Boosted Classifiers for Fast Parameter Estimation

Paul Viola

Microsoft Research

Estimation of hidden parameters from complex observations is notoriously 
difficult.  Examples include finding the pose of a human body in an image, or 
the expression of a face.  One approach is to use a very large database of 
examples to "represent" the relationship between the parameters and the 
corresponding image.  Parameter estimation is performed using some variant of 
nearest neighbor lookup.  There are two real problems with this approach.  One 
is the lack of sufficient training data.  The other is computational efficiency. 
I will describe an approach which links recent success in real-time object 
detection with Locality Sensitive Hashing to construct a system which requires 
less data and is extremely fast.

Bio: Before moving to Microsoft Paul Viola was a researcher at MERL and an 
Associate Professor of Computer Science and Engineering at the Massachusetts 
Institute of Technology.  He also spent two years as a visiting scientist in the 
Computational Neurobiology of the Salk Institute in San Diego.  Paul has a broad 
background in advanced computational techniques, publishing in the fields of 
computer vision, neurobiological vision, medical imaging, mobile robotics, 
machine learning, and automated drug design. Paul was a recipient of a National 
Science Foundation Career award in 1998. He has worked on research and 
development with a number of companies including:  Compaq, IBM Research, Arris 
Pharmaceuticals and Intarka.