Project Description

The goal of the projects is to dive deep into one or more of the topics and methods we discuss in this class and get some hands-on experience.

Projects can be one of two paths:

1. Path I: Novel research project. Perhaps integrating some form of self-supervised learning into your own research. Evaluation will be similar to that of a paper review process. (If the project works out especially well, the ICML deadline will be just around the corner by the end of the course.)

2. Path II: Reproduction of a published work. This path involves the participation in the ML Reproducibility Challenge 2020. There is a chance your submission will be selected for a special edition of a journal.

Projects will be done in groups of up to 4 students.

Elements:

• Project proposal: 1-2 pages indicating the selected path, describing the project including the projected contributions and evaluation methodology. To be handed in on the class Studium page.

• Written report: Standard paper submission: ICML paper format (path I) or the Reproducibility Program paper format (path II).

• Participation in a poster session

Project Timeline:

• Project proposals due: September 22nd 2020.

• Written report due: December 18th 2020.

• Poster session: (tentative) 18h30-20h30, Thursday December 10th 2020

Evaluation:

The course project counts for 50% of the final grade. The evaluation will be based on the following decomposition (out of 100%):

• project proposal: 10%

• poster presentation: 20%

• quality of the written report: 20%

• methodology: 25%

• the quality of the contributions: 25%

Office Hours:

If you want to discuss project ideas, feel free to book an office hour slot with Ankesh here.