Axioms for Information Leakage

on YouTube ↗41m

" summary

Claims that a small set of intuitive axioms (continuity, convexity/monotonicity, and a data‑processing-like property) exactly characterizes vulnerability measures used to quantify leakage, and proves G-vulnerability (gain‑function models) is a sound and complete characterization of all measures satisfying them. Uses hyperdistributions (channels as priors→distributions over posteriors) and elementary convex analysis to show posterior vulnerability is the expected gain and that those axioms collapse into equivalent constraints. Results are formal and constructive; no empirical system built.

AI-written summary · about

" chapters (18)

1. 0:00Intro
2. 0:33This talk in a nutshell • Quantitative information flow (QIF): measures the leakage of sensitive information by computer systems.
3. 3:26Organization of the Talk
4. 3:48Secrets and priors
5. 5:44Probabilistic systems are noisy channels: an output can correspond to different inputs, and an input can generate different outputs, according to a prob. distribution
6. 9:36Example of Anonymity Protocol DC Nets Chaum'88
7. 10:55Chaum's solution
8. 20:17Information theory: useful concepts • Entropy H(X) of a random variable X • A mesure of the degree of uncertainty of the events
9. 27:38Notion of entropy modeling the adversary in the example In the best strategy, the number of questions needed to determine the
10. 28:23Other entropies considered in the literature
11. 30:17Motivation: there are many kinds of adversaries that are not represented by the traditional measures
12. 31:13General framework: g-vulnerability The adversary is represented by a parameter g . The adversary makes a guess w E W about the secret
13. 32:26Background: channels and hypers Channel from a prior to the posteriors
14. 35:24Background: channels and hypers Vulnerability of Prior and Posteriors
15. 36:41Axiomatization of prior vulnerabilities
16. 38:22Axiomatization of posterior vulnerabilities
17. 39:58Posterior vulnerability as expectation
18. 40:54Conclusion

" related talks

• 01h 19mOpen Problem SessionAdam M. Smith, Sofya Raskhodnikova, Shay Moran, Raef Bassily, Thomas Steinke, Adam Sealfon, Matthew Reimherr, Vitaly Feldman, Sashko Nikolov · Privacy and the Science of Data Analysis 2019
• 41mReducibility and Statistical-Computational Gaps from Secret LeakageGuy Bresler · Computational Phase Transitions (Simons Institute, 2020)
• 45mQuantifying Privacy in LawFelix Wu · Data Privacy: From Foundations to Applications (Simons Institute workshop, 2019)
• 01h 02mFormal Methods and Proofs of Privacy Properties, Part IIIMarco Gaboardi · Data Privacy: Foundations and Applications Boot Camp
• 48mGeometry of Secure ComputationHemanta Maji · Minimal Complexity Assumptions for Cryptography 2023
• 48mTales of Obfuscation in Bounded Arithmetic, Metacomplexity, and Differential PrivacyRahul Ilango · Minimal Complexity Assumptions for Cryptography 2023
• 38mAdvisor-Verifier-Prover Games and the Hardness of Information Theoretic CryptographyBenny Applebaum · Minimal Complexity Assumptions for Cryptography 2023
• 01h 06mOn the Assumptions Used for ObfuscationBenny Applebaum · Simons Institute Workshop: New Developments in Obfuscation