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: v0.2...v1.0.0