A new reproducible research repository Tensorlab⁺ has been released. Tensorlab⁺ provides implementations of new algorithms, experiment files, demos and tutorials for 34 papers.

Features

Tensor decompositions

Compute the canonical polyadic decomposition, multilinear singular value decomposition, block term decompositions and low multilinear rank approximation.

Structured data fusion

Define your own (coupled) matrix and tensor factorizations with structured factors and support for dense, sparse, incomplete and structured data sets.

Complex optimization

Quasi-Newton and nonlinear least squares optimization with complex variables including numerical complex differentiation.

Global minimization of bivariate polynomials & rational functions

Real and complex exact line search (LS) and real exact plane search (PS) for tensor optimization.

Structured tensor representations

Obtain faster tensor operations and decompositions by exploiting the structure of the data, such as Hankel, Tensor Train and CPD structure.

And much more

Tensorize data, compute higher-order statistics, visualize tensors of arbitrary order, estimate a tensor's rank or multilinear rank, ...

Documentation

Visit the online user guide

or download the PDF version

Questions / bug reports

Demos

Basic Use of Tensorlab

Learn the basics of Tensorlab. This demo explains how to construct and visualize tensors. It also explains how to compute a CPD.

Multidimensional Harmonic Retrieval

The goal of this demo is to illustrate the use of the basic cpd command in an array processing application.

Independent Component Analysis

Independent component analysis finds latent variables that are statistically independent in observed data. The two related demos illustrate the computation of basic as well as constrained CPD.

Independent Vector Analysis

Independent vector analysis is a multi-set extension of independent component analysis. The problem is solved as an example of structured data fusion.

GPS: Predicting User Involvement

The goal of this demo is to predict whether a user participates in a certain activity in a certain place using a real life GPS dataset. This demo illustrates how to use the structured data fusion framework.

Additional Demos

Get an overview of all available demos. Other demos discuss the multilinear value decomposition, advanced independent component analysis and more.

Version History

Tensorlab 3.0 / Mar 28, 2016

Tensorlab 3.0 features dedicated algorithms for the decomposition in multilinear rank-$(L_r,L_r,1)$ terms, various tensorization techniques, a more flexible and expanded modeling language for structured data fusion problems, support for efficient representations of structured tensors in most optimization-based decomposition algorithms, and new algorithms for dealing with sparse, incomplete and/or large-scale datasets. A new visualization tool is introduced, many existing algorithms have received performance and flexibility updates, e.g., by using more lenient option parsing, and a number of bugs have been fixed. Finally, the user guide has been extended significantly and illustrated with practical demos.

3.0

Tensorlab 2.0 / Jan 31, 2014

Tensorlab 2.0 introduces the Structured Data Fusion framework and adds support for sparse and incomplete tensors. Speed improvements have been made across various methods, and a new method for low multilinear rank approximation has been added using adaptive cross-approximation. Tensorlab 2.01 and 2.02 bring various speed improvements and bug fixes.

2.0

Tensorlab 1.0 / Feb 11, 2013

First official version of Tensorlab introducing algorithms of various tensor decompositions, as well as different auxiliary tools such as multiplication methods, folding/unfolding commands and visualization techniques. Tensorlab 1.01 and 1.02 bring various speed improvements and bug fixes.

1.0