Timezone: »

First Order Generative Adversarial Networks
Calvin Seward · Thomas Unterthiner · Urs M Bergmann · Nikolay Jetchev · Sepp Hochreiter

Thu Jul 12 09:15 AM -- 12:00 PM (PDT) @ Hall B #37

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.

Author Information

Calvin Seward (Zalando Research)
Thomas Unterthiner (Johannes Kepler University Linz)
Urs M Bergmann (Zalando Research)
Nikolay Jetchev (Zalando Research)

PhD in Robotics and Machine learning. Research Scientist at Zalando with focus on generative models, computer vision, probabilistic time series modeling.

Sepp Hochreiter (Johannes Kepler University Linz)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors