Fast (and cheeky) differentially private gradient-based optimisation in PyTorch
Project description
deepee
deepee is a library for differentially private deep learning in PyTorch. More precisely, deepee implements the Differentially Private Stochastic Gradient Descent (DP-SGD) algorithm originally described by Abadi et al.. Despite the name, deepee works with any (first order) optimizer, including Adam, AdaGrad, etc.
It wraps a regular PyTorch model and takes care of calculating per-sample gradients, clipping, noising and accumulating gradients with an API which closely mimics the PyTorch API of the original model.
Check out the documentation here
For paper readers
If you would like to reproduce the results from our paper, please go here
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
