Protecting User Privacy in Private Set Intersection: A Journey Toward Mitigating User
summary
DPCA-PSI implements a redesigned DelPHI-style PSI for multi-ID matching so each key matches independently, turning exponential leakage into linear leakage, then adds a two-party differential privacy mechanism by appending randomized dummy records. The authors claim 27 adversarial PSI queries previously inferred ~3% membership; under DPCA-PSI those attacks reveal nothing across 30 operations. Benchmarks report practical runtime and low overhead (tested at ε=2 on ~10^7 records). The protocol is open-source as DPCA-PSI for researchers and practitioners.
AI-written summary · about
chapters (10)
- 0:00Introduction to PSI
- 0:30Understanding PSI concepts
- 2:46Multi-party computation
- 4:32Protocol vulnerabilities
- 5:37Membership inference attacks
- 8:00Practical PSI challenges
- 9:14Complexity of multi-ID PSI
- 11:45DPCA-PSI protocol design
- 13:48Experimental performance
- 15:15Lessons and conclusions
related talks
20mDesigning for User Privacy: Integrating Differential Privacy into Ad Measurement...
21m"SECRETS ARE LIES, SHARING IS CARING, PRIVACY IS THEFT." - A Dive into Privacy Preserving
34mThe Concrete Security of Two-Party Computation: Simple Definitions, and Tight Proofs for PSI and...
20mShare data without revealing personal information
19mPractical Considerations for Differential Privacy
29mPersonal Data Pseudonymization Versus Anonymization In The Age Of AI & Big Data
45mMaking Privacy Technology Accessible: Benchmarks and Platforms
26mAaron Roth, Associate Professor, University of Pennsylvania, NYC 2017