Doctoral Seminar: Gaby Dagher

Speaker: Gaby Dagher

Supervisors: Drs. J. Clark, B. Fung

Examining Committee: Drs. T. Eavis. A. Hamou-Lhadj, L. Wang

Title: Publicly Verifiable Data Integration with Differential Privacy

Date: Thursday, March 12, 2015

Time: 11:40 a.m.

Place: EV 3.309

ABSTRACT

Privacy-preserving data integration is a mechanism that enables multiple data owners to securely integrate their data for the purpose of data mining. Applying such a mechanism on set-valued data in a malicious environment, however, involves several challenges, including how to handle the high-dimensional nature of set-valued data, how to prevent malicious parties from inferring sensitive information during the integration process, how to guarantee an effective level of privacy on the released data while maintaining utility, and how to enable independent public verifiability of the protocol.

In this seminar report, we propose the first publicly verifiable protocol for integrating person-specific set-valued data from two or multiple data owners, while providing differential privacy guarantee and maintaining an effective level of utility on the released data. Our proposed approach can handle both horizontally and vertically partitioned set-valued data, and is secure in the malicious adversarial model with dishonest majority.