Title: The stability of SVMs
Authors: Daohong Xiang - Zhejiang Normal University (China) [presenting]
Abstract: In the practical application of kernel based methods like SVMs, practitioner often chooses model parameters, such as regularization parameter and scaling parameter of the kernels, empirically. It is natural to ask what would happen on SVMs estimator when the data points deviate from the structure of the data set. We will present many kernel based methods like SVMs have nice stability properties if simultaneously the distribution, the regularization parameter and the kernel change slightly.