Electrical Engineering and Computer Science Theory Seminar

Approximating covariance matrices

Roman Vershynin

Approximating covariance matrices

U-M Math

Friday, November 06, 2009
10:30 am - 11:30 am
3941 CSE

Abstract

How well can one approximate a covariance matrix of a distribution in R^n by sampling? This is a problem of multivariate statistics, but it also arises prominently in volume algorithms, going back to Kannan, Lovasz and Simonovits. I will describe some classic and recent approaches to this problem, ranging from quantum information theory to decoupling schemes.

Additional Details

Sponsor: Martin Strauss
Open To: Public