Using game theory to thwart multistage privacy intrusions when sharing data.

Area of Science:

• Computational Biology
• Privacy-Preserving Data Analysis
• Game Theory in Cybersecurity

Background:

• Increasing collection of person-specific biomedical data necessitates robust privacy measures.
• Existing re-identification risk assessments often fail to account for multi-resource attacks.
• Sharing sensitive health data requires balancing utility with stringent privacy protection.

Purpose of the Study:

• To develop a formal framework for assessing re-identification risk in biomedical data sharing.
• To propose an optimal data sharing strategy considering adversarial multi-stage attack capabilities.
• To evaluate the effectiveness of game theoretic models in mitigating privacy risks.

Main Methods:

• Modeling re-identification as a two-player Stackelberg game of perfect information.
• Incorporating adversarial multi-stage attack strategies into the game theoretic framework.
• Experimental validation using large-scale genomic datasets.

Main Results:

• The proposed game theoretic model accurately assesses re-identification risk under multi-stage attacks.
• Optimal data sharing strategies were identified, balancing privacy and data utility.
• Experiments demonstrated that most data can be shared with low re-identification risk.

Conclusions:

• Game theory provides a powerful tool for formalizing and assessing re-identification risks.
• Adversarial multi-stage attack considerations are essential for realistic privacy risk assessment.
• Effective strategies exist for sharing biomedical data securely, preserving utility and privacy.