Volume 43, pp. 244-254, 2014-2015.
On computing maximum/minimum singular values of a generalized tensor sum
Asuka Ohashi and Tomohiro Sogabe
Abstract
We consider the efficient computation of the maximum/minimum singular values of a generalized tensor sum $T$. The computation is based on two approaches: first, the Lanczos bidiagonalization method is reconstructed over tensor space, which leads to a memory-efficient algorithm with a simple implementation, and second, a promising initial guess given in Tucker decomposition form is proposed. From the results of numerical experiments, we observe that our computation is useful for matrices being near symmetric, and it has the potential of becoming a method of choice for other cases if a suitable core tensor can be given.
Full Text (PDF) [333 KB], BibTeX
Key words
generalized tensor sum, Lanczos bidiagonalization method, maximum and minimum singular values
AMS subject classifications
65F10
< Back