Graphics Processing Units and High-Dimensional Optimization
Zhou, Hua ; Lange, Kenneth ; Suchard, Marc A.
Statist. Sci., Tome 25 (2010) no. 1, p. 311-324 / Harvested from Project Euclid
This article discusses the potential of graphics processing units (GPUs) in high-dimensional optimization problems. A single GPU card with hundreds of arithmetic cores can be inserted in a personal computer and dramatically accelerates many statistical algorithms. To exploit these devices fully, optimization algorithms should reduce to multiple parallel tasks, each accessing a limited amount of data. These criteria favor EM and MM algorithms that separate parameters and data. To a lesser extent block relaxation and coordinate descent and ascent also qualify. We demonstrate the utility of GPUs in nonnegative matrix factorization, PET image reconstruction, and multidimensional scaling. Speedups of 100-fold can easily be attained. Over the next decade, GPUs will fundamentally alter the landscape of computational statistics. It is time for more statisticians to get on-board.
Publié le : 2010-08-15
Classification:  Block relaxation,  EM and MM algorithms,  multidimensional scaling,  nonnegative matrix factorization,  parallel computing,  PET scanning
@article{1294167962,
     author = {Zhou, Hua and Lange, Kenneth and Suchard, Marc A.},
     title = {Graphics Processing Units and High-Dimensional Optimization},
     journal = {Statist. Sci.},
     volume = {25},
     number = {1},
     year = {2010},
     pages = { 311-324},
     language = {en},
     url = {http://dml.mathdoc.fr/item/1294167962}
}
Zhou, Hua; Lange, Kenneth; Suchard, Marc A. Graphics Processing Units and High-Dimensional Optimization. Statist. Sci., Tome 25 (2010) no. 1, pp.  311-324. http://gdmltest.u-ga.fr/item/1294167962/