指数光滑支持向量分类机

Exponential smooth support vector machines

为了提高光滑支持向量机的分类性能,给出一种具有更强逼近正号函数能力的指数光滑函数,并利用光滑技术克服支持向量机模型的不可微性,得到指数光滑支持向量分类机(ESSVM)。通过理论分析,利用数学方法证明了指数光滑支持向量分类机的收敛性。数值实验表明,指数光滑支持向量机比多项式光滑支持向量机在分类性能上更有优势。

In order to improve smooth support vector machine classified performance,exponential smoothing function is introduced.The ability of exponential smoothing function to approximate plus function is stronger than those of existing smooth functions.Smooth technology is used to o-vercome the non-differentiable of support vector machine model and an Exponential Smooth Sup-port Vector Machine （ESSVM）is obtained by using the new smooth function.Theoretical and rigorous mathematical analyses are given to prove the convergence of Exponential Smooth Sup-port Vector Machine.Numerical experiments show that the Exponential smooth support vector machine has advantage on the classification performance than polynomial smooth support vector machine.