Fixing GAN optimization through competitive gradient descent - Anima Anandkumar