Statistical Disclosure Limitation: Where do we go from here?
Talkat Data Privacy: From Foundations to Applications (Simons Institute workshop, 2019) · 2019-03-04..03-08·
summary
Statistical agencies are moving established table builders toward non-interactive table builders that fix the release space in advance and apply perturbation via a discretized Laplace mechanism with caps (example cap ±7) rather than ad hoc suppression. Empirical experiments show Laplace beats a discretized normal and that ε≈1 gives good utility for many census-style tables (smaller ε inflates errors). Analysts must account for noise in likelihood to recover correct tests; challenges remain around weights, margins, synthetic data and allocating a privacy budget.
AI-written summary · about
chapters (13)
- 0:00Intro
- 1:08Traditional Statistical Outputs
- 1:59Types of Disclosure Risks
- 7:29Common SDL Methods
- 17:31Why Differential Privacy?
- 18:58Mechanisms in Differential Privacy
- 20:57Online Flexible Table Builder
- 27:00Differential Privacy Algorithm
- 29:06Exponential Mechanism
- 33:44Analysis under DP
- 35:31Table Builder for Survey Weighted Counts
- 36:45Future for Differential Privacy
- 39:52Challenges for Differential Privacy
related talks
01h 07mStatistical Inference and Privacy, Part I
01h 03mStatistical Inference and Privacy, Part II
01h 00mPANEL: Statistical Theory, Privacy and Data Analysis
27mPrivacy-preserving statistical analyses
01h 00mPractical Difficulties in the Application of Differential Privacy for the Release of Large-Scale...
48mRobust Statistics and Bias Correction: Ideas for Differential Privacy and Utility
01h 00mDifferential Privacy: Fundamentals to Forefront I
42mDifferential Privacy