A Note on the Complexity of Proximal Iterative Hard Thresholding Algorithm

The iterative hard thresholding(IHT)algorithm is a powerful and efficient algorithm for solving l_(0)-regularized problems and inspired many applications in sparse-approximation and image-processing fields.Recently,some convergence results are established for the proximal scheme of IHT,namely proximal iterative hard thresholding(PIHT)algorithm(Blumensath and Davies,in J Fourier Anal Appl 14:629–654,2008;Hu et al.,Methods 67:294–303,2015;Lu,Math Program 147:125–154,2014;Trzasko et al.,IEEE/SP 14th Workshop on Statistical Signal Processing,2007)on solving the related l_(0)-optimization problems.However,the complexity analysis for the PIHT algorithm is not well explored.In this paper,we aim to provide some complexity estimations for the PIHT sequences.In particular,we show that the complexity of the sequential iterate error is at o(1/k).Under the assumption that the objective function is composed of a quadratic convex function and l_(0)regularization,we show that the PIHT algorithm has R-linear convergence rate.Finally,we illustrate some applications of this algorithm for compressive sensing reconstruction and sparse learning and validate the estimated error bounds.

Geophysical data sparse reconstruction based on L0-norm minimization

Missing data are a problem in geophysical surveys, and interpolation and reconstruction of missing data is part of the data processing and interpretation. Based on the sparseness of the geophysical data or the transform domain, we can improve the accuracy and stability of the reconstruction by transforming it to a sparse optimization problem. In this paper, we propose a mathematical model for the sparse reconstruction of data based on the LO-norm minimization. Furthermore, we discuss two types of the approximation algorithm for the LO- norm minimization according to the size and characteristics of the geophysical data： namely, the iteratively reweighted least-squares algorithm and the fast iterative hard thresholding algorithm. Theoretical and numerical analysis showed that applying the iteratively reweighted least-squares algorithm to the reconstruction of potential field data exploits its fast convergence rate, short calculation time, and high precision, whereas the fast iterative hard thresholding algorithm is more suitable for processing seismic data, moreover, its computational efficiency is better than that of the traditional iterative hard thresholding algorithm.

Image reconstruction from few views by l_0-norm optimization

In the medical computer tomography （CT） field, total variation （TV）, which is the l1-norm of the discrete gradient transform （DGT）, is widely used as regularization based on the compressive sensing （CS） theory. To overcome the TV model＇s disadvantageous tendency of uniformly penalizing the image gradient and over smoothing the low-contrast structures, an iterative algorithm based on the l0-norm optimization of the DGT is proposed. In order to rise to the challenges introduced by the l0-norm DGT, the algorithm uses a pseudo-inverse transform of DGT and adapts an iterative hard thresholding （IHT） algorithm, whose convergence and effective efficiency have been theoretically proven. The simulation demonstrates our conclusions and indicates that the algorithm proposed in this paper can obviously improve the reconstruction quality.

Improved RIP-based bounds for guaranteed performance of two compressed sensing algorithms

Iterative hard thresholding(IHT)and compressive sampling matching pursuit(CoSaMP)are two mainstream compressed sensing algorithms using the hard thresholding operator.The guaranteed performance of the two algorithms for signal recovery was mainly analyzed in terms of the restricted isometry property(RIP)of sensing matrices.At present,the best known bound using the RIP of order 3k for guaranteed performance of IHT(with the unit stepsize)isδ3k<1/√3≈0.5774,and the bound for CoSaMP using the RIP of order 4k isδ4k<0.4782.A fundamental question in this area is whether such theoretical results can be further improved.The purpose of this paper is to affirmatively answer this question and to rigorously show that the abovementioned RIP bound for guaranteed performance of IHT can be significantly improved toδ3k<(√5−1)/2≈0.618,and the bound for CoSaMP can be improved toδ4k<0.5102.