Kernel Selection for Kernel Methods (KMs) and Support Vector Machines (SVMs)
Peter Haslinger
Peter Haslinger’s PhD, which is supervised by Dr. Ulrich Bodenhofer (Johannes Kepler University Linz), concentrates on the fundamental problem of Kernel Selection for Kernel Methods (KMs) and Support Vector Machines (SVMs). KMs and SVMs are a common tool for solving binary classification and regression problems. One of their main advantages is their good generalization aspect due to the underlying Vapnik-Chervonenkis theory. Particularly they have found widespread application in computer vision and image processing as a successful method for object detection due to their generalization behavior. Without doubt, one of the central open questions is still the lack of theory for an appropriate choice of the kernel. Interestingly in the field of applications only a few kernel classes have been investigated and implemented. In computer vision this is mainly the Gaussian Kernel, for which justification is still lacking. The goal of the PhD project is to study kernels for the purpose of gaining hints and criteria for their appropriate usage, especially in the field of computer vision applications.
back