 months on a self-designed research project. To keep you on track, we have established several specific checkpoints at which you will turn in preliminary reports that build on each other and will eventually evolve into a final report and final presentation.

• Initial Pitch Presentation
• Checkpoint 1
• Checkpoint 2
• Final Presentation
• Final Report