基于粒子群算法与图形处理器加速的支持向量机参数优化方法
Parameter Optimization of SVM Based on Particle Swarm Optimization Algorithm and GPU Acceleration
Abstract
支持向量机(SuPPOrT VECTOr MACHInE,SVM)的参数选择对其性能有着重要的影响,使用穷举法优化参数需要大量的计算时间.为快速寻找最优参数组合,利用粒子群算法(PArTIClE SWArM OPTIMIzATIOn,PSO)收敛速度快、简单易行等特点,将SVM参数作为粒子的解决方案.并利用图形处理器(grAPHICS PrOCESSIng unIT,gPu)并行化处理能力计算每个参数的分类准确率,从而提升了在一定的搜索空间内寻找最佳参数组合的计算速度.对uCI数据进行实验,对比结果显示,该方法能快速有效地获取优化结果. The parameter selection of the support vector machine(SVM)has significant impact on its performance.Brute-force method for finding the optimal parameters is time consuming.In this paper,we are making use of particle swarm optimization(PSO)′s character of fast convergence speed and could be implemented easily,adding the SVM parameters as the solution of the particles.And exploit the computing capability of the graphics processing unit(GPU)to calculate the classification accuracy of each parameter,thus enhancing the computing speed for finding the best parameter combination in a constraint solutions space.The comparison results on UCI data show that this method can obtain the optimal results quickly and efficiently.