Principal Investigator: Dapeng Oliver Wu
Sponsor: National Science Foundation
Start Date: July 15, 2020
End Date: June 23, 2023
Mobile crowdsensing leverages mobile devices (e.g., smartphones and wearables) to collect sensing data from users and measure spatiotemporal phenomena (e.g., air quality and traffic speed). Yet, existing crowdsensing solutions are mainly built on a cloud-centric approach that raises significant security and privacy challenges. For example, accurate and real-time situation awareness of flood and wildfire is important for incident commanders and residents to fight these natural hazards, and mobile crowdsensing can provide large-scale monitoring of hazards by pictures and/or input provided by mobile users. Although most users are willing to help, they may hesitate to participate in such a crowdsensing task due to privacy concerns, as their private information including GPS locations may be leaked during the transmissions to a cloud server or the storage on the server. This project investigates a hardware and software architecture for aggregation-free and privacy-aware mobile crowdsensing by integrating software and hardware design, edge computing, distributed spatiotemporal optimization, and machine learning based privacy protection. Without aggregating raw sensor data to a central server, it passes latent representation of user data among edge servers until they recover the data of all areas by spatiotemporal interpolation. The educational components of this project include local outreach programs (e.g., the University Minority Mentor Program at the University of Florida and the research week fair at the University of California, Merced) and summer internships to enhance research opportunities for underrepresentation of various populations, including minority and female students.
This project investigates a novel software and hardware architecture that integrates spatiotemporal prediction and distributed optimization into edge computing for aggregation-free and privacy-aware mobile crowdsensing. This project designs a machine-learning pipeline that predicts sensing measurements with partially-available crowdsensed data, at the same time providing privacy-awareness without aggregating sensor data to a central server. An edge computing platform is developed to efficiently manage the above machine learning pipeline and automatically scale up the computing resources of multiple edge servers. Two important applications, natural hazard (flood) and public health (body temperature) monitoring, will be implemented to evaluate the system effectiveness and demonstrate the societal impact.