RKHS中正则化学习算法的推广误差界

Generalization error bound for regularization-based learning algorithm in RKHS

研究了再生核希尔伯特空间(RKHS)中的正则化学习算法,证明了其推广误差可分解为两个部分:逼近误差和估计误差,并应用VC维和算法稳定性给出了相应界,最后联立这两个结果证明了正则化学习算法具有好的推广性.

Regularization-based learning algorithm in Reproducting Kernel Hilbert Space (RKHS) is studied.Its generalization error is proved to be decomposed into two parts:an approximation one and an estimation one. They are bounded respectively by VC dimension and algorithmic stability property. Finally,according to these two results,it can be proved that regularization-based learning algorithm has good generalization ability.