Is it meaningful to compare the Maximum Mean Discrepancy (MMD) values on different distributions? For example, if I have 3 distributions: A, B and C (all three are multivariate), can I compare MMD(A,B) with MMD(A,C) to determine if distribution B is more similar to A than C?
If not, why not? and which distribution distance metric can I use to make such conclusion?
Thanks a million in advance!