Location
Cherry Auditorium, Kirk Hall
Start Date
2-16-2012 1:00 PM
Description
The most popular methods for computing eigenvalues and singular values of a very large matrix are based on Sorensen's (1992) Implicitly Restarted Arnoldi (IRA) method. Currently, the IRA method is used by Mathworks, in the Matlab functions eigs and svds. We have developed several new fast SVD algorithms, based on Sorensen's method, but instead use the Lanczos bidiagonalization process. Our Matlab codes for the SVD problem are faster and more reliable than the MATLAB svds code. One of our Matlab codes, irlba.m has also been translated into the computer language R. For this talk, we will explore the mathematics of our newly developed SVD methods and show several computed numerical examples.
Methods for Computing a Partial Singular Value Decomposition of a Very Large Matrix
Cherry Auditorium, Kirk Hall
The most popular methods for computing eigenvalues and singular values of a very large matrix are based on Sorensen's (1992) Implicitly Restarted Arnoldi (IRA) method. Currently, the IRA method is used by Mathworks, in the Matlab functions eigs and svds. We have developed several new fast SVD algorithms, based on Sorensen's method, but instead use the Lanczos bidiagonalization process. Our Matlab codes for the SVD problem are faster and more reliable than the MATLAB svds code. One of our Matlab codes, irlba.m has also been translated into the computer language R. For this talk, we will explore the mathematics of our newly developed SVD methods and show several computed numerical examples.
Comments
Downloadable file is a PDF of the original event flier.