Fast Bayesian Matching Pursuit (FBMP) is an algorithm that rapidly performs Bayesian model averaging and minimum mean squared error (MMSE) estimation in the context of sparse linear regression. It can be readily applied to a wide variety of compressive sensing and sparse reconstruction problems. For regression tasks where model selection is the principal goal, FBMP's Bayesian framework allows it to provide the user with a set of high posterior probability models, rather than presenting a single maximum a posteriori (MAP) model as the only candidate model. For users who wish to perform sparse signal reconstruction, FBMP is able to offer approximate MMSE estimates of sparse signal vectors, yielding recoveries with lower mean squared error than MAP-based recoveries, when there is ambiguity regarding the true model.
*All authors are with The Ohio State University, Department of Electrical and Computer Engineering