Post date: Dec 13, 2018 1:7:19 PM
Dear students,
the end of the semester is approaching and I would like you to choose your final projects during the week 17-21.12.19.
Information common to both kinds of projects:
- you can work in groups of no more than 3 students
- before you start a project, please prepare a 1-2 paragraph long description about how your are going to approach it. This is especially important if you want to work on your own idea. Please discuss this during lab sessions.
There are two options for you:
You can work on the ICLR reproducibility challenge in which you will work on reproducing some experiments from submissions to the ICLR 2019 conference. Please note that the scope of the project depends on the provided resources - for some papers, the authors may provide code and data, in which case you can verify that it works as described and try applying the same principle to another problem. On the other extreme, a paper may not provide code nor data and your reproducibility effort may go towards implementing baselines or validating the idea in a limited setup.
If you choose to work on the reproducibility, please follow the procedure at the challenge website. Don't worry about the January 7 deadline - our course timeline is not well aligned with ICLR reviewing period and while your reports will not affect the acceptance process, they will benefit the scientific community nevertheless!
To help you choose a paper to reproduce I've selected a few papers in this spreadsheet. You can of course select any other paper, e.g. using this list to browse by reviewer scores.
You can work on you own idea - maybe you have a cool application project in mind, or would want to apply neural networks to some task? You can definitely do so during the project, but please consult with us and have your project accepted - we will try to keep the burden similar to the reproducibility track.
It is important to select the projects before the Christmas break, please think over them and prepare propositions for next lab sessions.