I am a researcher at Microsoft Research, part of the Mobile Computing Research Group. I am also an affiliate professor at the UW Department of Computer Science and Engineering. From 2001 to 2011, I was a researcher at Intel Labs.
I build systems for recognizing human state, especially human behavior, based on sensor data. My focus is on statistical algorithms for recognition, but I am interested in sensor design, model elicitation, infrastructure design and high-performance/low-power implementations as well. My early work showed how a new breed of sensors based on Radio Frequency Identification (RFID), when combined with state of the art statistical reasoning techniques, can dramatically improve practical human activity recognition. More recently, I have focused on using mobile vision and 3-D footage to improve recognition quality.
I have a strong parallel interest in applying behavior-monitoring
technologies to the problem of providing care, elder care in
particular. I have helped build, deploy and evaluate several such
systems in partnership with Intel's Digital Health business unit and
the US Department of Veterans Affairs.
My long-term vision is to build interactive machines with common
sense. To this end, I have shown how to automatically extract large
(~100,000 variables) sensor-based models of daily life from
web-based text and games. Perhaps surprisingly, these "machine-mined"
models can be used both to infer activities from sensor observations
of people in their homes and in turn, to refine the models
themselves. I believe it is possible in the next decade to build a
machine that reads widely, observes the world, understands much of
what it reads/sees and automatically gets better at these over
time.