Evaluation of the Fractional Approach for Iterative Algorithms in Compressed Sensing

Abstract:
In order to converge, algorithms for inference of any kind may require stabilization in difficult scenarios. For the expectation propagation (EP) framework, a method called fractional EP has been introduced for a stabilization purpose. In compressed sensing (CS) this approach is not very well known. This paper contains a thorough introduction and discussion of the fractional approach for CS. The effect of the stabilization procedure is examined and displayed in numerical simulations. Furthermore, we show in numerical simulations that a combination of the fractional approach and damping can be beneficial for the performance of the algorithm.