Efficient Context Inferences and Privacy-Aware Sharing of Sensory Information from Mobile Platforms [Report]

NESL Technical Report #: 2014-4-2


Abstract: Sensing and sharing of personal sensory information are key aspects in current pervasive mobile sensing applications. Sensing on mobile platforms poses a challenge in efficient resource usage because resource-constrained mobile platforms often perform context inferences that require complex machine learning algorithms. Sharing of personal sensory information also poses a challenge in privacy protection because users have no choice but to store their unfiltered data in third-party cloud services. Such challenges have been individually addressed by prior works, but we propose integrated middleware services on mobile and cloud platforms that provide resource-efficient context inferences and ensure ownership of user’s data. Our flow-based context inference service provides rich node connection and graph control mechanisms that achieve resource efficiency by reducing redundant computations. The inferred contexts are privately stored in our personal cloud service, which enforces privacy-aware sharing through user-defined expressive rules and differentially private aggregates. Our integrated system has been used for a user study with 12 participants, and we show how our system helps users feel less privacy concerned when they sense and share their personal data. We also present benchmark results to show resource savings achieved by the flow-based context service and practical performance of the rule processing on the personal cloud service.

Date: 2014-04-01

NESL Document?: Yes

Document category: Report