Some things about me. I have a couple cats who keep me company: the top kitten is Cody and the lower kitten is Kathryn. (They aren’t really still kittens, but I can’t help referring to them as such.) I adopted them from AnimalFriends when they were only about three months old. Above is a Java applet I wrote a long long long long time ago while working at the Applied Physics Laboratory (APL) at the University of Washington. Being paid to learn Java certainly has its perks. It seems I also play the piano... and grimace.... My parents started me when I was very young, and I've been playing for about 19 years now. The picture was taken during an informal recording session put together by a good friend of mine, Evan Gandy. Sadly, he doesn't have a webpage to link to. But he's a creative genius, so he may very well be famous one day. Perhaps you'll see him then.

I'm now a third year graduate student here at CMU working toward my Ph.D. in robotics from Robotics Institue under Drew Bagnell. My research interests lie at the intersection of higher level mathematics and Machine Learning; I've recently grown fond of kernel machines and structured learning. My latest research looks into new gradient based algorithms for solving structured learning problems, particularly as they relate to robotics (see Maximum Margin Planning) and specifically (or most recently) as they relate to problems in quadruped locomotion and terrain traversal. Earlier I worked on an algorithm called Kernel Conjugate Gradient (KCG) which boils down to performing conjugate gradient in the function space inherent to the particular differentiable kernel machine in question. In the end, when minimizing a differentiable regularized risk functional in a Reproducing Kernel Hilbert space, significant convergence improvements can be found via a few easy to implement tweaks to the more common parameter conjugate gradient algorithm.

Structured Prediction: Maximum Margin Techniques as author at Carnegie Mellon Machine Learning Lunch seminar, 154 views