From Theory to Practice: Crafting Differential Privacy Systems with Haskell

on YouTube ↗01h 01m

" summary

A Haskell eDSL called DPella eDSL in Haskell provides runnable queries that statically report privacy cost and accuracy: a symbolic interpreter computes a privacy budget, an accuracy estimator returns confidence intervals, and an executor emits Laplace/Gaussian-noised outputs (demo on the 33k-row Adult dataset shows ~10k females with adjustable epsilon). The system uses static taint-based accuracy tracking to infer statistical independence (enabling CH bounds vs union bounds), tracks stability at the type level, is ~450 LOC, and is productionized by a startup.

AI-written summary · about

" chapters (15)

1. 0:00Introduction
2. 2:51Personal connection
3. 3:34Data privacy challenges
4. 7:04Differential privacy and Liella
5. 13:53Privacy budget and composition
6. 17:30Live coding Liella in Haskell
7. 24:38Sensitivity and stability
8. 27:28Measuring accuracy
9. 31:24Accuracy with independence
10. 38:35Complex analytics with Liella
11. 39:40Liella architecture recap
12. 40:00Startup around differential privacy
13. 45:02Legitimacy reflections
14. 47:28Rethinking strategy
15. 50:00Q&A

" related talks

• 27mPrivacy-preserving statistical analysesElisabet Lobo-Vesga · foss-north 2020
• 19mPractical Considerations for Differential PrivacyAlex Kulesza · PEPR '25
• 01h 00mPANEL: Statistical Theory, Privacy and Data AnalysisThomas Steinke, John Duchi, Anand Sarwate, Yu-Xiang Wang, Aleksandra Slavkovic, Bin Yu · Privacy and the Science of Data Analysis 2019
• 45mProving Differential Privacy via Relational TypesDanfeng Zhang · Data Privacy: From Foundations to Applications (Simons Institute workshop, 2019)
• 30mAccuracy First: Selecting a Differential Privacy Level for Accuracy-Constrained...Katrina Ligett · Optimization, Statistics and Uncertainty
• 45mMaking Privacy Technology Accessible: Benchmarks and PlatformsMichael Hay · Data Privacy: From Foundations to Applications (Simons Institute workshop, 2019)
• 17mPrivacy in Theory, Bugs in Practice: Grey-Box Testing for Differential Privacy LibrariesJack Fitzsimons, Tudor Cebere, David Erb, Damien Desfontaines, Aurélien Bellet · PEPR '26
• 26mAaron Roth, Associate Professor, University of Pennsylvania, NYC 2017Aaron Roth · MLconf NYC 2017