Deploying DP: Experiences from the Front Lines and Open Challenges
Talkat Data Privacy: From Foundations to Applications (Simons Institute workshop, 2019) · 2019-03-04..03-08·
summary
Worked with the US Census’s OnTheMap product, formalizes Title 13 requirements into a pufferfish framework that defines which relational secrets must be hidden, implements DP algorithms for multi-table counts, and benchmarks utility against legacy SDL methods (often comparable, sometimes up to threefold worse). Builds a private residual diagnostics tool to verify regression fit on synthetic releases by spending privacy budget for one-bit verification. Open problems: choosing neighboring-database semantics with invariants, composing verification budgets, and detecting algorithmic artifacts vs. data artifacts.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:52Goals for this talk
- 2:03A Checklist for Deployment
- 6:03Underlying data: LODES
- 6:43Goal: Release Tabular Summaries
- 7:34Title 13 Section 9
- 10:00Current Interpretation
- 12:02Differential privacy
- 13:22Defining Privacy
- 14:24The Pufferfish Framework
- 25:08What is the utility cost?
- 29:31Open Questions
- 33:16A call to arms
- 39:13Differentially Private Residual Plots
related talks
01h 00mPractical Difficulties in the Application of Differential Privacy for the Release of Large-Scale...
19mPractical Considerations for Differential Privacy
17mPrivacy in Theory, Bugs in Practice: Grey-Box Testing for Differential Privacy Libraries
01h 11mCPDP ACADEMIC SESSION III
13mEvaluating the Usability of Differential Privacy Tools with Data Practitioners
16mTesting Differential Privacy with Dual Interpreters
01h 01mFrom Theory to Practice: Crafting Differential Privacy Systems with Haskell
26mAaron Roth, Associate Professor, University of Pennsylvania, NYC 2017