Binarized MNIST

We use the preprocessed binarized MNIST dataset from [49] which has a split of 50k/10k/10k.

BibTex: