Statistical Learning: From Theory to Applications
Ingo Steinwart (Los Alamos National Laboratory)
NICTA SML SEMINARDATE: 2009-05-22
TIME: 11:00:00 - 12:00:00
LOCATION: NICTA 7 London Circuit, Ground Floor Seminar Room
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
This talk will give provide an overview on some of the problems I have been working recently. The first part of the talk will deal with the statistical analysis of kernel machines, where in particular I will discuss learning rates. The second part will present a new solver for SVMs that, in part, is based on statistical insights. It turns out that the core solver is typically 2-4 times faster than an improved implementation of LIBSVM and even 4-8 times faster than the original LIBSVM implementation. In the final part of my talk, I will present an application in which anomalous changes between two images of the same scene are sought. Here a key challenge is to achieve a false alarm rate in the order of 10^-4 to 10^-5 by a fully automated SVM implementation.
BIO:
http://www.c3.lanl.gov/~ingo/
