I’m a grad student working on Deep Learning. In 2016 I completed Master of Science degree at Virginia Tech and started my PhD. At the beginning of 2017 I moved to Georgia Tech with my advisor, Dhruv Batra, to continue my PhD. I sit at desk 271 in the College of Computing Building and you can reach me at michael.a.cogswell at gmail.
I’m interested in AI, in understanding how intelligence can be constructed and through that how it works. In particular, the complex models allowed Deep Learning open up a number of new questions that interest me:
What representation of knowledge do these networks learn? How can we understand those representations and the process by which they are learned? How do we measure and evaluate this understanding?
Where do deep networks fail? Why?
What principles do deep networks take advantage of at the moment? What principles are missing? How can we incorporate new principles?
I like to explore these questions in the context of computer vision. My tool set and mode of thinking fits well into this domain, but the domain is also ripe with problems that interest me. Vision problems offer raw inputs which require complex models of perception to understand. Still, effective vision systems exist in animals, so at least part of the problem is simple enough to solve without more complex human-like abstractions.
I wrote an essay about how I got interested in the field a while ago.
Aside from research, I’ve been an active fencer since 2005, competing in club collegiate tournaments and USFA events (mostly in the Virginia Division). I’m currently a C2017 in foil. I fence at the Eagle Fencing Club and sometimes Georgia Tech’s Yellow Jacket Fencing Club.