The tensor train (TT) decomposition is a representation technique for arbitrary tensors, which allows efficient storage and computations. For a d-dimensional tensor with d ≥ 2, that decomposition consists of two ordinary matrices and d − 2 third-order tensors. In this paper we prove that the TT decomposition of an arbitrary tensor can be computed (or approximated, for data compression purposes) by means of a backward stable algorithm based on computations with Householder matrices. Moreover, multilinear forms with tensors represented in TT format can be computed efficiently with a small backward error.

Error Analysis of TT-Format Tensor Algorithms

Dario Fasino
;
2019

Abstract

The tensor train (TT) decomposition is a representation technique for arbitrary tensors, which allows efficient storage and computations. For a d-dimensional tensor with d ≥ 2, that decomposition consists of two ordinary matrices and d − 2 third-order tensors. In this paper we prove that the TT decomposition of an arbitrary tensor can be computed (or approximated, for data compression purposes) by means of a backward stable algorithm based on computations with Householder matrices. Moreover, multilinear forms with tensors represented in TT format can be computed efficiently with a small backward error.
File in questo prodotto:
File Dimensione Formato  
FT_INdAM_2019.pdf

non disponibili

Descrizione: Articolo
Tipologia: Versione Editoriale (PDF)
Licenza: Non pubblico
Dimensione 310.18 kB
Formato Adobe PDF
310.18 kB Adobe PDF   Visualizza/Apri   Richiedi una copia
FT_submitted.pdf

accesso aperto

Descrizione: Preprint
Tipologia: Documento in Pre-print
Licenza: Creative commons
Dimensione 195.58 kB
Formato Adobe PDF
195.58 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11390/1149281
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact