Law and CS
summary
Project builds formal links between legal terms and CS privacy: extracts a FERPA-derived mathematical definition, proves that standard differential privacy satisfies that conservative legal definition (sufficiency claim), and formalizes predicate singling out to capture GDPR language. Tools include differential privacy, k‑anonymity, leftover‑hash randomness arguments and IID data models. Main findings: naive singling‑out is unattainable (a trivial attacker succeeds ~37% trivial attacker), k‑anonymity/randomization likely fail, differential privacy resists predicate singling out but composition and IID and no‑auxiliary‑knowledge assumptions limit legal conclusions.
AI-written summary · about
chapters (13)
- 0:00Start
- 4:17Legal theorems of privacy
- 5:25Data privacy: the problem
- 14:01An ongoing project
- 19:44k-anonymity [Samarati Sweeney 98, Sweeney 02]
- 24:30The differential privacy desiderata [Dwork, McSherry, N, Smith '06]
- 34:27Related work (cont.)
- 37:41The GDPR (General Data Protection Regulation)
- 48:42Singling out Isolation ?
- 50:24Isolation with a trivial adversary
- 55:41Baseline: How well would a trivial adversary do?
- 57:06Security against predicate singling out (PSO)
- 58:47Properties of security against PSO
related talks
44mHybrid Legal-Technical Concepts of Privacy
38mPrivacy: the Last Stand for Fair Algorithms
29mPrivacy-preserving methods: Building secure projects
45mQuantifying Privacy in Law
01h 02mFormal Methods and Proofs of Privacy Properties, Part III
01h 03mStatistical Inference and Privacy, Part II
01h 00mPANEL: Statistical Theory, Privacy and Data Analysis
01h 04mPrivacy Dependencies