一种可减少训练时间的分层并行支持向量机方法

A Hierarchical and Parallel Support Vector Machines Algorithm for Reducing the Training Time

基于支持向量的本质和并行计算方法,提出了一种新的分层并行的机器学习方法以加速支持向量机的训练过 程.该方法首先按照分而治之的思想将原分类问题分成若干子问题,然后将支持向量机的训练过程分解成级联的两个层次, 在每层采用并行的方法训练各个子支持向量机.各层训练集中的非支持向量被逐步筛选掉,交叉合并的规则保证问题的一 致性.仿真结果表明该方法在保证分类器推广能力的同时,缩短了训练支持向量机的时间.

Based on the essence of support vectors and parallel algorithm, the paper proposes a novel strategy of filtering the training samples in a hierarchical and parallel way to speed up the training of support vector machines (SVMs). During the training process, the entire classification problem is divided into several small sub-problems that can be handled in a parallel way. Having hierarchically filtered out the non-support-vector data, we can obtain the final training data set, which is used to train a SVM that will be used as the final pattern classifier. In order to keep the consistency, the cross-combining principle is introduced. The simulation results illustrate that our method speeds up training while maintaining the generalization accuracy of SVMs.