Skip to Main content Skip to Navigation
Conference papers

Data Mining by NonNegative Tensor Approximation

Abstract : Inferring multilinear dependences within multi-way data can be performed by tensor decompositions. Because of the presence of noise or modeling errors, the problem actually requires an approximation of lower rank. We concentrate on the case of real 3-way data arrays with nonnegative values, and propose an unconstrained algorithm resorting to an hyperspherical parameterization implemented in a novel way, and to a global line search. To illustrate the contribution, we report computer experiments allowing to detect and identify toxic molecules in a solvent with the help of fluorescent spectroscopy measurements.
Document type :
Conference papers
Complete list of metadata

Cited literature [16 references]  Display  Hide  Download
Contributor : Pierre Comon Connect in order to contact the contributor
Submitted on : Monday, October 27, 2014 - 4:21:32 PM
Last modification on : Tuesday, October 19, 2021 - 11:22:38 PM
Long-term archiving on: : Wednesday, January 28, 2015 - 10:36:15 AM


Files produced by the author(s)

Les fichiers déposés sont bien ceux produits par les auteurs selon le style demandé par l'éditeur, et pas le fichier de l'éditeur. Le copyright et l'en-tete peuvent etre retirés si cela est imposé par HAL.


  • HAL Id : hal-01077801, version 1




Rodrigo Cabral Farias, Pierre Comon, Roland Redon. Data Mining by NonNegative Tensor Approximation. MLSP 2014 - IEEE 24th International Workshop on Machine Learning for Signal Processing, Sep 2014, Reims, France. ⟨hal-01077801⟩



Record views


Files downloads