Learning nonlinearities in the decoupling problem with structured CPD - Archive ouverte HAL Access content directly
Conference Papers Year :

Learning nonlinearities in the decoupling problem with structured CPD

Abstract

Decoupling of multivariate functions is an important problem in block-structured system identification. In the literature, different tensor-based solutions have been proposed to solve this problem using a canonical polyadic decomposition (CPD) of a Jacobian tensor. In a recent work, it has been proposed to add polynomial constraint on the model, which leads to a constrained CPD model with two factors depending nonlinearly on each other. In this work, we are interested in the estimation of the static nonlinearities with polynomial constraints. Using the reformulation as a structured CPD, we propose an alternating least squares algorithm with constrained rank-one terms. Numerical simulations show the performance and the usefulness of the proposed method compared to a competing approach.
Fichier principal
Vignette du fichier
SYSID21_YZ.pdf (1.34 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03223831 , version 1 (11-05-2021)

Identifiers

  • HAL Id : hal-03223831 , version 1

Cite

Yassine Zniyed, Konstantin Usevich, Sebastian Miron, David Brie. Learning nonlinearities in the decoupling problem with structured CPD. 19th IFAC Symposium on System Identification, SYSID 2021, Jul 2021, Padova (virtual), Italy. ⟨hal-03223831⟩
58 View
51 Download

Share

Gmail Facebook Twitter LinkedIn More