Nos partenaires



Accueil du site > Français > Evénements > Séminaires > Séminaires IRIT-UPS

Séminaires IRIT-UPS



Sparse coding from nonlinear compressive measurements: applications to declipping, de-quantization and 1-bit recovery

Lucas RENCKER - University of Surrey (Royaume-Uni)

Jeudi 15 Novembre 2018, 14h00 - 15h30
Version PDF :


Sparse coding and dictionary learning are well known techniques for linear inverse problems such as denoising, inpainting or deblurring. However very often in practice, the observed signal is measurement in a nonlinear and compressive way, such as in the case of clipped (i.e. saturated), quantized or 1-bit measurements. These three problems have often been addressed in the literature by solving constrained sparse coding optimization problems, which can be difficult and computationally expensive to solve. In this talk we will propose an unconstrained formulation, which provides a convex and differentiable cost function that is simple to optimize. We will then show how classical sparse coding algorithms can be extended to deal with clipped or quantized measurements, and how the proposed framework can be used to perform dictionary learning directly from nonlinearly compressed data. Finally, we will show how this approach generally fits within the framework of learning with uncertainties, and discuss open problems.

Short bio: Lucas Rencker is a PhD candidate at the Centre for Vision, Speech and Signal Processing (CVSSP), University of Surrey (UK), under the supervision of Wenwu Wang and Mark Plumbley. His research interests include sparse coding and dictionary learning for inverse problems in signal processing, with applications to audio and image reconstruction. During his PhD, he was also visiting research student at INRIA Paris, and at Cedar Audio Ltd, working on optimization and digital audio restoration respectively.