Phone: (213) 821-2070
Fax: (213) 821-2070
- 2005 – Present PhD student, Biomedical Eng., USC
- 2002 – 2005 MS in Biomedical Eng., USC Thesis : Virtual reality environment for human-in-a-loop simulation of FES reaching controllers
- 1995 – 1999 BS in Electronics Eng., Seoul National Univ., Korea
I’m interested in the brain mechanism of computation. The brain learns from limited number of examples but generalizes so gorgeously. (For instance, we learn language from some examples and start to speak every other sentences, we learn how to reach to a point from some practice and can reach everywhere else, and we learn a cartegory of an object from some examples of it and then can classify correctly from its every other instances.) I believe this beautiful characteristic of the brain should be at least partly due to the hardware the brain’s given to compute with – a network of neurons. Thus studying characteristics of the neural system with all its biological constraints (anatomical/physiological/developmental facts) should reveal some insights of how the brain does what it does. And also these insights should be helpful in letting the computers do what the brain does.
In an ideal case where we discover every detail of the neural system, i.e., how they’re connected, how they interact with each other and thus how their synapses are modified, then it should be possible with enough computing power that we feed input to the neural system and let them evolve and converge, to get the output cell characteristics that comply with known electrophysiological/psychophysical observations. But this is highly unplausible scenario, and one possible alternative would be such that through well-designed experiments we probe internal representations which the brain uses in each stage of its computation (assuming hierarchical functioning of the brain) to make a reasonable assumptions about them, and use these knowledge as constraints to fill in the unknown gaps in the system properties (its biology).
Currently I’m trying to tackle the problem in the domain of motion perception. Human exhibit consistent percept of motion under various configuration of stimulus (apparent, occluded, transparent motion to name a few). Electrophysiological recordings have shown much about the response properties of the cortical areas that are believed to be involved in motion processing, namely MT, MST as well as V1. Also recent psychophysical experiments in our lab show that human can metrically measure parameters of some global motions such as rotation and expansion, which seem to be consistent with response characteristics of some MST cells. I’m working on a computational theory that accounts for these new results based on the comprehensive theory developed by Yuille & Grzywacz . This approach in computational level, hopefully, may confine its underlying biological mechanism, which in turn could make insightful predictions that can be tested with experiments.
 Yuille, A. L., & Grzywacz, N. M. (1998). A theoretical framework for visual motion. In T. Watanabe (Ed.), High-level motion processing?computational, neurobiological, and psychophysical perspectives (pp. 187?211). Cambridge, Massachusetts: MIT Press.