3 ms·
This looks exciting, thank you for sharing! Implementing differential privacy correctly is famously difficult and easy to get wrong. Will you be making the priv
by TedTed 5y ago
This looks exciting, thank you for sharing! Implementing differential privacy correctly is famously difficult and easy to get wrong. Will you be making the privacy-critical part of your code available publicly for people to audit?
- ngrislain 5y agoYou are absolutely right, we are leveraging many open-source bricks to build our product, so that they can be reviewed, mainly: - https://github.com/google/differential-privacy https://github.com/google/differential-privacy (for basic mechanisms and PLD accounting) - https://github.com/tensorflow/privacy https://github.com/tensorflow/privacy (for DP-SGD and RDP accounting) - https://github.com/opendp https://github.com/opendp (for our SQL module) We actively contribute to some of them. We also open-sourced some tech bricks we are using: - https://github.com/sarus-tech/dp-xgboost https://github.com/sarus-tech/dp-xgboost (see also https://arxiv.org/pdf/2110.12770.pdf https://arxiv.org/pdf/2110.12770.pdf) We plan to continue building trust in the tools we are using by publishing some of them.