Measuring dataset similarity using optimal transport
Is FashionMNIST, a dataset of images of clothing items labeled by category, more similar to MNIST or to USPS, both of which are classification datasets of handwritten digits? This is a pretty hard question to answer, but the solution could have an impact on various…