Reproduce “Pufferfish Privacy Mechanisms for Correlated Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shuang Song, Yizhen Wang, Kamalika Chaudhuri
Researchers at UC San Diego developed general and efficient Pufferfish privacy mechanisms, the Wasserstein Mechanism and the Markov Quilt Mechanism, to protect correlated data. These mechanisms provide enhanced utility and broader applicability for privacy-preserving analysis of complex datasets, outperforming traditional differential privacy and concurrent approaches in empirical evaluations.
Key insights
- Pufferfish privacy mechanisms are robust to slight adversarial model misspecifications, with quantifiable bounds on privacy loss, enhancing their practical reliability.
- The proposed mechanisms, particularly the Wasserstein Mechanism, can achieve better utility (less added noise) for individual privacy in correlated data compared to global sensitivity-based Group Differential Privacy.
- The Markov Quilt Mechanism possesses sequential composition properties under specific conditions, allowing for multiple, successive analyses on the same dataset without excessive privacy degradation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.