Skip to content

twistedcubic/coin-press

Repository files navigation

CoinPress: Practical Private Mean and Covariance Estimation

A Python implementation of CoinPress: Practical Private Mean and Covariance Estimation. CoinPress stands for COnfidence-INterval-based PRivate EStimation Strategy. Published at NeurIPS 2020.

Instructions

We include a demo for both mean and covariance estimation, the parameters for which can be specified by either command-line arguments, or inside the python script.

For instance to run the covariance and mean estimation demo for a synthetic dataset with 3000 10-dimensional datapoints and total privacy budget 0.5:

python demo.py --n 3000 --d 10 --total_budget 0.5

To see usage details one can run:

python demo.py --h

For more examples please see mean_estimation.py and cov_estimation.py. More complete experiment details can be found in multivariate_covariance_experiments.ipynb and multivariate_mean_experiments.ipynb, calling on core functions from algos.py and utils.py. plot_mean.py and plot_cov.py produce the plots as shown in the paper, using data included in ./results. This data can be regenerated by running the aforementioned scripts or Jupyter notebooks. Exceptions are the files prefixed by dfmbg in ./results/synthetic_mean/, which are generated using code from Du et al's repository.

To run the map of Europe experiments (included our covariance experiments notebook), files from the following are required:

Reference

This repository is an implementation of our paper CoinPress: Practical Private Mean and Covariance Estimation, authored by Sourav Biswas, Yihe Dong, Gautam Kamath, Jonathan Ullman. Code contributed by all four authors.

If you use our code or paper, we ask that you please cite:

@incollection{BiswasDKU20,
  title         = {CoinPress: Practical Private Mean and Covariance Estimation},
  author        = {Biswas, Sourav and Dong, Yihe and Kamath, Gautam and Ullman, Jonathan},
  booktitle = {Advances in Neural Information Processing Systems 33},
  url       = {arXiv preprint arXiv:2006.06618},
  year          = {2020}
}

About

[NeurIPS 2020] Simple and practical private mean and covariance estimation.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published