Privacy-Preserving Database Fingerprinting.

Area of Science:

• Database Security
• Information Privacy
• Applied Cryptography

Background:

• Sharing relational databases requires balancing data utility with privacy and liability guarantees.
• Existing methods for simultaneous privacy and liability often degrade database utility.
• Orthogonal objectives of privacy (single noise addition) and liability (unique noise per copy) pose a challenge.

Purpose of the Study:

• To propose a novel entry-level differentially-private (DP) fingerprinting mechanism for relational databases.
• To achieve simultaneous privacy and liability guarantees without significant utility degradation.
• To provide provable privacy guarantees by leveraging fingerprinting randomization.

Main Methods:

• Devised a bit-level random response scheme for differential privacy on database entries.
• Developed an ϵ-entry-level DP fingerprinting mechanism based on the privacy scheme.
• Utilized a sparse vector technique to manage cumulative privacy loss across multiple shares.
• Theoretically analyzed privacy, fingerprint robustness, and utility using closed-form expressions.

Main Results:

• The proposed mechanism achieves simultaneous privacy and liability guarantees with minimal utility loss.
• Demonstrated strong fingerprint robustness, resisting over 50% of entry modifications.
• Achieved higher application-dependent database utility compared to baseline methods.
• Provided theoretical analysis connecting privacy, robustness, and utility.

Conclusions:

• The novel DP fingerprinting mechanism effectively balances privacy, liability, and utility in relational database sharing.
• The approach offers a robust solution for secure database distribution.
• Future work could explore further optimizations and applications of the technique.