- The course project will contribute to 30% of your overall grade.
- The project should be done in teams of three people. All the team members will receive the same grade and thus it is important that all team members take part in executing integral parts of the project. Towards this, the report should also mention the contributions of all the members for the project. Note that members of a team might be penalized against any strong evidence of delinquency.
The project schedule is given below:
Milestone Info Marks Deadline Register your team Check Moodle for Google Form Oct 27 Project Proposal 1-2 pages (excluding references) 10% Oct 27 Final Report 6-8 pages (excluding references) 60% Dec 15 Presentation 10 min spotlight presentation 30% Dec 13, 14, and 15
- Only one team member should fill the Google form to register the team. You can find the Google form in Moodle.
- Project proposal and final report are to be submitted in GradeScope. GradeScope will allow you to form teams. You must form the right team in GradeScope before you make the submission.
- The final presentation will be a 5 minute presentation by all the team members followed by 5 minutes of Q/A. This will happen remotely via zoom.
The teams should select only one of the following tracks for the project.
- Reproducibility Study: The objective of this track is to assess the reproducibility of published articles by (re)implementing the work and verifying the results from the original article. The study should involve a thorough experimental investigation of the work and report all necessary details, including but not limited to experiment setting, hyperparameters, and resources used to verify the original results. Further, the study should aim at validating the conclusions drawn in the paper with necessary additional analyses and ablation study. The results of the reproducibility study can entirely or partially affirm the results and findings reported in the original work.
- Ablations/Analysis: For this track, you will choose a paper with existing implementations and analyze the algorithm proposed in the paper. You should do meanighful analysis than simple hyper-parameter tuning. The objective is to study the proposed work rigorously to gather new insights on the same task as the original work.
- Applications/benchmarks: For this track, you can apply RL for new domains/applications/tasks. You can also propose new challenging benchmarks where existing RL algorithms fail.
- Research Track: You can conduct research in RL as part of the course project. The aim of this track would be to work on new ideas/algorithms that are worth publishing in a premier conference. You can work in teams of less than 3 only for this track.
Some example papers for the first two tracks are available in Moodle.
- Project Proposal [Deadline: Oct 27]:
- Project title and track.
- Motivation and Problem Definition.
- Brief Literature review for the original research track only.
- Summary of the paper: Model/Approach details for all tracks but original research.
- List Research/Analysis Questions that you will pursue.
- Experiment setup, Dataset/Environment details.
- Plan for contributions by each team member.
- Final Report [Deadline: Dec 15]:
- Introduction and Problem definition
- Background and Motivation
- Related works
- Detailed discussion of the Research/Analysis Questions studied
- Link to Code
- Contributions by each team member.
For the final report, you should use the ICLR 2022 LaTeX template which can be found here.