Extension of the semi-algebraic framework for approximate CP decompositions via non-symmetric simultaneous matrix diagonalization
Kristina Naskovska, Martin Haardt, Petr Tichavský, Gilles Chabriel, Jean Barrère
Abstract
With the increased importance of the CP decomposition (CANDECOMP/PARAFAC decomposition), efficient methods for its calculation are necessary. In this paper we present an extension of the SECSI (SEmi-algebraic framework for approximate CP decomposition via SImultaneous matrix diagonalization) that is based on new non-symmetric SMDs (Simultaneous Matrix Diagonalizations). Moreover, two different algorithms to calculate non-symmetric SMDs are presented as examples, the TEDIA (TEnsor DIAgonal-ization) algorithm and the IDIEM-NS (Improved DIagonalization using Equivalent Matrices-Non Symmetric) algorithm. The SECSI-TEDIA framework has an increased computational complexity but often achieves a better performance than the original SECSI framework. On the other hand, the SECSI-IDIEM framework offers a lower computational complexity while sacrificing some performance accuracy.
BibTeX
@inproceedings{icassp2016_extensionofthese,
title = {Extension of the semi-algebraic framework for approximate CP decompositions via non-symmetric simultaneous matrix diagonalization},
author = {Kristina Naskovska and Martin Haardt and Petr Tichavský and Gilles Chabriel and Jean Barrère},
booktitle = {ICASSP 2016},
year = {2016}
}