Nyström landmark sampling and regularized Christoffel functions

Selecting diverse and important items, called landmarks, from a large set is problem of interest in machine learning. As specific example, order to deal with training sets, kernel methods often rely on low rank matrix Nyström approximations based the selection or sampling landmarks. In this context, we propose deterministic randomized adaptive algorithm for selecting landmark points within data set. These landmarks are related minima sequence kernelized Christoffel functions. Beyond known connection between functions leverage scores, our method finite determinantal point processes (DPPs) also explained. Namely, construction promotes diversity among way similar DPPs. Also, explain how can influence accuracy Kernel Ridge Regression.