First Order Generative Adversarial Networks
This addresses a fundamental optimization issue in GANs for researchers and practitioners, though it is incremental as it builds on existing GAN methods.
The paper tackles the problem of suboptimal update directions in GANs by introducing a theoretical framework and a novel divergence that ensures unbiased steepest descent updates, achieving state-of-the-art results on language generation tasks like One Billion Word.
GANs excel at learning high dimensional distributions, but they can update generator parameters in directions that do not correspond to the steepest descent direction of the objective. Prominent examples of problematic update directions include those used in both Goodfellow's original GAN and the WGAN-GP. To formally describe an optimal update direction, we introduce a theoretical framework which allows the derivation of requirements on both the divergence and corresponding method for determining an update direction, with these requirements guaranteeing unbiased mini-batch updates in the direction of steepest descent. We propose a novel divergence which approximates the Wasserstein distance while regularizing the critic's first order information. Together with an accompanying update direction, this divergence fulfills the requirements for unbiased steepest descent updates. We verify our method, the First Order GAN, with image generation on CelebA, LSUN and CIFAR-10 and set a new state of the art on the One Billion Word language generation task. Code to reproduce experiments is available.