Final Project

9 minute read

As part of this course, students must complete a small research project instead of a final exam.

For this project you will either implement a current deep-learning research paper or implement a custom project that relates to your current research topic (subject to approval).

We provide you with a pool of project ideas mostly from recent papers. The idea is to select a paper that seems interesting you you, implement it, suggest at least one improvement and evaluate it as well. Some papers may be more advanced than others. Selecting a less advanced paper means we will expect you to also implement more of your own extensions and ideas to improve upon it.

During the project period you will be able to meet with course TAs for guidance during dedicated office hours. The project can be performed in groups of at most two students.

Project procedure

  1. Each group should read the topics and paper abstracts in the project topics spreadsheet.
  2. Each group must fill out the registration form with their top-3 priorities or custom project proposal until date 13/01/2019.
  3. Soon after you submit your priorities, we will approve one of them. Submitting the form sooner will increase your chance of getting your first priority.
  4. The final lecture of the course on 20/01/2019 will be dedicated to the projects. Each group will present a 5-minute (max) presentation about their topic. Feedback will be given by Prof. Alex Bronstein.
  5. You will work on your projects. Weekly (optional) guidance meetings with course TAs at dedicated office hours will be available for you.
  6. The official submission date for the project is 14/03/2019. It is advised that you finish well before this date in order to get a grade on time. However, in special circumstances we will allow submission up to 30/04/2019 (with prior approval).
  7. Your submission should be a detailed report. It should explain the problem and the paper you implemented (if relevant), explain your specific enhancements and modifications, showcase all your results (both reproduction of the paper and novel results, if any), include proper citations, etc.

Please view the project topics (updated with groups) and then fill out this form with your priorities.

Guidelines for presentations

  1. After you receive an email approving one of your project priorities, you should start working on your presentation.
  2. Presentations slides should be emailed to Aviv no later than 19/01/2019.
  3. We will publish the order of presentations in advance. Everyone will present from the same computer to reduce overhead between presenters.
  4. The presentation should address briefly the following points:
    • Full project title and description
    • Previous related work or main paper
    • Proposed method or improvement
    • Baseline you’re comparing to
    • Evaluation methods
  5. Please keep your presentation short! Just 5-6 slides are enough, up to 5 minutes total time.

Tuesdays 15:00 - 17:00, Taub 120 (VISTA lab).

Please email Chaim and Aviv beforehand if you plan to arrive.