Iterative thresholding algorithm based on non-convex method for modified lp-norm regularization minimization

Angang Cui,Jigen Peng,Haiyang Li,Meng Wen,Jiajun Xiong

DOI: https://doi.org/10.48550/arXiv.1804.09385

2018-04-25

Abstract:Recently, the $ł_{p}$-norm regularization minimization problem $(P_{p}^{\lambda})$ has attracted great attention in compressed sensing. However, the $ł_{p}$-norm $\|x\|_{p}^{p}$ in problem $(P_{p}^{\lambda})$ is nonconvex and non-Lipschitz for all $p\in(0,1)$, and there are not many optimization theories and methods are proposed to solve this problem. In fact, it is NP-hard for all $p\in(0,1)$ and $\lambda>0$. In this paper, we study two modified $ł_{p}$ regularization minimization problems to approximate the NP-hard problem $(P_{p}^{\lambda})$. Inspired by the good performance of Half algorithm and $2/3$ algorithm in some sparse signal recovery problems, two iterative thresholding algorithms are proposed to solve the problems $(P_{p,1/2,\epsilon}^{\lambda})$ and $(P_{p,2/3,\epsilon}^{\lambda})$ respectively. Numerical results show that our algorithms perform effectively in finding the sparse signal in some sparse signal recovery problems for some proper $p\in(0,1)$.

Optimization and Control