Letters: Numerical optimization of a sum-of-rank-1 decomposition for n-dimensional order-p symmetric tensors

Authors:
Olexiy Kyrgyzov;Deniz Erdogmus
Affiliations:
Northeastern University, Department of Electrical and Computer Engineering, 380 Huntington Avenue, Boston, MA 02115, USA;Northeastern University, Department of Electrical and Computer Engineering, 380 Huntington Avenue, Boston, MA 02115, USA
Venue:
Neurocomputing
Year:
2010

Citing 7
Cited 1

Polynomial roots from companion matrix eigenvalues

Mathematics of Computation
Decomposition of quantics in sums of powers of linear forms

Signal Processing - Special issue on higher order statistics
The Jacobi Method for Real Symmetric Matrices

Journal of the ACM (JACM)
A Multilinear Singular Value Decomposition

SIAM Journal on Matrix Analysis and Applications
On the Best Rank-1 and Rank-(R1,R2,. . .,RN) Approximation of Higher-Order Tensors

SIAM Journal on Matrix Analysis and Applications
Canonical decomposition of ictal scalp EEG and accurate source localisation: principles and simulation study

Computational Intelligence and Neuroscience - EEG/MEG Signal Processing
Tensor Decompositions and Applications

SIAM Review

Topological features in 2D symmetric higher-order tensor fields

EuroVis'11 Proceedings of the 13th Eurographics / IEEE - VGTC conference on Visualization

Quantified Score

Hi-index	0.01

Visualization

Abstract

In this paper, we present a sum-of-rank-1 type decomposition and its differential model for symmetric tensors and investigate the convergence properties of numerical gradient-based iterative optimization algorithms to obtain this decomposition. The decomposition we propose reinterprets the orthogonality property of the eigenvectors of symmetric matrices as a geometric constraint on the rank-1 matrix bases, which leads to a geometrically constrained eigenvector frame. Relaxing the orthogonality requirement, we developed a set of structured-bases that can be utilized to decompose any symmetric tensor into a similar constrained sum-of-rank-1 decomposition.