Deep Learning (DL)-powered personalization holds great promise to fundamentally transform the way people live, work and travel, but poses high risk to people's individual privacy. This project will address the privacy risks arising in DL-powered contextual mobile services by developing solutions that facilitate the use of personal information while maintaining explicit user control over use of the information. The developed learning methods will enable learning from mobile devices in a manner flexible enough to enable current and future DL-powered contextual services, while maintaining explicit user control over how that information is used by third-party service providers.
This research will design and implement PADLOCK, a Privacy-Aware Deep Learning Of Contextual Knowledge engine. PADLOCK executes DL computation over users' personal data in a sandbox environment, while performing lightweight static and runtime analysis to ensure that mobile apps comply with users' privacy policies. The design of PADLOCK explores the tradeoff among privacy protection, communication cost, system overhead and service quality, providing solutions with different provable privacy and efficiency features for a wide range of contextual mobile services. For further information see the project web site at: http://x-machine.github.io/project/padlock
