Sylvain Lesage

Thesis: Learning structured dictionaries for sparse modeling of multichannel signals Results: + state of the art on signal representation with a dictionary of atoms + new atomic representation formalism that integrates invariance to phase offset, time translation, filter and pitch shifting + adaptation of the Matching Pursuit decomposition algorithm + machine learning algorithms (K-SVD, MoTIF) for creating the dictionary of atoms + source separation algorithms (Demixing Pursuit, « soft » Demixing Pursuit) + part of the development of Matching Pursuit ToolKit : http://mptk.gforge.inria.fr (C++) + seven conference articles