dpwgan icon indicating copy to clipboard operation
dpwgan copied to clipboard

Published 20 hours ago • verified publisher icon civisanalytics
→

Metadata

Differentially-private Wasserstein GAN implementation in PyTorch

Results 1 dpwgan issues
Sort by recently updated
recently updated
newest added

Differential privacy needs gradient clipping not weight clipping

In line 99 of dpwgan.py you clip discriminator weights "for privacy guarantees". Weight clipping is needed to ensure Lipschitz condition for Wasserstein GAN and not for privacy. For DP you...

MahedAb avatar MahedAb

← Metadata

26
Stars
17
Forks
Watchers

Owner

verified publisher icon civisanalytics

Metadata

Differentially-private Wasserstein GAN implementation in PyTorch

Back