APPLIED MATH SEMINAR

Title: A randomized algorithm for approximating an SVD of a matrix

Speaker: Mark Tygert, Yale University

When/where: Tuesday, November 7th, 4:15PM, AKW 200

Abstract:
This talk will describe a robust, efficient randomized algorithm for computing an approximation to a singular value decomposition (SVD) of a matrix A for which A and its transpose may be applied rapidly to arbitrary vectors. Given any positive integer k, the algorithm constructs a rank-k approximation whose accuracy is of the same order as the accuracy of the best possible rank-k approximation. The algorithm has a rather negligible probability of failure (1e-17 is typical), and operates reliably independently of the structure of A (unlike the classical Lanczos method for computing an approximation to an SVD of a matrix A). The results will be illustrated via numerical examples.