求解大型线性最小二乘问题的贪婪随机坐标下降法  

A Greedy Randomized Coordinate Descent Method for Solving Large Linear Least Squares Problem

作　　者：董勤 

机构地区：[1]温州大学数理学院,浙江 温州

出　　处：《应用数学进展》2024年第6期2780-2790,共11页Advances in Applied Mathematics

摘　　要：贪婪随机坐标下降法(GRCD)是求解大型线性最小二乘问题的有效迭代方法之一。本文在GRCD算法中引入松弛因子,构造了一种含参数的贪婪随机坐标下降法。并证明了当线性最小二乘问题的系数矩阵为列满秩时该方法依期望的收敛性。数值实验表明,当选取适当的松弛因子时,该算法在迭代步数和计算时间比GRCD方法更有效。The greedy randomized coordinate descent method (GRCD) is one of the effective iterative methods to solve large linear least squares problem. A greedy randomized coordinate descent method with parameters was constructed by introducing a relaxation parameter in the GRCD algorithm. It is also proved that the method has the expected convergence when the coefficient matrix of the linear least squares problem is of full column rank. Numerical experiments show that the proposed algorithm is more effective than the GRCD method in terms of iterative steps and calculation time when the appropriate relaxation parameter is selected.

关 键 词：最小二乘问题 贪婪随机坐标下降法 松弛因子 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]