Designing the Pilot Release of Israel's National Registry of Live Births: Reconciling
summary
Project trains differentially private generative models on the 2014 singleton births registry (about 160k singletons) and outputs a six-column CSV after postprocessing. Utility is enforced by predefined acceptance criteria — marginal and regression error thresholds — plus a one-to-one faithfulness matching computed by linear programming to ensure released rows correspond to real births. Privacy accounting uses private selection and yields total epsilon 9.98; release includes the CSV, a README, and forthcoming code and technical report.
AI-written summary · about
chapters (7)
- 0:00National Registry introduction
- 1:55Differential privacy requirements
- 3:41Solution concepts and criteria
- 6:54Record-level correspondence
- 7:56Face privacy guarantees
- 9:21Privacy loss budget results
- 10:26Conclusion and future
related talks
01h 00mStepping-up: The Census Bureau Sets an Example of How to Be a Good Data Steward in the 21st Century
45mA Practical Method to Reduce Privacy Loss when Disclosing Statistics Based on Small Samples
30mPrivate Generative Models: Balancing Privacy and Utility in AI
17mNegotiating Privacy/Utility Trade-Offs under Differential Privacy
47mDifferential Privacy from a Statistical Perspective–Obtaining Valid Inferences...
43mAnalyzing Administrative Data with Privacy Protection in Place
16mPrvTel: Lightweight Models for Private and Accurate Telemetry Data Retention
22mEnhancing Privacy through an Interactive On-demand Incremental Information