Discussion class format

  • There will be twelve discussion classes centered on important papers in the field. Each class will begin with a short presentation by the instructor and then move into discussion.
  • Discussions use rotating student roles, adapted from a course design by Alec Jacobson and Colin Raffel.
  • Students present in small groups (typically 2–3 students sharing a role on a given day). Groups must coordinate to create a coherent mini-talk.
  • Each student will serve in a presenter role for four of the twelve classes, contributing an approximately 10-minute segment (followed by ~2 minutes of Q&A for the group).
  • For the remaining eight classes, each student completes the non-presenting assignment before class.

Presenter roles

  • Archaeologist: Explain where the paper sits in the literature. Find and report on one older paper cited by the current paper and one newer paper citing the current paper.
  • Researcher: Propose a plausible follow-up project that is possible specifically because of this paper’s ideas, results, data, code, or benchmarks. Your presentation should include answers (implicitly or explicitly) to the following simplified version of the Heilmeier Catechism:
    • What are we trying to do?
    • How is it done today, and what are the limitations?
    • What’s new in our idea, and why might it succeed?
    • If successful, what difference will it make?
    • What’s the first thing we can try to test the idea?
  • NeurIPS Reviewer: Provide a balanced, evidence-based review with at least three strengths and at least three weaknesses. Touch on the following dimensions from the NeurIPS reviewer guidelines:
    • Quality: Is the submission technically sound? Are claims well supported (e.g., by theoretical analysis or experimental results)? Are the methods appropriate? Is this a complete piece of work or work in progress? Are the authors careful and honest about strengths and weaknesses?
    • Clarity: Is the submission clearly written and well organized? Does it adequately inform the reader (enough detail for reproduction)?
    • Significance: Are the results impactful? Will others build on the ideas? Does it address a difficult task better than prior work or advance understanding in a demonstrable way? Does it provide unique data, conclusions, or approaches?
    • Originality: Is it clear how this work differs from prior contributions, with relevant citations? Does it introduce novel tasks or methods, or a well-reasoned novel combination of techniques? (Novel insights or improved efficiency/fairness can also count as originality.)

Non-presenter assignment

By 1pm on the day of the discussion, post at least one question about the paper to Ed Discussion. Your question can be something you found confusing or a topic you’d like the class to explore.

Grading

  • Presenter contributions: 4 presentations × 9 points each = 36 points.
  • Non-presenter assignments: 8 submissions × 3 points each = 24 points.
  • Total: 60 points.