[20%] Written Paper Critiques

[Due] Before each lecture, on Canvas
No late submissions. A missed critique is a zero.

You will write a critique on one of the two (Required) papers for that day. Your choice, no need to tell me in advance. (Optional) papers do not count. Note that this is not light pleasure reading; once you are used to it, expect 2-3 hours per paper.

Your critique MUST have the following fields:

• Summary of the paper (a paragraph)
• Contributions (2-3 bullets, and detailed justifications)
• Strengths and weaknesses (2-3 bullets each, and detailed justifications)
• Your opinions (on the strengths and weaknesses)

Please make your best effort to write concise and precise statements. For example:

• Summarize, don't describe. "This paper proposed a X..." is useless to you later; "the key idea is XYZ" is not.
• Extract the essence. "The work is well-evaluated..." is not a contribution. It is the authors' job. Put these under "opinions."
• Push on the threat model. Weak agent-security papers are usually weak because the threat model was chosen to make the result come out.

I look for evidence that you thought about the problem, not that you read it. Do not accept the authors' claims as-is—it is their job to convince you!

Here is a sample critique of mine for reference [example].

Grading Policy

Each critique is scored out of 2 points; your total over the term is scaled to the 20% critiques are worth.

• 0 points: No submission before the deadline / a simple paraphrase (or AI-slobs) of the abstract and introduction.
• 1 point: You made an effort to understand the paper and thought carefully about its pros and cons.
• 2 points: You make an insightful comment.

Submission Instructions

Upload your critique to Canvas as a PDF.

[15%] In-Class Paper Discussion

[Note on Paper Selection]

• Choose one paper relevant to the topic covered on the day you are leading (see the syllabus).
• The paper must be from one of the following venues:
  • Security: IEEE S&P, USENIX Security, ACM CCS, NDSS
  • ML: NeurIPS, ICML, ICLR
  • NLP: ACL, EMNLP, NAACL
  • Systems / HCI: OSDI, SOSP, CHI, SOUPS
• It must NOT already be on the syllabus, required or optional. Find it yourself—that is part of the exercise.

[Note on Sign-up]
Opens 09/23, closes 09/30 on the course Google Sheet (link on Canvas; use your ONID).
If you are not signed up by 09/30 you score 0 for this component, so do it in the first week.

• The first two lectures (09/23, 09/28) fall before sign-up closes, so they are instructor-led.
• Up to 2 studentsper day, first come first served.
• You MUST meet with Sanghyun once, 3–4 days before your session. Bring draft slides and 2–3 discussion questions.
• Your slot is 25–30 min: 20 min presenting the paper as if you were the author, then 10–15 min leading the discussion.
• Do not spend your slot summarizing; assume the room has read it. The common failure mode is a 25-min summary and a 5-min silence.

Grading Policy

• 0%: No sign-up, or no presentation.
• 7%: You signed up and presented, but the presentation or discussion lacked quality.
• 12%: You presented a paper well and led a real discussion.
• 15%: Outstanding—the room argued about something substantive and left with a question it did not have before.
• Miss the meeting with Sanghyun: your score for this component drops by 3%.