Data and location privacy: challenges and solutions
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Driven by the emerging Internet-of-Things paradigm and the deep penetration of mobile devices into people's everyday life, a massive amount of data in all different forms is being generated daily. On the one hand, there is a growing interest for companies to collect, analyze, and mine the deluge of data to gain valuable insight, facilitate business decision making, and improve their products and services. On the other hand, people have growing concerns about their sensitive information being revealed or abused, and they may be reluctant to contribute their data without a strong privacy guarantee. It is therefore important to not only design and develop effective privacy enhancing techniques to protect users' data privacy but also discover new attacks on data privacy. ☐ Despite significant research efforts on both fronts in the past, there are still several remaining challenges with respect to specific types of data and applications. For example, how can we protect the location privacy of a user without affecting the Quality of Service (QoS) he/she receives from a Location-Based Service Provider? And how can we leverage the temporal correlation to infer a user's location trace if it is protected by Location Privacy Preserving Mechanism (LPPM)? Finally, how can we achieve the same level of data utility across different categories while ensuring Local Differential Privacy (LDP) in data aggregation? In this dissertation, we seek to tackle several key challenges in data and location privacy. ☐ First, we investigate privacy-preserving spatial crowdsourcing. Spatial crowdsourcing is a promising paradigm for collecting location-specific information by outsourcing spatial sensing tasks to a group of mobile workers. While several LPPMs have been proposed to protect workers' location privacy, directly applying them would result in task misassignment and increased workers' travel distance. To tackle this challenge, we design a context-aware privacy-preserving spatial crowdsourcing framework based on a novel Elliptical Laplacian mechanism, which cannot only provide strong location privacy protection but also effectively reduce the chance of task misassignment as well as the total travel distance. ☐ Second, we study the location inference in the presence of temporal correlation. Existing inference attacks either suffer from low inference accuracy or incur exponential computation complexity, making them impractical. To overcome their limitations, we introduce a novel location inference attack that strikes a good balance between inference accuracy and computational complexity by effectively exploiting temporal correlation. ☐ Finally, we study the locally differentially private and fair key-value data aggregation. Current schemes for LDP key-value aggregation primarily rely on uniform sampling for mean estimation, i.e., a single key-value pair is selected randomly from each user's set. However, this approach achieves high estimate accuracy for highly frequent keys while diminishing the accuracy for infrequent ones. To tackle this issue, we present the design and evaluation of a novel two-phase LDP key-value aggregation scheme that can deliver uniformly high estimation accuracy across all keys regardless of their frequency.
Description
Keywords
Data, Location privacy, Mobile devices, Local Differential Privacy, Data privacy, Location Privacy Preserving Mechanism