Efficiency of linesearch strategies in interior point methods for linearly constrained signal restoration

Emilie Chouzenoux 1 Saïd Moussaoui 2 Jérôme Idier 2
1 LIGM - Laboratoire d'Informatique Gaspard-Monge
2 IRCCyN - Institut de Recherche en Communications et en Cybernétique de Nantes
Abstract : We discuss in this paper the influence of line search on the performances of interior point algorithms applied for constrained signal restoration. Interior point algorithms ensure the fulfillment of the constraints through the minimization of a criterion augmented with a barrier function. However, the presence of the barrier function can slow down the convergence of iterative descent algorithms when general-purpose line search procedures are employed. We recently proposed a line search algorithm, based on a majorization-minimization approach, which allows to handle the singularity introduced by the barrier function. We present here a comparative study of various line search strategies for the resolution of a sparse signal restoration problem with both primal and primal-dual interior point algorithms.
Keywords : Signal restoration constrained optimization interior points linesearch
Type de document :
Communication dans un congrès
IEEE Workshop on Statistical Signal Processing (SSP'2011), Jun 2011, Nice, France. pp.101-104, 2011, 〈10.1109/SSP.2011.5967631〉
Liste complète des métadonnées
https://hal.archives-ouvertes.fr/hal-00625901
Contributeur : Saïd Moussaoui <>
Soumis le : jeudi 22 septembre 2011 - 18:24:17
Dernière modification le : mercredi 19 décembre 2018 - 15:02:03

Identifiants

Collections

UNIV-NANTES | IRCCYN-ADTSI | IRCCYN | PARISTECH | ENPC | UNIV-MLV | LIGM_SIGNAL | LIGM | UPEC-UPEM | UNAM

Citation

Emilie Chouzenoux, Saïd Moussaoui, Jérôme Idier. Efficiency of linesearch strategies in interior point methods for linearly constrained signal restoration. IEEE Workshop on Statistical Signal Processing (SSP'2011), Jun 2011, Nice, France. pp.101-104, 2011, 〈10.1109/SSP.2011.5967631〉. 〈hal-00625901〉

Exporter

BibTeX TEI DC DCterms EndNote

Partager

Métriques

Consultations de la notice

290

Contact

support.ccsd.cnrs.fr
hal.support@ccsd.cnrs.fr
Logo CCSD