JAX-Privacy: Algorithms for Privacy-Preserving Machine Learning in JAX

Installation | Reproducing Results | Citing

This repository contains the JAX implementation of algorithms that we develop in our research on privacy-preserving machine learning. This research code is open-sourced with the main objective of transparency and reproducibility, so (some) rough edges should be expected.

Installation

Note: to ensure that your installation is compatible with your local accelerators such as a GPU, we recommend to first follow the corresponding instructions to install TensorFlow and JAX.

Option 1: Static Installation

This option is preferred for the purpose of re-using functionalities of our codebase without modifying them. The package can be installed by running the following command-line:

pip install git+https://github.com/google-deepmind/jax_privacy

Option 2: Local Installation (Allowing Edits)

This option is preferred to either build on top of our codebase or to reproduce our results.

The first step is to clone the repository:

git clone https://github.com/google-deepmind/jax_privacy

Then the code can be installed so that local modifications to the code are reflected in imports of the package:

cd jax_privacy
pip install -e .

Reproducing Results

Unlocking High-Accuracy Differentially Private Image Classification through Scale

Instructions: experiments/image_classification.
arXiv link: https://arxiv.org/abs/2204.13650.
Bibtex reference: link.

Unlocking Accuracy and Fairness in Differentially Private Image Classification

Instructions: experiments/image_classification.
arXiv link: https://arxiv.org/abs/2308.10888.
Bibtex reference: link.

How to Cite This Repository

If you use code from this repository, please cite the following reference:

@software{jax-privacy2022github,
  author = {Balle, Borja and Berrada, Leonard and De, Soham and Ghalebikesabi, Sahra and Hayes, Jamie and Pappu, Aneesh and Smith, Samuel L and Stanforth, Robert},
  title = {{JAX}-{P}rivacy: Algorithms for Privacy-Preserving Machine Learning in JAX},
  url = {http://github.com/google-deepmind/jax_privacy},
  version = {0.3.0},
  year = {2022},
}

Acknowledgements

License

All code is made available under the Apache 2.0 License. Model parameters are made available under the Creative Commons Attribution 4.0 International (CC BY 4.0) License.

See https://creativecommons.org/licenses/by/4.0/legalcode for more details.

Disclaimer

This is not an official Google product.

jasonj2525/jax_privacy