MMD-FUSE: Learning and Combining Kernels for Two-Sample Testing Without Data Splitting

Neural Information Processing Systems 

The fundamental problem of non-parametric two-sample testing consists in detecting the difference between any two distributions having access only to samples from these.