Parallel Training Strategy Based on Support Vector Regression Machine

In this paper, we investigate the parallel training strategy and propose a parallel support vector regression machine algorithm that integrates model segmentation and data space decomposition. The major aim is to explore the new data space decomposition scheme that can solve computation intensive problem about the long time training based on SVR´s classification by using low-dimension algorithms. The strategy, which divides the whole task into several sub-tasks based on the sample division strategy, uses master-slave mode on the design of parallel program, and finally the master node produce a regression mode by collecting training results. The performance of this algorithm has been analyzed and evaluated with KDD99 data on the high-performance computer of ZQ3000 cluster. The results on this paper prove that the algorithm can guarantee the high precision in the regression and reduce the training time.