An enhanced disprover algorithm which aims to resolve the limitations in counterexample construction of the latest work DP-Sniper and produce higher privacy violations.
pip install -r requirements.txtWe used python version 3.8 to run our evaluations.
To run the evaluation, run the following command:
python dpopt/experiments/__main__.pyThe main algorithms DP-Opt and PowerSearcher from the paper can be found in dpopt/search/powersearcher.py and
This is the implementation of the algorithms from the following research paper:
Niu B, Zhou Z, Chen Y, et al. DP-Opt: Identify High Differential Privacy Violation by Optimization[C]// L. Wang, M. Segal, J. Chen, et al. Wireless Algorithms, Systems, and Applications, Cham: Springer Nature Switzerland, 2022: 406–416.
MIT License, see LICENSE.
This repository is derived from DP-Sniper, marked as MIT License, Copyright (c) 2021 SRI Lab, ETH Zurich.