%0 Generic %1 gulrajani2017improved %A Gulrajani, Ishaan %A Ahmed, Faruk %A Arjovsky, Martin %A Dumoulin, Vincent %A Courville, Aaron %D 2017 %K GAN discrete language training wasserstein %T Improved Training of Wasserstein GANs %U http://arxiv.org/abs/1704.00028 %X Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but can still generate low-quality samples or fail to converge in some settings. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to pathological behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.

@misc{gulrajani2017improved, abstract = {Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but can still generate low-quality samples or fail to converge in some settings. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to pathological behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.}, added-at = {2017-08-18T10:36:07.000+0200}, author = {Gulrajani, Ishaan and Ahmed, Faruk and Arjovsky, Martin and Dumoulin, Vincent and Courville, Aaron}, biburl = {https://www.bibsonomy.org/bibtex/27b75ae0614ef34d6f0139ea225f3a843/daschloer}, description = {Improved Training of Wasserstein GANs}, interhash = {ad4cf89e36acfc121df044f6284b1ed8}, intrahash = {7b75ae0614ef34d6f0139ea225f3a843}, keywords = {GAN discrete language training wasserstein}, note = {cite arxiv:1704.00028Comment: New CIFAR-10 and LSUN image generation experiments}, timestamp = {2017-10-04T16:28:59.000+0200}, title = {Improved Training of Wasserstein GANs}, url = {http://arxiv.org/abs/1704.00028}, year = 2017 }