Privacy-preserving statistical analyses
summary
Explains using differential privacy implemented as a Haskell library called Pella that lets analysts write SQL-like queries, symbolically estimate accuracy on a symbolic dataset (no data, 90% confidence), and then have a data owner run them under a global privacy budget. Demo uses a 30-row CSV with true count 7: epsilon 0.5 yields ~5.9 error, epsilon 1 yields ~2 error, larger epsilon reduces error; composition sums local epsilons and the owner rejects queries that exceed the global budget.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:53Presentation
- 5:07Differential privacy
- 7:16How to use differential privacy
- 8:54Accuracy
- 10:28PLALA
- 10:53DEMO
- 16:43Data Owner
- 25:26Questions
related talks
01h 01mFrom Theory to Practice: Crafting Differential Privacy Systems with Haskell
30mAccuracy First: Selecting a Differential Privacy Level for Accuracy-Constrained...
01h 03mStatistical Inference and Privacy, Part II
42mDifferential Privacy
27mYou've got trust issues, we've got solutions: Differential Privacy
01h 00mPANEL: Statistical Theory, Privacy and Data Analysis
17mNegotiating Privacy/Utility Trade-Offs under Differential Privacy
35mDifferential Privacy Made Practical