Kernel Stein discrepancies certify Rényi differential privacy
Original title: Task-aware Kernel Stein certificates for Rényi differential privacy
The paper gives exact identities for Gaussian and Laplace mechanisms and bounds for exponential mechanisms.
Researchers use kernel Stein discrepancies based on escort distributions to certify Rényi differential privacy when the divergence is hard to compute. Designing the kernel around relevant privacy directions yields separate task-aware certificates. Exact identities cover Gaussian and Laplace mechanisms, with bounds for exponential mechanisms, and a real-data example shows how the comparison set and released subspace shape the certificates.
- Task-aware certificates target settings such as Rényi Pufferfish privacy.
- Exact identities are given for Gaussian and Laplace mechanisms.
- Bounds are given for exponential mechanisms.
- A real data set illustrates joint effects of the comparison set and released subspace.
From the original article
1 min read at arxiv.orgRènyi differential privacy, based on Rènyi divergence, has become a notion of choice for privacy guarantees. However, outside the Gaussian setting, Rènyi divergence can be difficult to compute. In particular, Rènyi differential privacy guarantees that exploit task-specific information, as in Rènyi Pufferfish privacy, can be challenging.
By Leoni Carla Wirth · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.