Making Privacy Technology Accessible: Benchmarks and Platforms
Talkat Data Privacy: From Foundations to Applications (Simons Institute workshop, 2019) · 2019-03-04..03-08·
summary
Empirical benchmark DPBench/DP Comp shows that contradictory paper results trace to signal strength (epsilon, record count) rather than universal algorithm superiority, so testing across regimes matters. An operator framework EktaLo packages vetted primitives like the Laplace mechanism and HDMM, letting reuse and proof-of-privacy reduce errors. Private Sequel picks views, builds differentially private synopses for multi‑relational queries, uses truncation to bound sensitivity for household vs person privacy, and reports better accuracy than per-query Flex-style releases.
AI-written summary · about
chapters (14)
- 0:00Start
- 1:46Analysis & Implementation • Query: 2-way marginal between race of child and race of householder
- 3:11Challenges to deployment
- 4:52Sound evaluation is hard
- 6:46Sound evaluation is important!
- 7:38Challenges of DP Deployment
- 19:24Operator classes and instances
- 22:26Architecture for Private Computation?
- 25:46Outline
- 26:13Motivations for Private SQL
- 26:52Complex Queries
- 28:31Goals of Private SQL
- 33:03Addressing view sensitivity • View is complex SQL query evaluation is hard
- 37:09Summary
related talks
17mPrivacy in Theory, Bugs in Practice: Grey-Box Testing for Differential Privacy Libraries
19mPractical Considerations for Differential Privacy
01h 01mFrom Theory to Practice: Crafting Differential Privacy Systems with Haskell
01h 00mPANEL: Statistical Theory, Privacy and Data Analysis
29mKeynote Presentation: Minding the (Privacy Research/Practice) Gap
26mAaron Roth, Associate Professor, University of Pennsylvania, NYC 2017
14mToward Provably Private Insights into AI Use
16mBalancing Privacy: Leveraging Privacy Budgets for Privacy-Enhancing Technologies