Meta Two-Sample Testing: Learning Kernels for Testing with Limited Data

In realistic scenarios with very limited numbers of data samples, it can be challenging to identify a kernel powerful enough to distinguish complex distributions.