A Keras example on consistency regularization

Hi folks,

I hope you are doing well. I wanted to let you know about my new example on keras.io:

It shows how to perform consistency regularization to improve the performance of vision models under distribution shifts. It also provides a template for conducting semi-supervised / weakly supervised learning.

With this minimal example, I got promising improvements on CIFAR-10-C:


My full-scale experiments are here. As always, happy to address feedback.