The reading project is an opportunity to engage with research on machine learning theory. You are free to pick any topic you like, within reason. You may work individually or in pairs. The expectation for projects done in pairs is naturally higher than for those done individually.

Requirements

  1. Read a few research papers (2-3 papers, depending on how substantial they are) on a topic in machine learning theory (broadly construed). Please see the list linked below for some paper suggestions.

    Paper suggestions

  2. Write a short report that describes the main results, ideas, techniques, etc. Try your best to keep it to 5-6 pages (at 11 pt font, single space).

    1. The report should clearly demonstrate a solid understanding of the papers you have selected to read, including the main problems and solutions (whether they are algorithms, lower bounds, etc.). The report doesn’t have to exhaustively repeat everything in the papers, especially if the papers contain large number of results.
    2. The report should be understandable to another student who has taken COMS 4773 but has not read the selected papers. For instance, this means you should make sure concepts and problems that we have not covered in the class are carefully defined.

  3. Do something “new” / “creative”. This might be to extend a result from a paper, or find a simpler proof or algorithm. Or you could find a better way to explain a key idea from a paper (e.g., an animation, interactive web application, interpretive dance). Almost anything within reason is fine! Depending on what you do, you can either include this with report or just provide a web link.

    1. Since many people have asked: One thing that isn’t really appropriate for this course is to simply implement an algorithm from a paper and try it out on some dataset. (That is more of a project for an applied ML course.)

Milestones

  1. Friday, March 22: Please submit a 1-2 page project proposal in which you list the papers you plan to read, and why you have chosen these papers. You should have at least looked at the papers to know what they are about, but of course you don’t have to have read them in detail. One of the purposes of this proposal is to solicit feedback from the instructor / TAs. This will be done on Gradescope. [If you are working in pairs, only one person needs to submit, but please make sure you use the Gradescope feature for specifying who is in the group.]
  2. Friday, April 19 (11:59 pm): Send a short progress report by email to the instructor. Use the subject line “COMS 4773 Progress Report”. Just briefly explain what you’ve done so far. I hope the progress is nonzero. And please give some indication of what you are doing for the “new” part of the project.
  3. Friday, May 3 (11:59 pm): Submit your final report on Gradescope.