Bayesian Methods for Finding Sparse Representations

Finding the sparsest or minimum `0-norm representation of a signal given a (possibly) overcomplete dictionary of basis vectors is an important problem in many application domains, including neuroelectromagnetic source localization, compressed sensing, sparse component analysis, feature selection, image restoration/compression, and neural coding. Unfortunately, the required  optimization is typically NP-hard, and so approximate procedures that succeed with high probability are sought.