Altman Micah, Cohen Aloni
CREOS, MIT Libraries, Massachusetts Institute of Technology, Cambridge, MA, United States.
Computer Science, University of Chicago, Chicago, IL, USA.
PeerJ Comput Sci. 2023 Sep 28;9:e1576. doi: 10.7717/peerj-cs.1576. eCollection 2023.
We introduce "Natural" differential privacy (NDP)-which utilizes features of existing hardware architecture to implement differentially private computations. We show that NDP both guarantees strong bounds on privacy loss and constitutes a practical exception to no-free-lunch theorems on privacy. We describe how NDP can be efficiently implemented and how it aligns with recognized privacy principles and frameworks. We discuss the importance of formal protection guarantees and the relationship between formal and substantive protections.
我们引入了“自然”差分隐私（NDP）——它利用现有硬件架构的特性来实现差分隐私计算。我们表明，NDP既能保证对隐私损失的严格限制，又构成了隐私领域无免费午餐定理的一个实际例外情况。我们描述了NDP如何能够高效实现，以及它如何与公认的隐私原则和框架相一致。我们讨论了形式化保护保证的重要性以及形式化保护与实质性保护之间的关系。
