Guideline & Policy for Paper Presentation

The tentative course schedule lists 20 papers across four presentation sessions: October 27, November 3, November 17, and November 24, with five presentations per session. Depending on enrollment, papers will be presented individually or jointly by two students sharing one 30-minute slot. For 25 students, this gives 15 individual presentations and five paired presentations. The instructor will confirm assignments and presentation roles after enrollment is finalized; details will be released on Piazza and Canvas.

For a joint presentation, both students must read the agreed material, contribute to the slides, present a substantial part, and be prepared for questions on the whole paper. Each student is graded individually on their contribution and understanding using the same rubric below.

Read the main paper and any appendix material essential to its central claims. For a long paper or technical report, agree on a focused reading and presentation scope with the instructor in advance. That scope should cover the central method, supporting evidence, limitations, and essential appendix material. Every student is graded individually using the same rubric below.

Paper Reading

Before reading the paper, I strongly recommend you read this essay on how to read papers and even the references therein. You could imagine yourself being the reviewer of this paper. If you have never done review before, you can find some examples here and more from OpenReview. For example, you can see the public review comments after clicking any paper in ICLR 2026. Of course, since not all reviewers are doing a great job, you should do some filtering and learn from good ones.

Slides (50%)

Prepare one deck per paper and submit it on Canvas at least 24 hours before the presentation day. Joint presenters should identify their respective contributions. Use enough slides to support the allotted time, prioritizing a clear technical account over a fixed slide count. PowerPoint and LaTeX Beamer are preferred.

Slides should roughly contain the following parts:

  1. Problem Setup and Motivation (20%)
    • Explain the problem the paper addresses, why it matters, and its main technical and empirical contributions.
  2. Contributions (50%)
    • Summarize the main (both technical and empirical) contributions.
    • Explain the technical contributions, such as the methodology, implementation, and theoretical results. You do not need to present every proof or implementation detail; use a proof sketch or backup slides when helpful.
    • Explain the experimental setup, results, and analysis.
  3. Take-home messages and additional comments (30%)
    • Describe the strength and the weakness. Typical criteria include: soundness of the claims (theoretical grounding, empirical evaluation), significance and novelty of the contribution, and clarity of the writing.
    • Summarize important things that you learned.
    • Raise questions you have.
    • Propose suggestions about how to improve the work.
  4. (Optional) Backup slides for answering questions from the professor and other students.

You can use LLMs and online materials like pictures, animation, and so on. But please make sure you give credits clearly.

Presentation (50%)

Every paper has a 30-minute slot, including questions, whether presented by one or two students. Plan for 20–25 minutes presenting and the remaining 5–10 minutes for Q&A: either 20 minutes plus 10 minutes, or 25 minutes plus 5 minutes. Joint presenters divide the presentation time within this shared slot. Finish within the allotted slot.

Each presentation meeting has five slots (150 minutes), 10 minutes for transitions and overflow, and a 20-minute break after the second presentation. This fits the 4:00–7:00 p.m. meeting, with 160 minutes of teaching time.

  1. Presentation (50%)
    • You should practice the presentation to make it smooth.
  2. QAs (50%)
    • Be prepared to answer questions from the professor and other students. Use backup slides when an answer is difficult to explain verbally.

See the course policy on AI tools and attribution for requirements that also apply to slides.