摘要
梯度硬阈值追踪算法是求解稀疏优化问题的有效算法之一.考虑到算法中投影对最优解的影响,提出一种比贪婪策略更好的投影算法是很有必要的.针对一般的稀疏约束优化问题,利用整数规划提出一种迭代投影策略,将梯度投影算法中的投影作为一个子问题求解.通过迭代求解该子问题得到投影的指标集,并以此继续求解原问题,以提高梯度硬阈值追踪算法的计算效果.证明了算法的收敛性,并通过数值实例验证了算法的有效性.
The gradient hard threshold pursuit algoritlun is considered as one of the most,effective algorithms for solving sparse optimization problems.Since greedy projec-tion operator affects t he performance of the algorithm,it is desirable to propose a better projection algorithm besides the greedy strategics.In this paper,we first formulate the projection problem as an integer programming subproblem.Then,by iteratively solving the subproblem,we obtain a better support set of the sparse optimization problem,thus the original problem is solved to improve the performanee of the gradient,hard threshold pursuit algorithm.Finally,the convergence of the improved algorithm is proved,and the effectiveness of the algorithm is verified by numerical experiments.
作者
陈薪蓓
朱明康
陈建利
CHEN Xinbei;ZHU Mingkang;CHEN Jianli(Key Labwatory of Discrete Mathe-matics with.Applications Ministry of Education,Fuzhou University.Fuzhou 350108,China;Fuzhou No.l High School,Fuzhou 350116,China)
出处
《运筹学学报》
北大核心
2019年第1期1-14,共14页
Operations Research Transactions
基金
国家自然科学基金(No.A11501115)
福建省高校杰出青年人才培育计划(No.SX2016-21)
关键词
稀疏约束
整数规划
梯度硬阈值追踪
迭代投影
sparsity constraint
integer program
gradient hard thresholding pursuit algorithm
iterative projection
