Tag: Towards a Better Understanding and Regularization of GAN Training Dynamics