Computer Science · cs.LG
Rényi Differential Privacy of the Sampled Gaussian Mechanism
Ilya Mironov, Kunal Talwar, Li Zhang
See on the mapAbstract
The Sampled Gaussian Mechanism (SGM)---a composition of subsampling and the additive Gaussian noise---has been successfully used in a number of machine learning applications. The mechanism's unexpected power is derived from privacy amplification by sampling where the privacy cost of a single evaluation diminishes quadratically, rather than linearly, with the sampling rate. Characterizing the precise privacy properties of SGM motivated development of several relaxations of the notion of differential privacy. This work unifies and fills in gaps in published results on SGM. We describe a numerically stable procedure for precise computation of SGM's Rényi Differential Privacy and prove a nearly tight (within a small constant factor) closed-form bound.
References
The reference list for this paper has not been collected yet.
Cited by 10
10 papers in the index cite this (Semantic Scholar counts 360 in total).
The cited-by list is still filling in while reference lists are collected.