CRII: SaTC: Local Differential Privacy under Correlation

Rochester Institute Of Tech, Rochester NY

Investigators

Abstract

As data has become the fuel that drives business growth, an increasing number of service providers collect large volumes of data from users to gain insights for better business decision-making. Such data may contain or reveal sensitive personal information, and disclosing such information raises significant privacy concerns among the general public. Various Local Differential Private (LDP) data analysis techniques have been proposed to allow a data collector to gain helpful information from the data while ensuring users' privacy. Still, these methods exhibit an inherent trade-off between individual data privacy and data utility, i.e., strong data privacy for individual data contributors comes at the cost of reduced data utility for the data collector, which has been hindering their broad adoption. This project's novelties lie in exploiting the correlation that commonly exists in multi-attribute data, e.g., a person's age and salary, and new correlated random perturbation techniques to develop effective LDP techniques with much-improved privacy and utility tradeoff. The project's broader significance and importance include new tools for service providers to improve how they collect and utilize user data to drive their business decisions and growth while ensuring strong privacy guarantees to individual users as well as privacy-preserving data analysis techniques in various web, mobile, and IoT-based applications and services. This project develops novel LDP techniques to significantly improve the privacy and utility tradeoff by exploiting the correlation in multi-attribute data and the correlation that can be introduced into different users' random perturbations. The project will: (1) develop novel LDP techniques for correlated multi-attribute data via sequential random perturbation for improving data utility without sacrificing privacy guarantee, and (2) design novel LDP techniques with improved privacy and utility tradeoffs by exploiting correlated random perturbation among randomly formed groups of data contributors. The findings from this project will enrich the scientific knowledge of privacy-preserving data analysis and privacy-enhancing technologies. Insights gained from and outputs of the project will be made publicly shared through online tutorials, talks, publications, and software toolkits. The project will integrate research outputs in curriculum development, and will contribute broadly through undergraduate and graduate mentoring, and outreach to K-12 and underrepresented students. 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.