TY  - CONF
CY  - Budapest, Hungary
ID  - SisLab2193
UR  - http://dx.doi.org/10.1109/EUSIPCO.2016.7760629
A1  - Nguyen, Viet Dung
A1  - Abed-Meraim, Karim
A1  - Nguyen, Linh Trung
Y1  - 2016///
N2  - Tensor decomposition is an important tool for many applications in diverse disciplines such as signal processing, chemometrics, numerical linear algebra and data mining. In this work, we focus on PARAFAC and Tucker decompositions of three-way tensors with non-negativity and/or sparseness constraints. By using an all-at-once optimization approach, we propose two decomposition algorithms which are robust to tensor order over-estimation errors, ? a desired practical property when the tensor rank is unknown. Different algorithm versions are proposed depending on the desired constraint (or property) of the tensor factors or the core tensor. Finally, the performance of the algorithms are assessed via insightful simulation experiments on both simulated and real-life data.
TI  - New robust algorithms for sparse non-negative three-way tensor decompositions
SP  - 2151
M2  - Budapest, Hungary
AV  - restricted
EP  - 2155
T2  - 24th European Signal Processing Conference (EUSIPCO)
ER  -