A Novel Algorithm to Train Support Vector Machine Based on Convex Optimization Theory

Liang-zhi GAN, Mi HE

Abstract


A novel algorithm to train support vector machine was proposed. This algorithm was based on the convex optimization theory. With the help of the theory, the training samples were divided into two classes: the possible support vectors and the impossible support vectors. In linear separable case, the impossible vectors do not affect the training results of support vector machine and hence can be removed from the training samples. Linear inseparable training data in low dimensional space are linear separable in high dimensional space. By this way, the number of training samples is greatly reduced and training support vector machine is simplified.

Keywords


Statistical learning theory, Support vector machine, Convex optimization theory


DOI
10.12783/dtssehs/emass2018/20446

Full Text:

PDF