Ian's GAN list 02/2018

Ians_gans_02_2018.md

The list by @goodfellow_ian, as per Twitter discussion with @timnitGebru: https://twitter.com/timnitGebru/status/968242968007200769

1. Progressive GANs:
   • Video: https://www.youtube.com/watch?v=XOxxPcy5Gr4
   • Link: https://arxiv.org/abs/1710.10196
   • Description: Probably the highest quality images so far
2. Spectral normalization:
   • Link: https://openreview.net/forum?id=B1QRgziT-&noteId=BkxnM1TrM
   • Description: Got GANs working on lots of classes, which has been hard
3. Projection discriminator
   • Link: https://openreview.net/forum?id=ByS1VpgRZ
   • Description: from the same lab as #2, both techniques work well together, overall give very good results with 1000 classes
4. pix2pixHD
   • Link (including video): https://github.com/NVIDIA/pix2pixHD
   • Description: GANs for 2-megapixel video
5. Are GANs created equal?
   • Link: https://arxiv.org/abs/1711.10337
   • Description: A big empirical study showing the importance of good rigorous empirical work and how a lot of the GAN variants don't seem to actually offer improvements in practice
6. WGAN-GP:
   • Link: https://arxiv.org/abs/1704.00028
   • Description: probably the most popular GAN variant today and seems to be pretty good in my opinion. Caveat: the baseline GAN variants should not perform nearly as badly as this paper claims, especially the text one
7. StackGAN++:
   • Link: https://arxiv.org/abs/1710.10916
   • Description: High quality text-to-image synthesis with GANs
8. Making all ML algorithms differentially private by training them on fake private data generated by GANs
   • Link: https://www.biorxiv.org/content/early/2017/07/05/159756
   • Description: Privacy-preserving generative deep neural networks support clinical data sharing
9. "GANs with encoders", one of my favorites:
   • Link: https://arxiv.org/abs/1701.04722
10. "theory of GAN convergence", one of my favorites:
    • Link: https://arxiv.org/abs/1706.04156