jax-privacy - v1.0.0 Jax Privacy Release

First official release of Jax Privacy.

Jax Privacy library now includes: - Full support for DP-SGD training with Jax, Flax Linen and Keras - E2E examples and Jupyter notebooks showing how to use the library - Algorithms for DP-FTRL - DP accounting for ML (e.g. calculating stddev of noise given eps, delta and other training parameters)

Experimental folder contains features that will be part of the next release: - Refactoring of the API making it more streamlined and easier to use - Poisson sampling

Full Changelog: https://github.com/google-deepmind/jax_privacy/compare/v0.2...v1.0.0

- Python
Published by RamSaw 8 months ago