Ziyue Huang, Yuting Liang, and Ke Yi. Instance-optimal mean estimation under differential privacy (NeurIPS 2021).
| Folder | Description |
|---|---|
| data | contains MNIST data |
| lpme | implementation for the methods in locally private mean estimation |
| coinpress | a copy of the code from https://github.com/twistedcubic/coin-press, containing implementation of coinpress |
| quantile_binary_search | implementation of our methods |
numpy v1.23.5
scipy v1.9.3
torch v2.1.0+cpu
joblib v1.3.2
The joblib library is used for computing some functions on different coordinates of the data in parallel (para='0'). Alternatively, the multiprocessing library can be used (para='1'), or sequential computations can be used (para='2').
To reproduce the experiments in the central model (Fig. 1-8), run:
python central_tests.py
To reproduce the experiments in the local model (Fig. 10-12), run:
python local_tests.py
To reproduce the study on the clipping threshold (Fig. 9), run:
python run_syn_qt.py