Can we have extension of Mercer theorem to interpolation?

71 Views Asked by At

This question is related to Mercer theorem, Reproducible kernel Hilbert space(RKHS) and interpolation. The wikipedia links are https://en.wikipedia.org/wiki/Mercer%27s_theorem and https://en.wikipedia.org/wiki/Reproducing_kernel_Hilbert_space. As I understand Mercer theorem confirms the existence of RKHS with appropriate kernel. My understanding is that the kernel takes two vector and returns a scalar. Therefore RKHS is nice for classification type application. My question is can we extend the idea to interpolation? In other words if the 'kernel' can take a matrix and a vector it would return a function or a vector. Can we have a similar RKHS space for this case? Is this a valid question? In case it is not, would you please let me know what is the problem in this view?