Paper-Reading Presentation Guidelines

Objectives.

1. Read and present a research paper.

2. Upload a recorded video presentation that is at least 15 minutes.

Teams and Papers. The instructor will provide a list of papers. Each paper will be presented

by a group of 1-2 students. All students in the same team will receive the same grade for the

paper-reading presentation.

With the instructor’s approval, you can present a research paper that is not on the provided list

(e.g., a paper related to robustness from recent ICLR, ICML, or NeurIPS).

Submission Instructions.

• Submit a publicly accessible link to your recorded presentation in a Google spreadsheet.

• The length of your video presentation should be at least 15 minutes per student.

• All recorded presentations will be shared with the class and possibly future students.

Important Dates.

• Sep 30: Decide which paper to present.

• Oct 21: Submit your presentation video.

• Nov 4: Watch and rate all presentations (optional).

(All deadlines are 11:59pm on the due date.)

Grading. Paper-readingpresentationswillbegradedoutof40points(withupto9bonuspoints):

• Meeting the deadlines (10 points):

– (5 points) Sign up for teams and papers by Sep 30.

– (5 points) Submit your recorded video presentation by Oct 21.

• Recorded Presentation (30 points): We will run a lightweight peer-review process to grade

your presentation. We will ask each student to watch and rate all presentations.

The following criteria are provided as guidelines. You can discuss/post your criteria on Ed.

1

– Explain why the topic and the problem being discussed are important.

– Provide a roadmap outlining what will be covered in your presentation.

– Define the problem clearly, along with any notations and terminology.

– State the main results of the paper and the necessary assumptions.

– Emphasize the most significant contributions and insights of the paper.

– Discuss background, previous work, and the current state of the field.

– Provideintuitionorproofsketchesforthekeytheorems/lemmas/algorithmsinthepaper.

– Make sure that your presentation does not have factual errors.

– Speak at a reasonable pace. Avoid reading from the slides.

– Include key points on slides. Avoid overly dense slides.

– Use figures and examples to make the theorems and proofs more accessible.

– Help audience unfamiliar with the field understand the main points of the paper.

• Peer review (up to 9 bonus points):

– Eachstudentcansubmitascore(21,24,27,or30points)ordeclareconflictsofinterestfor

each presentation. The instructor will participate as a reviewer (with twice the weight).

– Scores will be normalized to have the same mean and variance across reviewers. The

final score of a presentation is the average of the normalized scores of all reviewers.

– (Top projects) The top 20% of presentations will receive 3 bonus points. The top 10%

of presentations will receive 3 additional bonus points (so 6 points in total).

– (Top reviewers) If your evaluation is among the top 20% most accurate evaluations

(measured by ℓ -norm of the difference), you will receive 3 bonus points.

2

2