AG Kommunikationstheorie


Second-Order Hard Thresholding Algorithms for Sparse Recovery.


In this talk, we propose two variants of the well-known Iterative Hard Thresholding (IHT) and Hard Thresholding Pursuit (HTP) algorithms for the reconstruction of sparse vectors from subsampled measurements. While both algorithms aim at improving a smooth data fidelity measure at each iteration via a gradient-descent step followed by a projection onto the set of k-sparse vectors, the proposed algorithms update the current iterate based on a second-order method. Barring the calculation of the Moore-Penrose pseudoinverse of the sensing matrix, both modifications incur the same computational complexity as their original versions. After presenting the IHT and HTP algorithms in detail and motivating the proposed change of their respective updating rules, some preliminary numerical simulation results are presented which demonstrate that both variants significantly outperfom the original algorithms over a wide range of parameters in a Gaussian setting.

zurück zur Terminübersicht