|
Author(s): |
Xiaolei Su, Institute for Information and System Sciences & Ministry of Education Key Lab for Intelligent Networks and Network Security, Xi’an Jiaotong University, Xi’an, Shaanxi,710049, China
Hui Li, Institute for Information and System Sciences & Ministry of Education Key Lab for Intelligent Networks and Network Security, Xi’an Jiaotong University, Xi’an, Shaanxi,710049, China
Zongben Xu, Institute for Information and System Sciences & Ministry of Education Key Lab for Intelligent Networks and Network Security, Xi’an Jiaotong University, Xi’an, Shaanxi,710049, China
Jinshan Zeng, Institute for Information and System Sciences & Ministry of Education Key Lab for Intelligent Networks and Network Security, Xi’an Jiaotong University, Xi’an, Shaanxi,710049, China |
|
Abstract: |
The superiority of l1/2 regularization has been investigated on sparse optimization problems (particularly, on compressed sensing) in recent studies. The iterative half thresholding algorithm and the iterative reweighed l1 algorithm are two successful paradigms based on l1/2 regularization. However, both algorithms may suffer from getting trapped in local optima since l1/2 regularization leads to a multi-modal optimization problem. To overcome this weakness, we propose a global iterative half thresholding algorithm (HALF) for sparse optimization problems, denoted by HALF-EA, which incorporates iterative half thresholding algorithm into an evolutionary algorithm. Our experimental results demonstrate that HALF-EA is capable of recovering sparse signals with fewer measurements over the existing l1/2 algorithms.
|