Paper Reviews#

Each week, you will be asked to read at least four papers. You are required to choose 2 of them to write a paper review on. Paper reviews consist of 3 paragraphs:

  • 1 paragraph summarizing the key takeaways from the first paper
  • 1 paragraph summarizing the key takeaways from the second paper
  • 1 paragraph on connections between the two papers, such as:
    • Comparing and contrasting
    • How ideas from one paper can be applied to solve the problem in the other
    • A new method incorporating ideas from both paper

Paper reviews should be submitted on Gradescope (code: PXXDRW) and are due Tuesday night anywhere on Earth (equivalently, 8am ET on Wednesday) before the corresponding paper presentations. We do not allow any late days for paper reviews, since they will not be useful after the papers have been presented. However, you are allowed to miss one review without losing credit.

Paper Presentations#

On paper presentation days, we will discuss 2-3 papers. Each paper is assigned to one student, who has 15 minutes to present. After all of the presentations for the day, we will have small-group and large-group discussion. Your presentation should 1) motivate the work well, 2) place it in context of previous work, 3) explain high level ideas and technical details, 4) go over experimental results and 5) discuss limitations and potential impacts of the work. Presenters are expected to follow the guidelines in this slide template. Presenters should also submit 5 discussion questions about their paper the day before as a direct message to the instructors on Piazza.

Scribing#

As an alternative to presenting a paper, students may instead create scribe notes for one of the lectures. Since there are only 13 total lectures, only 13 students can scribe. A scribe's role is to create a record of lecture content and discussion that clearly explains the material. Scribe notes will be posted online and will serve as a useful reference for your fellow students and researchers. For more instructions, refer to our guidelines for scribing.

Homework#

There will be one open-ended homework, which is intended to give everyone a taste of how research is done. You may pick any topic related to the course and do any of the following:

  • Re-implement the algorithm proposed in a paper and study the reproducibility of some of the reported experiments
  • If a paper has a code release, first run this code to verify results in the paper. Then, investigate new variations/ablations on the algorithm or investigate how the algorithm fares in new environments
  • Conduct experiments to compare two or more methods that address the same class of problems

The homework should be done in teams of 2-3 students. You will submit a 4-page PDF (including references) describing your investigation in a self-contained way. Make sure it has the following sections:

  • Statement of What Idea/Hypothesis is Being Investigated
  • Motivation Behind the Investigation
  • Description of the Experiments (and/or Theoretical Analysis)
  • Conclusions and Opportunities for Future Work resulting from your investigation.

The homework will be due on October 5th at 11:59pm through Gradescope. Late submissions will be penalized 10% every day, at the granularity of 1 day (e.g., 1 day and 2 hours late would be rounded up to 2 days late).

Final Project#

The final project will be your opportunity to explore some of the topics introduced in the course more deeply.

  • Your project should be related to the course. This is a chance to show off what you've learned.
  • Your idea should be fun/cool: systems vs theoretical doesn’t matter as long as it has a cool demonstration.
  • Do not commit to an idea too early. Try open-source code and make sure your idea makes sense.
  • You should work on the project in a team of 2-3 people.
Deadline
Project proposal (1-2 pages)10/12/2022
Project midterm report (4-8 pages)11/14/2022
Project presentation12/07/2022
Project final report (6-8 pages)12/14/2022