East Asian J. Appl. Math., 12 (2022), pp. 406-420.
Published online: 2022-02
Cited by
- BibTex
- RIS
- TXT
A fast block coordinate descent method for solving linear least-squares problems is proposed. The method is based on a greedy criterion of the column selection used at each iteration. It is proven that if the coefficient matrix of the corresponding system has full column rank, the method converges to the unique solution of the linear least-squares problem. Numerical experiments show the advantage of this approach over similar methods in terms of CPU time and computational cost, does not matter whether the coefficient matrix is of full column rank or not.
}, issn = {2079-7370}, doi = {https://doi.org/10.4208/eajam.160721.160122}, url = {http://global-sci.org/intro/article_detail/eajam/20261.html} }A fast block coordinate descent method for solving linear least-squares problems is proposed. The method is based on a greedy criterion of the column selection used at each iteration. It is proven that if the coefficient matrix of the corresponding system has full column rank, the method converges to the unique solution of the linear least-squares problem. Numerical experiments show the advantage of this approach over similar methods in terms of CPU time and computational cost, does not matter whether the coefficient matrix is of full column rank or not.