In this lecture, Prof. Dr. Bernard Haasdonk will give a motivation of kernel methods for pattern analysis, machine learning and function approximation. Then we will mainly concentrate on kernels, their properties, functional analytic interpretation in feature spaces, and provide kernel construction principles. In particular kernels are an elegant way to encode problem-specific prior knowledge, e.g. about invariances.
A subsequent ML-session unit in the next semester will focus on a variety of specific kernel methods for data analysis and machine learning.
The next session will take place:
Feb 5, 2019, 2pm: Gaussian Processes (M. Toussaint)