Today is: Thu, May 23, 2013
October 23, 2012, 11:10 am to 12:00 pm
"Flag Mean: A Geometric Variable-Dimension Subspace Average," by Justin Marks (Common Hour - Lecture and Discussion)
Abstract: Given a set of subspaces, perhaps of varying dimension, of a vector space, we present a pair of algorithms to compute a mean subspace of flexible dimension. The distinction between the pair of formulations is the choice of metric in the optimization problem. We call these means Flag Means, due to the nexted relationship demonstrated by flag means of increasing dimension. The prerequisite mathematical tools for this analysis are found in a introductory-level Linear Algebra class, and include subspaces, dot products, eigenvalues and eigenvectors. We introduce the Grassmann manifold as one domain of application for the flag mean, and illustrate by computing the flag mean of subspaces generated from digital photographs of human faces.
Location: Hayes Hall 311
Sponsor: Department of Mathematics
Contact: Connie M. Miller
|Help||powered by VT Calendar|