The Weizmann Institute of Science
Faculty of Mathematics and Computer Science
Vision and Robotics Seminar
Greg Shakhnarovich
CS and AI Lab, MIT
will speak on
Parameter-Sensitive Hashing: relating regression,
classification and retrieval for very large data sets
Abstract:
Example-based methods are effective for parameter estimation problems when the underlying system is simple
or the dimensionality of the input is low. For complex and high-dimensional problems such as pose estimation,
the number of required examples and the computational complexity rapidly become prohibitively high. I will
describe a new algorithm that learns a set of hashing functions that efficiently index examples in a way
relevant to a particular estimation task, thus essentially embedding the parameter similarity in a Hamming space.
This is done by learning a set of simple classifiers that operate on example pairs and classify them as
similar or dissimilar. Our algorithm, called Parameter-Sensitive Hashing (PSH), extends locality-sensitive
hashing, a recently developed technique to find approximate neighbors in time sublinear in the number of examples.
PSH has been successfully applied to articulated pose estimation tasks, allowing fast and accurate estimation based
on very large example sets.
Joint work with Paul Viola and Trevor Darrell.
The lecture will take place in the
Lecture Hall, Room 1, Ziskind Building
on Sunday, September 26, 2004
at noon
Please note the unusual day