An Introduction to Model Building with Reproducing Kernel Hilbert Spaces

Notes for a shortcourse given at Interface 2000. We assume no knowledge of reproducing kernel Hilbert spaces, but review some basic concepts, with a view towards demonstrating how this setting allows the building of interesting statistical models that allow the simultaneous analysis of heterogenous, scattered observations, and other information. The abstract ideas will be illustrated with several specific data analyses, including modeling risk factors for eye diseases.