Homeworks - 20%. This will involve coding + theoretical problems.
We will have short biweekly quizzes, just to ensure that you are keeping up with the material. Most of these quiz problems should be solvable by just following the lecture. Total weight: 20%
Writing a review/survey article - 30%. The aim should be to summarize a research direction, including at least 3-4 papers. The students should aim to submit the work to a workshop such as the ML Retrospective Workshop (NeurIPS 2020). This will also involve giving a presentation about the topic.
Reproducibility challenge - 30%. You will pick up a paper from recent ICLR/NeurIPS/ICML and implement it. See here for details about the 2020 edition--- ML Reproducibility challenge 2020. The aim again is to have a decent submission to the ReScience journal. Note that getting accepted in the challenge is not the same as getting accepted in the journal. We will share sample reviews from last year to illustrate the expectations.