Statistical physics approach to inference in signal processing

Internship proposed by (Stage proposé par) Lenka ZDEBOROVA

Many problems in signal processing can be viewed as models of interacting particles in statistical mechanics.

Examples of problems in signal processing is reconstruction of images in NMR or computed tomography, or learning a basis in which a certain datasets are sparse and can hence be represented and measured with lower cost.

One very promising approach to solve such problems more efficiently is based on the so-called message passing algorithms. In statistical physics terms the message passing algorithms are closely related to the replica and cavity method that are traditionally used to understand materials such as glasses or disordered magnets. The goal of this internship is to get familiar with this interdisciplinary field and apply the message passing method to one of the problems in signal processing.

Contact: ; tel: +33 (0)1 6908 8114.

#656 - Last update : 10/18 2012


Retour en haut