We will hold about three sessions to discuss influential papers related to the course (two papers per session). These discussions will follow a role-based discussion format inspired by Colin Raffel and Alec Jacobson’s role-playing student seminars, which is also used in Aditi Raghunathan’s course. Reading groups are common in machine learning research community for keeping up with new research and discovering fresh research directions.
For each session, students will adopt specific “roles”:
Positive Reviewer: Act as though the paper is under review at a top peer-reviewed conference/journal, and you advocate for acceptance.
Negative Reviewer: Act as though the paper is under review at a top peer-reviewed conference/journal, and you advocate for rejection.
Archaeologist: Place the paper in the context of earlier and later work. Identify and report at least one older paper that substantially influenced the current one (and is cited within), and one newer paper that cites the current paper (it would be of particular interest if the follow-up work challenges the original claim).
Academic Researcher: Propose potential follow-up projects that are not just based on the current paper but also only possible thanks to the current paper’s breakthroughs or success.
Hacker: Either verify a theoretical claim through experiments or attempt to reproduce the paper’s empirical results. Note any challenges in reproducing the work (e.g., sensitivity to hyperparameters), and point out potential new research questions suggested by your experiments.
The reading group component will constitute 25% of your final grade. The breakdown is:
The evaluation for presentations non-presenter summaries are below.
Submit the following on Canvas by noon the day before the discussion (about 26 hours prior):
Your grade will be based on both the quality of this written submissions and your presentation.
Submit the following on Canvas by noon the day before the discussion (about 26 hours prior):