Share data without revealing personal information
summary
Shows how differential privacy (Dwork 2006) adds calibrated noise so removing one person changes results only within a formal privacy budget, used by the US Census to stop linkage attacks like Latanya Sweeney’s 1990s reidentification (≈85% in one study) and Netflix/IMDb correlations. Describes a trusted-curator model vs. MPC, key limits (base rates, degraded accuracy from noise, exhausted budgets via repeated queries) and practical mitigations (randomized subsets, budget accounting) and points to the OpenMined Python library for hands-on experiments.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 2:56Netflix and IMDb
- 4:49What can you do
- 8:32Aggregate data
- 10:59Requirements
- 12:05Differential Privacy
- 16:06Limitations
related talks
26mDoes using data mean giving up privacy?
42mDifferential Privacy
20mDesigning for User Privacy: Integrating Differential Privacy into Ad Measurement...
27mYou've got trust issues, we've got solutions: Differential Privacy
47mDifferential Privacy from a Statistical Perspective–Obtaining Valid Inferences...
45mPractical Experience with Making Synthetic Data Differentially Private
01h 00mPANEL: Statistical Theory, Privacy and Data Analysis
10mCommunicating Differential Privacy Guarantees to Data Subjects