A Joint Exponential Mechanism For Differentially Private Top-$k$

Jennifer Gillenwater, Matthew Joseph, Andres Munoz, Monica Ribero Diaz
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:7570-7582, 2022.

Abstract

We present a differentially private algorithm for releasing the sequence of $k$ elements with the highest counts from a data domain of $d$ elements. The algorithm is a "joint" instance of the exponential mechanism, and its output space consists of all $O(d^k)$ length-$k$ sequences. Our main contribution is a method to sample this exponential mechanism in time $O(dk\log(k) + d\log(d))$ and space $O(dk)$. Experiments show that this approach outperforms existing pure differential privacy methods and improves upon even approximate differential privacy methods for moderate $k$.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-gillenwater22a, title = {A Joint Exponential Mechanism For Differentially Private Top-$k$}, author = {Gillenwater, Jennifer and Joseph, Matthew and Munoz, Andres and Diaz, Monica Ribero}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {7570--7582}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/gillenwater22a/gillenwater22a.pdf}, url = {https://proceedings.mlr.press/v162/gillenwater22a.html}, abstract = {We present a differentially private algorithm for releasing the sequence of $k$ elements with the highest counts from a data domain of $d$ elements. The algorithm is a "joint" instance of the exponential mechanism, and its output space consists of all $O(d^k)$ length-$k$ sequences. Our main contribution is a method to sample this exponential mechanism in time $O(dk\log(k) + d\log(d))$ and space $O(dk)$. Experiments show that this approach outperforms existing pure differential privacy methods and improves upon even approximate differential privacy methods for moderate $k$.} }
Endnote
%0 Conference Paper %T A Joint Exponential Mechanism For Differentially Private Top-$k$ %A Jennifer Gillenwater %A Matthew Joseph %A Andres Munoz %A Monica Ribero Diaz %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-gillenwater22a %I PMLR %P 7570--7582 %U https://proceedings.mlr.press/v162/gillenwater22a.html %V 162 %X We present a differentially private algorithm for releasing the sequence of $k$ elements with the highest counts from a data domain of $d$ elements. The algorithm is a "joint" instance of the exponential mechanism, and its output space consists of all $O(d^k)$ length-$k$ sequences. Our main contribution is a method to sample this exponential mechanism in time $O(dk\log(k) + d\log(d))$ and space $O(dk)$. Experiments show that this approach outperforms existing pure differential privacy methods and improves upon even approximate differential privacy methods for moderate $k$.
APA
Gillenwater, J., Joseph, M., Munoz, A. & Diaz, M.R.. (2022). A Joint Exponential Mechanism For Differentially Private Top-$k$. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:7570-7582 Available from https://proceedings.mlr.press/v162/gillenwater22a.html.

Related Material