From Majorization to Interpolation: Distributionally Robust Learning using Kernel Smoothing