Differential Privacy

on YouTube ↗42m

" summary

Survey of real attacks shows reidentification from anonymized records (Massachusetts medical release, Netflix Prize, GWAS/Homer). Formal guarantee treats the release algorithm as randomized and demands epsilon-differential privacy (with optional epsilon-delta slack): outputs from databases differing by one person are nearly indistinguishable. Simple mechanism adds Laplace noise to counts (Laplace mechanism), and privacy degrades under repeated use according to a composition theorem; typical epsilons around 0.1–2 give usable accuracy on large datasets but hurt small clinical cohorts.

AI-written summary · about

" chapters (4)

1. 0:00Start
2. 13:08Resolving Individuals Contributing Trace Amounts of DNA to Highly Complex Mixtures Using High-Density SNP Genotyping Microarrays
3. 30:31INTERPRETATION OF DIFFERENTIAL PRIVACY
4. 31:17LAPLACE MECHANISM

" related talks

• 01h 00mDifferential Privacy: Fundamentals to Forefront ICynthia Dwork · Cryptography Boot Camp 2015
• 30mAccuracy First: Selecting a Differential Privacy Level for Accuracy-Constrained...Katrina Ligett · Optimization, Statistics and Uncertainty
• 20mShare data without revealing personal informationGordon Haff · DevConf.CZ 2021
• 01h 03mStatistical Inference and Privacy, Part IIGautam Kamath · Data Privacy: Foundations and Applications Boot Camp
• 47mDifferential Privacy from a Statistical Perspective–Obtaining Valid Inferences...Jörg Drechsler · Data Privacy: From Foundations to Applications (Simons Institute workshop, 2019)
• 26mAaron Roth, Associate Professor, University of Pennsylvania, NYC 2017Aaron Roth · MLconf NYC 2017
• 01h 43mPerspectiveKobbi Nissim · Data Privacy: Foundations and Applications Boot Camp
• 27mYou've got trust issues, we've got solutions: Differential PrivacyVikram Waradpande, Sarthika Dhawan · PyCon Italia 2023