STA 250 - Theoretical Foundations of Modern AI

Reading group details

The reading group discussions will be modeled after Colin Raffel and Alec Jacobson's role-playing student seminars, also used by Aditi Raghunathan. Researchers in machine learning often attend reading groups to learn about recent research as well as to discover new problems to work on. In these discussions, we will provide additional structure whereby students adopt different "roles" that they take, taken from the following roles:

• Positive reviewer: someone who advocates for the paper to be accepted
• Negative reviewer: someone who advocates for the paper to be rejected
• Archaeologist: someone who determines where this paper sits in the context of previous and subsequent work. They must find and report on at least one older paper cited within the current paper that substantially influenced the current paper and at least one newer paper that cites this current paper. It is of particular interest if follow-up work contests a claim in the original paper.
• Academic researcher: someone who proposes potential follow-up projects, not just based on the current paper but also only possible due to the existence and success of the current paper.
• Hacker: someone who either verifies a theoretical claim with experiments or who attempts to reproduce some of the empirical results presented in the paper. Were there any difficulties in reproducing their work? How sensitive to small modifications of the hyperparameters were the results? Do your experiments suggest any potential interesting problems to work on?

Every student must participate in at least one paper presentation, which will have 1-2 students per role for 2-5 roles depending upon enrollment. Further details will be provided in due time.

Evaluation

30% of your final grade will come from the reading group. The grade will be determined as follows:

• 30% non-presenter summaries
• 60% presentation
• 10% in-class participation

The evaluation for presentations non-presenter summaries are below.

• Non-presenters, 24 hours before discussion: Submit (on Canvas/Gradescope; details TBD) responses to the following questions.

  • Summarize the paper in 1-2 paragraphs
  • Outline one part of the paper that you found to be interesting or that you were confused by.
  • Propose two discussion questions for the reading group.

• Presenters, 24 hours before discussion: Submit (on Canvas/Gradescope; details TBD) a writeup of what you plan to present in your role specified above.

  • The writeup should either be a LaTeX'd PDF document of 1-3 pages single spaced, or a PDF of 4-6 slides.
  • Your grade will be based on your writeup as well as your presentation skills.