Quantifying Differential Privacy under Temporal Correlations.

Area of Science:

• Computer Science
• Data Privacy
• Cybersecurity

Background:

• Differential Privacy (DP) is a robust privacy framework, but traditional mechanisms often assume data independence.
• Real-world continuous data frequently exhibit temporal correlations, which adversaries can exploit.
• Existing DP methods may not adequately protect against privacy loss in the presence of temporal data dependencies.

Purpose of the Study:

• To investigate and quantify privacy loss in traditional DP mechanisms when applied to temporally correlated continuous data.
• To introduce the concept of temporal privacy leakage and its accumulation over time.
• To develop methods for measuring and bounding this temporal privacy leakage.

Main Methods:

• Modeling temporal correlations using a Markov model to analyze privacy leakage.
• Developing an efficient polynomial-time algorithm to calculate temporal privacy leakage.
• Proposing novel mechanisms to convert existing DP mechanisms for protection against temporal privacy leakage.

Main Results:

• Privacy loss can accumulate and increase over time in DP mechanisms due to temporal correlations, termed temporal privacy leakage.
• An efficient algorithm was developed to measure this leakage, with a bounded supremum identified in certain scenarios.
• Experimental results validated the efficiency and effectiveness of the proposed mitigation mechanisms.

Conclusions:

• Temporal correlations pose a significant privacy risk for continuous data release under traditional DP.
• The proposed methods provide effective means to analyze, measure, and bound temporal privacy leakage.
• This work enhances the security of DP mechanisms for dynamic and time-series data environments.