A simple implementation of the Counterbalancing Teacher (CT) model.

For our CT model on MNIST with red and blue squers, run `train_mnist_ct_sup.py` to train and test on the four different test sets of RB, R, B, and None as explained in the paper.

For ERM models with/Without batch normalization, run `train_mnist_w_wo_bn_sup.py` and make sure to set `use_bn` flag to False to run ERM without batch normalization, by defult it will train and test ERM with batch normalization.


Requirements:

- Tensorflow 1.14
- Keras  2.2.4



