Title: Regularized kernel estimation from noisy, incomplete, inconsistent
dissimilarity information.

Speaker: Grace Wahba


Abstract: We extract information in the form of a positive definite
kernel matrix from possibly crude, noisy, incomplete,
inconsistent dissimilarity information between pairs of
objects, obtainable in a variety of contexts. Any
positive definite kernel defines a consistent
set of distances, and the fitted kernel provides a set of
coordinates in Euclidean space that attempts to respect
the information available while controlling for complexity
of the kernel. Two versions are available: The first uses
global information and the resulting set of coordinates
is highly appropriate for visualization and as input to
classification and clustering algorithms. The second version
uses only local distances and is suitable for the case
where the data is believed to lie in a low dimensional
generally nonlinear manifold in a higher dimensional
space. This version, a solution to the manifold
unrolling problem, is applicable to motion and image analysis
and semisupervised learning problems.
Several applications will be discussed. This is work
with Fan Lu, Sunduz Keles, Stephen Wright and Yi Lin.