Given that the dataset of YC companies is starting to be somewhat sizable, I'd be incredibly curious to see if people tried to run a regression between lots of plausible variables that affect success (founder demographic variables, subject area, exact time, etc) to see if there are any non-trivial correlations.
The dataset is not quite big enough to train a very complex deep net, but even some simple linear regressions would be fascinating.
The dataset is not quite big enough to train a very complex deep net, but even some simple linear regressions would be fascinating.