Mean Estimation in the Add-Remove Model of Differential Privacy

Abstract

Differential privacy is often studied under two different models of neighboring datasets: the add-remove model and the swap model. While the swap model is frequently used in the academic literature to simplify analysis, many practical applications rely on the more conservative add-remove model, where obtaining tight results can be difficult. Here, we study the problem of one-dimensional mean estimation under the add-remove model. We propose a new algorithm and show that it is min-max optimal, achieving the best possible constant in the leading term of the mean squared error for all $\epsilon$, and that this constant is the same as the optimal algorithm under the swap model. These results show that the add-remove and swap models give nearly identical errors for mean estimation, even though the add-remove model cannot treat the size of the dataset as public information. We also demonstrate empirically that our proposed algorithm yields at least a factor of two improvement in mean squared error over algorithms frequently used in practice. One of our main technical contributions is a new hourglass mechanism, which might be of independent interest in other scenarios.

Cite

Text

Kulesza et al. "Mean Estimation in the Add-Remove Model of Differential Privacy." International Conference on Machine Learning, 2024.

Markdown

[Kulesza et al. "Mean Estimation in the Add-Remove Model of Differential Privacy." International Conference on Machine Learning, 2024.](https://mlanthology.org/icml/2024/kulesza2024icml-mean/)

BibTeX

@inproceedings{kulesza2024icml-mean,
  title     = {{Mean Estimation in the Add-Remove Model of Differential Privacy}},
  author    = {Kulesza, Alex and Suresh, Ananda Theertha and Wang, Yuyan},
  booktitle = {International Conference on Machine Learning},
  year      = {2024},
  pages     = {25642-25661},
  volume    = {235},
  url       = {https://mlanthology.org/icml/2024/kulesza2024icml-mean/}
}