Group proposal feedback

[andrewpbray]

These are three exciting topics.

The one that I think would work least well is the Beatles one. My sense is that there simply isn't much data there - either in terms of observations or potential parameters - so you'd probably end with quite simple models that wouldn't give you the chance to try out all of the tools from this class.

The other two are more promising and I'll let you all decide which one to move forward with. The first is very ambitious and rather vague right now. You'll certainly want to flesh that out - think through what your X and Y would look like, how you would gather it, what the models might look like, and how you would use that model to generate new pictures. It is certainly possible, there are just many moving parts that you'll want to nail down.

The instrument ID thing is a great idea; actually one I've wanted to see someone do for years, ever since I came across a paper proposing something called local harmonic estimation, which models the pitch, vibrato, and tremolo of an instrument and then postulates that the residuals from this model are what we mean when we discuss the "timbre" of an instrument. Maybe you can run LHE on a bunch of instruments, pull out their residuals, then built a classifier?

Here's a site that the author put together awhile back: http://www.biostat.jhsph.edu/~ririzarr/Demo/