Paper Reviews
You will need to submit one paper review per class based on the assigned readings. This helps ensure you do the readings and are primed for in-class discussion.
Note that anything you write in the reviews may be anonymously brought up in class at some point or another. Therefore, treat the review as if others (in a conference setting, the paper authors or even the general public) are going to see it.
Review Guidelines
There are many resources for how to write good reviews (some provided below). I will not enforce any particular format but can recommend including the following components:
- Problem and Motivation
Think carefully about why this work exists. In no more than a few sentences, express:
- What problem(s) does the paper address? Why does this matter? What is their importance?
- How is the problem motivated (e.g., by a real-world problem, theoretical predictions)?
- Who cares about and benefits from the problem(s) being solved? How?
- What kind of impact can solving this problem have?
- Paper summary
- What research questions do the authors raise?
- What are the paper's key ideas, claims, insights, and contributions? Be brief but specific.
- How does it advance the state-of-the-art?
- Strengths and weaknesses
Think critically about how well the paper does its job and provide 3-5 concrete bullet points for each. Try to avoid making vague or subjective criticisms. Some prompts to get you thinking:
- Does the paper address the problem? How well?
- How well does the paper substantiate its claims?
- How relevant, applicable, and extensible is the work?
- How reproducible are the paper's findings?
- Could its ideas be communicated better?
- Does the paper open up new topics to study?
- Are there any flaws that you observe in the work?
- New ideas and impact
What are your personal thoughts on the paper? Some prompts to get you thinking:
- How is the paper connected to other research or papers you know?
- What did you learn from the paper?
- What are your most important takeaways?
- What did you like/dislike the most? Why?
- How might you approach things differently?
- How might you improve or extend the paper?
- Do you think it is likely to have impact? Has it already done so?
- Other comments
Any other questions (e.g., things you didn't understand), comments, or discussion points you might like to raise. These are helpful for me to plan future lectures.
Format and Submission
- Feel free to use your favorite text editor and submit files in plain text or PDF.
- Please limit reviews to 1-1.5 pages long (no more than ~800 words).
- Submit reviews online to Canvas.
Grading
Reviews will be graded based on how well they demonstrate your critical thinking. Although we are not reviewing for the benefit of the authors, practice in doing so is an important component of this graduate seminar.
I will score papers on a 4-point system with 1 point of possible extra credit:
- 5: Exceptional review providing deep original insight into the work or related works.
- 4: Review shows a clear understanding of the paper's contributions and its place in the broader context of the literature and course topics.
- 3: Some effort made to understand the paper's ideas and interpret them using the student's own thoughts and words.
- 2: Wordy submission without depth of understanding. Minor effort to interpret the paper
- 1: Terse and/or shallow review; minimum-viable submission.
- 0: Late or missing submission.
Note: You may skip 3 reviews throughout the semester without penalty. This is to give you flexibility (e.g., when you have an upcoming deadline and need to suddenly drop everything).
External Resources and Examples
- Advice on reviewing research papers
- Smith, A.J., "The Task of the Referee," Computer, 1990. [HTML version]
- Hill and McKinley, "Notes on Constructive and Positive Reviewing."
- Cormode, G., "How NOT to review a paper: The tools and techniques of the adversarial reviewer,", SIGMOD, 2008.
- Keshav, S., "How to Read a Paper,", 2016.
- Roscoe, T., "Writing Reviews for Systems Conferences,", 2007.
- Example conference review guidelines
- Example conference review results (ICLR'21)
- ICLR'21: Example review data
- ICLR'21: Reviews and scores
- OSDI'18: Reviews for Frank McSherry's paper