Differentially Private Algorithms for 2020 Decennial Census Detailed DHC Race & Ethnicity
summary
Proposal produces DDHC race and ethnicity counts by adaptively choosing which breakdowns to release based on noisy group size, trading off accuracy and confidentiality with per-level privacy budget tuning. Implementation uses integer counts with a discrete Gaussian mechanism and ZCDP accounting, yielding lower composed privacy loss (example comparison roughly 15.3 under pure DP vs 12.2 under approximate/ZCDP). Prototype generation uses 2010 census data in January 2023 and targets product release August 2023.
AI-written summary · about
chapters (7)
- 0:00Start
- 3:57Detailed Demographic and Housing Characteristics File A (Detailed DHC-A)
- 5:18Disclaimer
- 5:39Takeaways
- 8:13Additional criteria
- 8:39Simple Algorithm
- 9:22Optimization 3
related talks
01h 00mPractical Difficulties in the Application of Differential Privacy for the Release of Large-Scale...
01h 00mStepping-up: The Census Bureau Sets an Example of How to Be a Good Data Steward in the 21st Century
20mShare data without revealing personal information
44mDifferentially Private Inference for Regression Coefficients
36mGenerating realistic, differentially private data sets using GANs
44mFederated Heavy Hitters Discovery with Differential Privacy
01h 01mFrom Theory to Practice: Crafting Differential Privacy Systems with Haskell
01h 14mStatistical Inference and Privacy, Part III