Review for NeurIPS paper: Differentiable Augmentation for Data-Efficient GAN Training

The contribution is not enough. This paper address overfitting problem of training GAN with limited data, and proposed the differentiable augmentation. I think it is important factor, but still limited. The generated image with limited diversity is also leveraged to train the discriminator (classifier). In this case, the learned discriminator is not suitable to evaluate the validation data.