Deep Learning Semester Project
In this project you will apply deep learning to some data set. You may use your own data or a data set that I provide.
You may work alone or in a team. If you work in a team, I will expect a little more from your project. You may team up with someone outside the class if you are working on a research problem joutside of this class.
If you are using this class project to work on research outside the class, I strongly recommend that you try to publish your work. I will work with you to identify an appropriate publication venue, design your research methodology, and write your paper.
You will produce the following deliverables, due at appropriate points during the semester (see the schedule):
- Project proposal. Describe your data, where it comes from, what you want to get from the data, and why it's interesting.
- (Intermediate) Results. The purpose of this deliverable is to ensure that you get your deep learning pipeline working end-to-end early enough that you can act on feedback from preliminary results.
- Report the results (so far) of the deep learning model you developed, using confusion matrices, ROC curves, and any other appropriate performance measures.
- What baseline are you using to conclude that your model has learned something?
- Is your test error significantly higher than your training error? Is it higher than baseline error?
- If your model is not performing well, why?
- If your model is not performing well, what is your strategy to improve it?
- Finished research paper. A finished research paper ready to submit for publication. Of course, you may continue to work on it after this class is over.
- Presentation. A 20-minute conference-style presentation to the class during the last two or three class meetings, depending on how many groups we have.
You will write your proposal, results and finished research paper in using the style files of the venue to which you intend to submit your work, or NeurIPS style files if you don't (yet) have a target venue. Your code should be written in Python using the PyTorch deep learning framework. You should keep your code, data, and paper up to date in a private GitHub repository to which you grant me access.