A Gauss-Newton Approach for Min-Max Optimization in Generative Adversarial Networks