Abstract

We study kernel quadrature rules with positive weights for probabilitymeasures on general domains. Our theoretical analysis combines the spectralproperties of the kernel with random sampling of points. This results ineffective algorithms to construct kernel quadrature rules with positive weightsand small worst-case error. Besides additional robustness, our numericalexperiments indicate that this can achieve fast convergence rates that competewith the optimal bounds in well-known examples.

 

Quick Read (beta)