CFE 2017: Start Registration
View Submission - CMStatistics
B0737
Title: Stability of pairwise learning methods Authors:  Andreas Christmann - University of Bayreuth (Germany) [presenting]
Abstract: Pairwise kernel learning methods are often used to solve ranking problems and other non-standard problems beyond classification and regression. It will be shown that many pairwise learning methods based on kernels are not only statistically robust with respect to small changes of the probability measure or the data set, but that such methods have additionally nice stability properties with respect to small changes of the regularization parameter and with respect to the kernel or its reproducing kernel Hilbert space.