The proposed test is distribution-free in finite samples. While most existing two-sample tests perform poorly or are even inapplicable to high-dimensional data, our test can be conveniently used in ...