Foundations of numerical multilinear algebra: decomposition and approximation of tensors

  • Authors:
  • Gene H. Golub;Gunnar E. Carlsson;Lek-Heng Lim

  • Affiliations:
  • Stanford University;Stanford University;Stanford University

  • Venue:
  • Foundations of numerical multilinear algebra: decomposition and approximation of tensors
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

The subject of this thesis is best described as the study of a few new problems in multilinear algebra that are analogous to important classical problems in numerical linear algebra. Among other things, we will study the problem of finding a best low rank approximation to a tensor, a symmetric tensor, and a nonnegative tensor; define a notion of eigenvalues and eigenvectors for symmetric tensors and a notion of singular values and singular vectors for general tensors; we apply our multilinear spectral theory to hypergraphs in a way that parallels spectral graph theory, and to nonnegative tensors in a way that parallels the results of Perron-Robenius; we will also study an extension of nonnegative matrix factorization to nonnegative tensors.