Statistic Maximal Leakage

Area of Science:

• Privacy-preserving data release
• Information theory
• Cryptography

Background:

• Existing privacy measures like maximal leakage protect unknown secrets.
• There is a need for privacy measures that protect known functions of public variables.

Purpose of the Study:

• Introduce statistic maximal leakage, a novel privacy measure.
• Extend the maximal leakage framework to protect known secrets.
• Analyze privacy-utility tradeoffs of data release mechanisms.

Main Methods:

• Define statistic maximal leakage, quantifying worst-case leakage over all priors.
• Prove composition and post-processing properties.
• Develop efficient computation for deterministic mechanisms.
• Analyze quantization and randomized response mechanisms.

Main Results:

• Statistic maximal leakage protects known functions of public variables.
• The measure satisfies composition and post-processing properties.
• Quantization mechanism offers better privacy-utility tradeoffs than randomized response.

Conclusions:

• Statistic maximal leakage provides a robust framework for protecting known secrets in data release.
• The framework aids in assessing leakage without exact priors, enhancing data utility.
• Practitioners can use this to balance privacy and utility more effectively.