基于K最近邻的支持向量机快速训练算法

传统支持向量机训练大规模样本时间和空间开销大,使其应用受到了很大限制。为了提高支持向量机的训练速度,根据支持向量机的基本原理,应用K最近邻思想来筛选训练样本集,提出了基于K最近邻的支持向量机快速训练算法(KNN-SVM)。算法首先选取一部分最有可能成为支持向量的样本——边界向量,然后用边界向量集代替训练样本集进行支持向量机训练,大幅度减少了训练样本的数量,使支持向量机的训练速度显著提高。同时,由于边界向量包含了支持向量,因此,支持向量机的分类能力没有受到影响。仿真实验结果表明,与传统支持向量机相比,在分类精度相同的情况下,算法能够有效地提高支持向量机的训练速度,而且还可以提高支持向量机的分类速度和推广能力。

A fast training algorithm for support vector machines based on K nearest neighbors

It is very costly in terms of time and memory consumption to process the large training sample using traditional Support Vector Machines(SVM),thus the application of SVM is limited severely.In order to improve the training speed of SVM,a fast training algorithm for SVM is presented based on K Nearest Neighbors(KNN-SVM) according the fundamental principle of SVM,which selects training sample by applying the idea of K nearest neighbors.At first,the algorithm extracts some samples that are most likely to become support vectors,the border vectors,from the training samples,and trains SVM by substituting the border vector set for training set.The method reduces training samples greatly and advances training speed markedly.The classification capability of SVM to is not affected because the border vectors contain support vectors.Experiment results showed that improved algorithm enhances the speed of training SVM effectively comparing with conventional SVM under the same classification precision.Furthermore,the algorithm may improve the speed for classification and generalization ability of SVM.