Institutional tag:

Thematic tag(s):

In the past two decades, the interest in low-rank approximations for numerical solvers has been growing very quickly in order to reduce both the memory footprint and the operation count of matrix factorizations. Theory has shown that the inverse of both sparse and dense matrices coming from particular applications (e.g., finite element methods) exhibit low-rank structures, i.e., appropriately chosen sub-blocks can be well approximated using Singular Value Decomposition. Many strategies have been developed to exploit this property and we will present the most relevant ones. We will then briefly present potential applications to machine learning.

Dates:

Thursday, August 31, 2017 - 11:00 to 12:00

Location:

Inria B21

Speaker(s):

Clément Weisbecker

Affiliation(s):

Livermore Software Technology Corporation

Speaker's URL: