We propose two different types of course projects, you need to pick one by 29th October (Moodle form):
A - Labs Project. Implement several extensions of the Lab assignments (proposed within each lab description).
B - Research Project. Implement a ML based solution for a problem of your choice based on the work presented on a research paper (based on the official repository code released with that paper).
Common requirements for both types of project:
The projects are individual (although of course we encourage discussion with your classmates, we just expect you to be honest and submit your own work).
Mid-semester check point with the assigned project supervisor.
Prepare a "spotlight" final presentation of your results (5 min) of your work.
Write a final report of the developed work following the corresponding template.
PROJECT TYPE A TEMPLATE: https://www.overleaf.com/read/nccjycmbcnbj
PROJECT TYPE B TEMPLATE: https://www.overleaf.com/read/wdtkwwkyxyzv
Specific requirements:
Project A: you need to perform 10 points in total from the proposed "project tasks" from all the Labs. You'll have 3 points available at least in each of the labs. You need to take at least 2 points from each of the four labs, the other 2 points you can pick the ones you prefer. In the final project report you need to describe your results from tasks accounting only for 10 points (even if you decided to do more just to learn, sorry but we'll only evaluate 10 points for the project).
Project B: you'll have an initial check point around 1 week after submitting the proposal (the week of Monday 8th November). Before this initial meeting, you need to have done the following:
Try to run the demo code (maybe all, maybe just the inference part, you may just run 1 iteration of training … Just something reasonable, for you to find out and let us know which parts you can run or not from the base repo).
Download the dataset or simulator you think of using in the project, to make sure it’s actually what you expected and available and “downloadable”
Make sure that the document you have as “core” reading related to your project is an actual published paper (and therefore reviewed paper. This means if the paper is only in arxiv is has not been reviewed and published yet. If you have doubts about this, it can be discussed in the following individual meetings)
In the meeting with your assigned project tutor, you will agree wether the project sounds reasonable and feasible, and what are the tasks to be implemented in the project.
Project tutor: everyone (project A and B) has an assigned faculty as tutor for the project. You have the updated list in moodle (Project Type Selection + Tutors assigned). This is your contact point for general issues with the project as well as for the intermediate checks, etc …
Project A: for particular technical questions on each of the tasks within your project, please refer to the faculty that taught that lab.
Project B: make sure you had agreed on the planning and details of project B yet with your tutor
Check point (for everyone). We will do an intermediate check-point (each of you with your assigned tutor) around the end of November. It will be a very short meeting (we will pass you slots in Meet to pick) just to show your progress to the tutor and get advice on how to prioritize what’s left.
If you have done more half of your project or more, that would be amazing!
If you bring “something” already done (e.g., some experiments, and some sections already drafted in the report, that would be considered a great check point outcome! )
If you have barely started … we will all start to get a bit worried about that project
Submission & Evaluation.
1) You’ll have to submit to moodle the following, by the official date of the exam:
Your report
The code corresponding to your solution or solutions
2) You’ll have to give around 5 min presentation (January, dates to be decided: either last week of class or continuous evaluation period) of your results and submit this presentation to moodle
The best place to start is to access the paper repositories of the major conferences of machine learning, computer vision, robotics and graphics for inspiration. Some of the papers already have a code link in the repository or include a repository link in the PDF. Here, we include a list of many top conferences with open repositories for papers and code.
- ML conferences: NeurIPS, ICML, AISTATS, IJCAI, UAI...
- CV conferences: ICCV, CVPR, ECCV...
- Robotics conferences: RSS, CoRL...
This site is a compilation of many papers with available repository, it would be a good place to quickly start looking: https://paperswithcode.com/
Kaggle.com is a good place to find datasets from public competitions and challenges related to Machine Learning, BUT remember you also need an associated research paper and code repository for your project proposal.
Papers that are included ONLY in arXiv or similar paper repositories. That usually means that the work has not been properly reviewed and verified.
-Papers that require many resources. You should be able to make it work in a personal computer with open datasets. Big projects from large companies are usually a red flag.
-Media over science. Popularity might not directly correlated with scientific quality.
-Do not underestimate classical methods. Recent does not always mean better. Classical methods can also be easier to reproduce in your computer with smaller datasets/simpler scenarios.