In the past, when I read Generative Adversarial Network research, I found a veritable zoo of approaches to regularizing the discriminator, such as Least-Squares-GAN, Wasserstein-GAN, Wasserstein-Gradient-Penalty, and Instance Noise. However, in the past few years I’ve noticed that the simple gradient penalty labelled ‘R1’ suddenly became so dominant that papers using it don’t draw attention to it, they just say ‘R1 regularization’ and cite this work.