pull down to refresh

It has nothing to do w/ copyright, the issue is the method used to train the model. Details in the paper.