Your team picks one question from the course and answers it with data and code: extend a result from a paper or a lecture, add a missing control, apply a method we used to new data, or test a model's prediction on people. A carefully analyzed negative result is welcome; we grade the question, the method and the interpretation, not whether the effect shows up.
Teams have two students (one team of three only if the class count is odd). Register on Canvas by Thu 11/19 (People → Groups → Project Teams).
The research trail was meant to get you reading papers critically. If one of your entries turned up a question you would like to pursue, it can become your project; otherwise, come up with your own.
You have about four weeks: plan for one model, one or two datasets, and one comparison that could fail. Start from a model or pretrained features rather than a long training run. Use any data or models from the lectures and assignments, or anything else you can get your hands on; say in the proposal what your data can and cannot show. Run on the course's Colab; if a project truly outgrows it, talk to a TA early about Oscar, Brown's cluster. Track your runs; we show Weights & Biases for this in recitation.
Updated Tue 11/17: the proposal is now due Sun 12/6 at 11:59 PM (was Wed 12/2, the day after Exam 2). Changes are in red.
| Date | What | Weight |
|---|---|---|
| Thu 11/19 | Team registered on Canvas | none |
| Thu 12/3 – Thu 12/10 | Project work sessions in class | none |
| Sun 12/6, 11:59 PM (was Wed 12/2) | One-page proposal, one per team; feedback at the Tue 12/8 session | 3% |
| Mon 12/21, 9:00 AM | Poster session and two-page report | 12% |
Proposal (one page). Answer, in order: your question in one sentence; why it matters for how a model and a brain or mind relate; your data; your model or method; your baseline or control, and what a skeptic would say is missing; what result would change your mind; who does what.
Report. Two pages in the CCN extended-abstract format (template and example), abstract of at most 300 words; references don't count toward the limit.
Poster. Four letter-sized slides, printed one per sheet and mounted on the wall, plus an optional appendix slide. Half the teams present while the others tour, then you switch. Give each visitor a three-minute walk-through, then take questions. Both teammates must be able to present the whole project alone.
Unlike on the assignments, AI tools (ChatGPT, Copilot, Claude, or similar) are permitted on the final project, including for writing and debugging your code. Three conditions apply:
In class we will show you how to use AI safely and productively for a research project, and go over in more detail what is and is not permitted. If you are unsure whether a use is allowed, ask before submitting.
Grades are individual and reflect each student's contribution and understanding, judged in part from your own walk-through and answers at the poster. The 12% splits into research quality (half), report and poster (a quarter), and your own presentation (a quarter).