[Resource Topic] 2024/004: Practical Two-party Computational Differential Privacy with Active Security

Welcome to the resource topic for 2024/004

Practical Two-party Computational Differential Privacy with Active Security.

Authors: Fredrik Meisingseth, Christian Rechberger, Fabian Schmid


Distributed models for differential privacy (DP), such as the local and shuffle models, allow for differential privacy without having to trust a single central dataholder. They do however typically require adding more noise than the central model. One commonly iterated remark is that achieving DP with similar accuracy as in the central model is directly achievable by \textit{emulating the trusted party}, using general multiparty computation (MPC), which computes a canonical DP mechanism such as the Laplace or Gaussian mechanism. There have been a few works proposing concrete protocols for doing this but as of yet, all of them either require honest majorities, only allow passive corruptions, only allow computing aggregate functions, lack formal claims of what type of DP is achieved or are not computable in polynomial time by a finite computer. In this work, we propose the first efficiently computable protocol for emulating a dataholder running the geometric mechanism, and which retains its security and DP properties in the presence of dishonest majorities and active corruptions. To this end, we first analyse why current definitions of computational DP are unsuitable for this setting and introduce a new version of computational DP, SIM$^*$-CDP. We then demonstrate the merit of this new definition by proving that our protocol satisfies it. Further, we use the protocol to compute two-party inner products with computational DP and with similar levels of accuracy as in the central model, being the first to do so. Finally, we provide an open-sourced implementation of our protocol and benchmark its practical performance.

ePrint: https://eprint.iacr.org/2024/004

See all topics related to this paper.

Feel free to post resources that are related to this paper below.

Example resources include: implementations, explanation materials, talks, slides, links to previous discussions on other websites.

For more information, see the rules for Resource Topics .