Abstract: In this talk, we will focus on the relations between RKHS and probability measures on Rd. We will be interested in kernels with a translation invariant Maximum Mean Discrepancy (MMD) which allows unbounded kernels. We will use the fact that these kernels k are not necessarily bounded to consider a stronger distance than usual, the Wasserstein distance W1.

Keywords: RKHS, MMD, Wasserstein distance