Paper: Augenstein, Sean, H. Brendan McMahan, Daniel Ramage, Swaroop Ramaswamy, Peter Kairouz, Mingqing Chen, Rajiv Mathews, and Blaise Aguera y Arcas. 2019. “Generative Models for Effective ML on Private, Decentralized Datasets”
./demos/federated-gan-training-loop.ipynb
: training a federated GAN on Colab./demos/ablation-results.ipynb
: visualizing the runs
-
dp_l2_norm_clip
: 0.1,dp_noise_multiplier
: 0.01 -
dp_l2_norm_clip
: 0.08036,dp_noise_multiplier
: 0.00555 -
dp_l2_norm_clip
: 0.105667,dp_noise_multiplier
: 0.001130