MM Adapted MH Methods

Yosra Marnissi (1) , Emilie Chouzenoux (2) , Amel Benazza-Benyahia (3) , Jean-Christophe Pesquet (4, 2)

1 SAFRAN [Paris]
2 OPIS - OPtimisation Imagerie et Santé
3 SUP'COM [TUNIS] - Ecole supérieure des communications de Tunis
4 CVN - Centre de vision numérique

In this paper, we derive a novel MH proposal, inspired from Langevin dynamics, where the drift term is preconditioned by an adaptive matrix constructed through an MM strategy. We propose several variants of low-complexity curvature metrics applicable to large scale problems. The proposed method is shown to exhibit a good performance in an image recovery example.

Domaines: Traitement du signal et de l'image [eess.SP]

Yosra Marnissi, Emilie Chouzenoux, Amel Benazza-Benyahia, Jean-Christophe Pesquet. MM Adapted MH Methods. BASP 2019 - International Biomedical and Astronomical Signal Processing Frontiers workshop, Feb 2019, Villars sur Ollon, Switzerland. ⟨hal-02314412⟩