This repository to implements the framework described by Papernot Et al. [1], for training differentially private machine learning models. The framework only applies to a select few types, those that solve the classification problem for only a few classes (< 100).
For those interested in a less academic description of the paper, Nicolas Papernot and Ian Goodfellow (both authors of the paper) did a details blog post on the paper and the relationship between differential privacy and machine learning.
DISCLAIMER:
- THIS REPOSITORY IS STILL A WORK IN PROGRESS
- As the author of the repository I take no responsibility for any use or the correctness of this code. Be careful when working with private data.