Independent component analysis for tensor-valued data

Joni Virta, Bing Li, Klaus Nordhausen, Hannu Oja

Research output: Contribution to journalArticle

8 Citations (Scopus)

Abstract

In preprocessing tensor-valued data, e.g., images and videos, a common procedure is to vectorize the observations and subject the resulting vectors to one of the many methods used for independent component analysis (ICA). However, the tensor structure of the original data is lost in the vectorization and, as a more suitable alternative, we propose the matrix- and tensor fourth order blind identification (MFOBI and TFOBI). In these tensorial extensions of the classic fourth order blind identification (FOBI) we assume a Kronecker structure for the mixing and perform FOBI simultaneously on each direction of the observed tensors. We discuss the theory and assumptions behind MFOBI and TFOBI and provide two different algorithms and related estimates of the unmixing matrices along with their asymptotic properties. Finally, simulations are used to compare the method's performance with that of classical FOBI for vectorized data and we end with a real data clustering example.

Original languageEnglish (US)
Pages (from-to)172-192
Number of pages21
JournalJournal of Multivariate Analysis
Volume162
DOIs
StatePublished - Nov 1 2017

Fingerprint

Independent component analysis
Independent Component Analysis
Tensors
Fourth Order
Tensor
Vectorization
Data Clustering
Asymptotic Properties
Preprocessing
Alternatives
Estimate
Simulation

All Science Journal Classification (ASJC) codes

  • Statistics and Probability
  • Numerical Analysis
  • Statistics, Probability and Uncertainty

Cite this

Virta, Joni ; Li, Bing ; Nordhausen, Klaus ; Oja, Hannu. / Independent component analysis for tensor-valued data. In: Journal of Multivariate Analysis. 2017 ; Vol. 162. pp. 172-192.
@article{e8c5e0711f204abfae250c27d09a6ff6,
title = "Independent component analysis for tensor-valued data",
abstract = "In preprocessing tensor-valued data, e.g., images and videos, a common procedure is to vectorize the observations and subject the resulting vectors to one of the many methods used for independent component analysis (ICA). However, the tensor structure of the original data is lost in the vectorization and, as a more suitable alternative, we propose the matrix- and tensor fourth order blind identification (MFOBI and TFOBI). In these tensorial extensions of the classic fourth order blind identification (FOBI) we assume a Kronecker structure for the mixing and perform FOBI simultaneously on each direction of the observed tensors. We discuss the theory and assumptions behind MFOBI and TFOBI and provide two different algorithms and related estimates of the unmixing matrices along with their asymptotic properties. Finally, simulations are used to compare the method's performance with that of classical FOBI for vectorized data and we end with a real data clustering example.",
author = "Joni Virta and Bing Li and Klaus Nordhausen and Hannu Oja",
year = "2017",
month = "11",
day = "1",
doi = "10.1016/j.jmva.2017.09.008",
language = "English (US)",
volume = "162",
pages = "172--192",
journal = "Journal of Multivariate Analysis",
issn = "0047-259X",
publisher = "Academic Press Inc.",

}

Independent component analysis for tensor-valued data. / Virta, Joni; Li, Bing; Nordhausen, Klaus; Oja, Hannu.

In: Journal of Multivariate Analysis, Vol. 162, 01.11.2017, p. 172-192.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Independent component analysis for tensor-valued data

AU - Virta, Joni

AU - Li, Bing

AU - Nordhausen, Klaus

AU - Oja, Hannu

PY - 2017/11/1

Y1 - 2017/11/1

N2 - In preprocessing tensor-valued data, e.g., images and videos, a common procedure is to vectorize the observations and subject the resulting vectors to one of the many methods used for independent component analysis (ICA). However, the tensor structure of the original data is lost in the vectorization and, as a more suitable alternative, we propose the matrix- and tensor fourth order blind identification (MFOBI and TFOBI). In these tensorial extensions of the classic fourth order blind identification (FOBI) we assume a Kronecker structure for the mixing and perform FOBI simultaneously on each direction of the observed tensors. We discuss the theory and assumptions behind MFOBI and TFOBI and provide two different algorithms and related estimates of the unmixing matrices along with their asymptotic properties. Finally, simulations are used to compare the method's performance with that of classical FOBI for vectorized data and we end with a real data clustering example.

AB - In preprocessing tensor-valued data, e.g., images and videos, a common procedure is to vectorize the observations and subject the resulting vectors to one of the many methods used for independent component analysis (ICA). However, the tensor structure of the original data is lost in the vectorization and, as a more suitable alternative, we propose the matrix- and tensor fourth order blind identification (MFOBI and TFOBI). In these tensorial extensions of the classic fourth order blind identification (FOBI) we assume a Kronecker structure for the mixing and perform FOBI simultaneously on each direction of the observed tensors. We discuss the theory and assumptions behind MFOBI and TFOBI and provide two different algorithms and related estimates of the unmixing matrices along with their asymptotic properties. Finally, simulations are used to compare the method's performance with that of classical FOBI for vectorized data and we end with a real data clustering example.

UR - http://www.scopus.com/inward/record.url?scp=85032711931&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85032711931&partnerID=8YFLogxK

U2 - 10.1016/j.jmva.2017.09.008

DO - 10.1016/j.jmva.2017.09.008

M3 - Article

AN - SCOPUS:85032711931

VL - 162

SP - 172

EP - 192

JO - Journal of Multivariate Analysis

JF - Journal of Multivariate Analysis

SN - 0047-259X

ER -