Journal of Frontiers of Computer Science and Technology ›› 2009, Vol. 3 ›› Issue (4): 368-377.DOI: 10.3778/j.issn.1673-9418.2009.04.004

• 学术研究 • Previous Articles     Next Articles

GPU-based Parallel SVM Algorithm

DO Thanh-Nghi1, NGUYEN Van-Hoa2, POULET François3+   

  1. 1. College of Information Technology, Can Tho University, Can Tho, Vietnam
    2. INRIA Rennes, Campus de Beaulieu, 35042 Rennes Cedex, France
    3. Université de Rennes I, IRISA, Campus de Beaulieu, 35042 Rennes Cedex, France
  • Received:1900-01-01 Revised:1900-01-01 Online:2009-07-15 Published:2009-07-15
  • Contact: DO Thanh-Nghi

GPU的并行支持向量机算法

DO Thanh-Nghi1, NGUYEN Van-Hoa2, POULET François3+   

  1. 1. College of Information Technology, Can Tho University, Can Tho, Vietnam
    2. INRIA Rennes, Campus de Beaulieu, 35042 Rennes Cedex, France
    3. Université de Rennes I, IRISA, Campus de Beaulieu, 35042 Rennes Cedex, France
  • 通讯作者: DO Thanh-Nghi

Abstract: A new parallel and incremental support vector machine (SVM) algorithm for the classification of very large datasets on graphics processing units (GPUs) is presented. SVM and kernel related methods have shown to build accurate models but the learning task usually needs a quadratic program so that this task for large datasets requires large memory capacity and long time. A recent least squares SVM (LS-SVM) proposed by Suykens and Vandewalle for building incremental and parallel algorithm is extended. The new algorithm uses graphics processors to gain high performance at low cost. Numerical test results on UCI and Delve dataset repositories show that this para-llel incremental algorithm using GPUs is about 130 times faster than its CPU implementation and often significantly faster (over 2 500 times) than state-of-the-art algorithms like LibSVM, SVM-perf and CB-SVM.

Key words: support vector machine (SVM), general purpose graphics processing units (GP-GPU), least squares SVM (LS-SVM)

摘要: 提出了一种新的并行增量式支持向量机算法来解决图形处理单元(GPU)中大规模数据集的分类问题。SVM 以及核相关方法可以用来创建精确分类模型,但学习过程需要大量内存和很长时间。扩展了Suykens和Vandewalle提出的最少次方SVM(LS-SVM)方法来建立增量和并行算法。新算法使用图形处理器以低代价获得高系统性能。实现表明,在UCI和Delve数据集上,基于GPU并行增量算法较CPU实现方法快130倍,而且比现行算法,如LibSVM、SVM-perf 和CB-SVM等快的多(超过2500倍)。

关键词: 支持向量机, 图形处理器, 最少次方SVM