Thanks a lot for your suggestions!
We went around a few different directions and ended up proposing this project:
Through your eyes – reconstruct visual stimuli from EEG data
Inspired by recent successes in reconstructing images people see with their eyes from their fMRI recordings (thanks @Bitking!)
and claims of proof-of-concept results with EEG instead, we thought it would make for a very interesting and attractive hackathon effort to try and corroborate such claims. The scope is likely much more than we can afford to complete in 2 days, but we chose it rather aesthetically: it’s exciting!
We hope to attract an EEG experimentalist [1], as our core team has ML, software engineering and medical expertise. If not, we’ll pivot to the fMRI paper and try something new with their model and data, both of which are public.
[1] We can also work on open EEG datasets, but so far we haven’t found anything relevant to our questions.