Linked Tucker2 Decomposition for Flexible Multi-block Data Analysis

Tatsuya Yokota, Andrzej Cichocki

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

2 Citations (Scopus)


In this paper, we propose a new algorithm for a flexible group multi-way data analysis called the linked Tucker2 decomposition (LT2D). The LT2D can decompose given multiple tensors into common factor matrices, individual factor matrices, and core tensors, simultaneously. When we have a set of tensor data and want to estimate common components and/or individual characteristics of the data, this decomposition model is very useful. In order to develop an efficient algorithm for the LT2D, we imposed orthogonality constraints to factor matrices and applied alternating least squares (ALS) algorithm to the optimization criterion. We conducted some experiments to demonstrate the advantages and convergence properties of the proposed algorithm. Finally, we discuss potential applications of the proposed method.

Original languageEnglish
Title of host publicationNeural Information Processing - 21st International Conference, ICONIP 2014, Proceedings
EditorsChu Kiong Loo, Keem Siah Yap, Kok Wai Wong, Andrew Teoh, Kaizhu Huang
PublisherSpringer Verlag
Number of pages8
ISBN (Electronic)9783319126425
Publication statusPublished - 2014
Externally publishedYes
Event21st International Conference on Neural Information Processing, ICONIP 2014 - Kuching, Malaysia
Duration: 3 Nov 20146 Nov 2014

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference21st International Conference on Neural Information Processing, ICONIP 2014


  • Alternating Least Squares (ALS)
  • Common and individual components
  • Group component analysis
  • Group data analysis
  • Tensor decompositions
  • Tucker2 decomposition


Dive into the research topics of 'Linked Tucker2 Decomposition for Flexible Multi-block Data Analysis'. Together they form a unique fingerprint.

Cite this