CRII: SaTC: Foundations for Differentially Private Provenance

Provenance tracks the origin, creation, usage, and modifications of data, and is essential for verifying data integrity and conducting security audits. It supports compliance with strict privacy regulations in sectors like healthcare, finance, and government by maintaining detailed records of data activities. However, detailed provenance graphs can expose sensitive information and are vulnerable to adversarial attacks. Differential privacy provides a way to minimize privacy risks by reducing the impact of individual data contributions on analytical outcomes, offering robust protection against adversaries with extensive knowledge. However, applying differential privacy in practice remains challenging due to the need of balancing data utility with privacy. The project's novelty lies in bridging the gap between differential privacy's theoretical benefits and practical implementation in provenance graphs. The project's broader significance and importance include translating the findings into refined methods and synthetic datasets, and disseminating outcomes to sectors like healthcare, finance, and public safety through educational workshops and seminars.

This project is structured around two main thrusts. Thrust 1 focuses on 1) identifying privacy risks in current provenance-based machine learning anomaly detectors by systematically applying property inference attacks, membership inference attacks, and graph reconstruction attacks; and 2) developing snapshot-level differential privacy with implications that align with enterprise privacy policies, compliance requirements, and performance needs . Thrust 2 aims to develop practical solutions by generating differentially private graphs, through the design of a subgraph synthesis method that addresses the dense correlations and large scale of raw provenance graphs.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Award Number: 2451662
Principal Investigator: Joann Chen
Funds Obligated: $174,995
State: CA