We are pleased to welcome Bradley WORLEY (Institut Pasteur) next Thursday 22/02, 14h30, at LIX (building Alan Turing, 1 Rue Honoré d'Estienne d'Orves, 91120 Palaiseau) room Philippe Flajolet.
Abstract: Sparse recovery is a cornerstone of nonuniform sampling (NUS) methods in nuclear magnetic resonance (NMR), magnetic resonance imaging (MRI), and other instrumental techniques. While most recovery algorithms can find globally optimal solutions by means of convex optimization, they generally do not provide error estimates, which are critically important in scientific applications. We introduce the variational method of constructing approximate Bayesian inference algorithms, and show recent results of its use in NUS NMR reconstruction. Connections with existing compressed sensing algorithms and current efforts towards improved performance will also be presented.