Final project for COMS 4774 Spring 2021

Overview

The purpose of this project is for you to engage with the research literature in machine learning.

You will read and understand a substantial research paper on some topic related to machine learning (and ideally on a topic that goes beyond plain-old supervised learning). This paper must have a significant and meaningful theoretical component (e.g., theorems, proofs).

You will also synthesize something “new” based on the paper. This can range anywhere from illustrative examples of the ideas from the paper, to extensions of the theoretical results in the paper. They need not be at the level of “publishable” contributions, but you are also welcome to be ambitious and creative.

You will write a report that reviews the research paper and describes your new contribution.

You can do this project individually or in teams of two (i.e., pairs).

Report

Write a report completely in your own words containing the following.

  1. A high-level review of the paper. This should include a brief summary of the paper, what problem the paper tries to solve (or question the paper tries to answer), the main theoretical results, the key ideas/insights, etc. It should also mention strengths and/or weaknesses of the paper (in your opinion). Here is an example of a high-level review.

  2. Something “new”. This can be a new insight you have from thinking about the paper, an illustrative example of some idea from the paper, an extension of a theoretical result in the paper, an empirical study based on the paper, a simplification of a proof from the paper, etc. Your report must very clearly indicate what is your new contribution. Whatever you come up with, you should write it up as clearly and precisely as possible.

The report should be at most 7 pages (with additional pages for references and appendices). The key components of your report must fit within the seven pages. The appendices are only for additional material that you feel is important to include, but it will not be graded.

Please use the following \(\LaTeX\) template: project-report.tex, project-report.pdf. (If you don’t want to use \(\LaTeX\), please ensure your report has similar formatting.)

Milestone 1: Paper selection

Select 3-5 candidate papers for your project. Submit your selections on Gradescope, ordered from most preferred to least preferred. I will choose one of them for you (assuming there is a suitable candidate; if there isn’t, I will assign a different paper).

  • For each paper:
    • Provide the paper title, authors, bibliographic information, and the URL for a freely-accessible version of the paper.
    • The paper must have a substantial and meaningful theoretical component.

Deadline: February 19 at 10 PM.

If you are working in pairs, only one person needs to submit, but that person should select (on Gradescope) the other team member.

Here are the assigned papers based on your proposals.

Milestone 2: Progress report

Submit a very brief progress report on Gradescope.

If you are working in pairs, only one person needs to submit, but that person should select (on Gradescope) the other team member.

Deadline: March 15 at 10 PM.

Milestone 3: Final report

Submit the final report on Gradescope.

Deadline: April 15 at 10 PM.

Grading

The grade will be based on (1) the successful completion of the various milestones and (2) the quality of the final report.

If you are working in pairs, then the default is that both members will receive the same grade. But if there is any issue with work-share (e.g., your partner disappears), please let me know ASAP.