Differential Privacy and Verification of Results

• Activity Number: 640
• Type: Topic Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Business and Economic Statistics Section
• Author(s): David McClure* and Jerry Reiter and Ashwin Machanavajjhala
• Companies: and Duke University and Duke University
• Keywords: differential privacy ; synthetic data

Abstract

Organizations collect private data of great interest to public researchers, which means finding ways to release information about these data while maintaining the confidentiality of study participants is paramount. Synthetic data, generated from models built on the true data, is one of the favored methods for doing this, but it is typically difficult to give theoretical bounds on privacy loss due to the release of these data. We explore the privacy properties of several synthesis models (including differential privacy levels). We also explore the properties of various "fidelity measures", which gives researchers an indication of how close the results of a query on the synthetic data would be to the results that would come from using the true data.