– S. Kitic (Inria) : Sparsity & Co. : Regularization of the physics-driven and audio inverse problems Carte non disponible Date/heure Date(s) - 27 mars 2015 Catégories Pas de Catégories Sparsity & Co. : Regularization of the physics-driven and audio inverse problems\nBy Sr?an Kiti?, Inria.\n\nLow dimensional data models are powerful tools for regularizing ill-posed inverse problems. In this work, we investigate the means and performance of the sparse and cosparse (aka sparse analysis) data models for physics-driven and audio inverse problems.\n\nPhysics-driven inverse problems naturally arise from physical laws expressed through linear partial differential equations. We introduce a regularization framework based on the two data models, and show that, despite nominal equivalence, the two models significantly differ from the computational perspective. Our findings are illustrated on two example applications : sound and brain source localization (electroencephalography – EEG). \n\nIn the second part of the talk, we explore the ill-posed problem of de-saturation (de-clipping) of audio signals. We compare the performance of the two data models embedded in greedy heuristics, and show that they outperform state-of-the-art methods in the field.\n[
