Does Label Differential Privacy Prevent Label Inference Attacks?

This repo contains the official implementation for the paper Does Label Differential Privacy Prevent Label Inference Attacks?. In this paper, we analyze in depth the connection between label differential privacy and label inference attacks with both theoretical and empirical results.

Environment Setup

To install required packages, run the command below (we recommend python 3.7 and pytorch 1.7):

pip install -r requirements.txt

Folder Structure

criteo/
    data/
    LP-MST/
    PATE/
image/
    LP-MST/
    PATE/
simulation/

image/ contains experiment code for Section 3.

simulation/ contains experiment code for Section 5.1.

criteo/ contains experiment code for Section 5.2.

Please see individual folder for more information.

Citation

If you find this code useful in your research, please consider citing:

@inproceedings{wu2022does,
  title={Does Label Differential Privacy Prevent Label Inference Attacks?},
  author={Wu, Ruihan and Zhou, Jin Peng and Weinberger, Kilian Q and Guo, Chuan},
  booktitle={International Conference on Artificial Intelligence and Statistics},
  year={2023}
}