new support vector algorithms（新的支持向量算法）.pdfVIP

LETTER Communicated by John Platt New Support Vector Algorithms Bernhard Sch¨olkopf Alex J. Smola GMD FIRST, 12489 Berlin, Germany, and Department of Engineering, Australian National University, Canberra 0200, Australia Robert C. Williamson Department of Engineering, Australian National University, Canberra 0200, Australia Peter L. Bartlett RSISE, Australian National University, Canberra 0200, Australia We propose a new class of support vector algorithms for regression and classification. In these algorithms, a parameter lets one effectively con- trol the number of support vectors. While this can be useful in its own right, the parameterization has the additional benefit of enabling us to eliminate one of the other free parameters of the algorithm: the accuracy parameter in the regression case, and the regularization constant C in the classification case. We describe the algorithms, give some theoretical re- sults concerning the meaning and the choice of , and report experimental results. 1 Introduction Support vector (SV) machines comprise a new class of learning algorithms, motivated by results of statistical learning theory (Vapnik, 1995). Originally developed for pattern recognition (Vapnik Chervonenkis, 1974; Boser, Guyon, Vapnik, 1992), they represent the decision boundary in terms of a typically small subset (Sch¨olkopf, Burges, Vapnik, 1995) of all training examples, called the support vectors. In order for this sparseness property to carry over to the case of SV Regression, Vapnik devised the so-called -insensitive loss function, y f x max0 y f x (1.1) which does not penalize errors below some 0, chosen a priori. His algorithm, which we will henceforth call -SVR, seeks to estimate functions, f x w x b w x RN b R