Integrating Differential Privacy and Contextual Integrity
summary
Proposes a concrete hybrid: add a new transmission property parameter to contextual integrity and use it to choose differential privacy epsilon/delta by modeling utility and participation tradeoffs. Builds a contextual rubric (roles, purposes, population size, adversary model, architecture) and encodes it as a causal influence diagram / Bayesian network with utilities to solve an optimization for optimal noise. Applies the method to US Census, medical data sharing, and federated learning and reports how DP tunings change design and participation.
AI-written summary · about
chapters (5)
- 0:00Start
- 0:09Integrating Differential Privacy and Contextual Integrity
- 0:55Contextual Integrity (CI) - Contexts A social theory of privacy for interdisciplinary research, (a) Privacy is appropriate information flow. (+ appropriate flow) and (- inappropriate flow)
- 4:57Contributions: Privacy Theory • New formalization of CI. Based on a systematic review of previous computer science implementations of Cl(Benthall et al., 2017) and our use case of tuning and communicating PET parameters • Integrated rubric for privacy analysis
- 6:17Contributions: Parameter Tuning Procedure . Privacy Modeling Components of integrated privacy rubric combine into contextualized model of information flows and threats.
related talks
41mSome Applications of Contextual Integrity
01h 05mMarkets and Incentives for Transacting on (Personal) Information, Part III
01h 00mPANEL: Statistical Theory, Privacy and Data Analysis
19mPractical Considerations for Differential Privacy
20mShare data without revealing personal information
44mHybrid Legal-Technical Concepts of Privacy
06mLessons Learned in Communicating Differential Privacy to Data Subjects
20mDesigning for User Privacy: Integrating Differential Privacy into Ad Measurement...