摘要
针对光滑孪生支持向量回归机(Smooth Twin Support Vector Regression,STSVR)中Sigmoid函数逼近精度不高的问题,将正号函数展开为无穷多项式级数,由此得到一族光滑函数.采用该多项式光滑函数逼近孪生支持向量回归机的不可微项,并用Newton-Armijo算法求解相应的模型,提出了多项式光滑孪生支持向量回归机(Polynomial Smooth Twin Support Vector Regression,PSTSVR).不仅从理论上证明了PSTSVR的收敛性和满足任意阶光滑的性能,而且在人工数据集和UCI数据集上的实验表明了PSTSVR比STSVR具有更好的回归性能.
Sigmoid function is used as the smoothing function in Smooth Twin Support Vector Regression （STSVR）. However, the approximation accuracy of Sigmoid function is low. In order to overcome this problem, in this paper, a new smooth twin support vector regression, term as Polynomial Smooth Twin Support Vector Regression （PSTSVR）, is proposed. In PSTSVR, plus function is transformed to an infinite polynomial series. Thus a family of smoothing functions is derived. Polynomial function is used to approximate the non-differential term of twin support vector regression. Then Newton-Armijo algorithm is used to solve the corresponding model. We have proved that PSTSVR is not only convergent, but also can meet the arbitrary order smooth performance. Meanwhile, the experimental results on several artificial and benchmark datasets show that PSTSVR has better regression performance than STSVR.
出处
《微电子学与计算机》
CSCD
北大核心
2013年第10期5-8,共4页
Microelectronics & Computer
基金
国家重点基础发展计划(2013CB329502)
国家自然科学基金项目(41074003)
