Sep 30
(We didn't meet - Tim sent a 9:30am email saying he was sick.)
Jim - 9:40am
A Markov model is a common tool for working with tunes, looking
at probabilities of the next note based on the previous sequence
in some way. This is what Abe Stimson did with me in his 2008
plan with me, which is on a poster outside my office.
I look forward to seeing what you come up with.
Tim - Fri 3:30am
I mostly spent the week thinking about how to break down the problem of detecting motifs, but I wasn't able to do much coding. I started with the idea of finding phrases, before realizing that the way most literature defines phrases is not how I was thinking of them. Instead, the goal became finding (possibly short) motifs that can be incorporated into newly generated phrases--maybe more reasonably defined as melodies, though that word has certain tonal connotations I would like to avoid. The first methods I played with were largely brute force style list checking, but since that's not very elegant and would probably result in performance issues down the road, I have been looking for more interesting ways to do this and I returned to Russell and Norvig's Artificial Intelligence: A Modern Approach, mostly their sections on Markov models and Bayesian networks (chapter 15). It's not a lot of tangible work, but I have a clearer picture of how I should be utilizing the system's input.
