Smart devices with built-in sensors, computational capabilities, and networkconnectivity have become increasingly pervasive. The crowds of smart devicesoffer opportunities to collectively sense and perform computing tasks in anunprecedented scale. This paper presents Crowd-ML, a privacy-preserving machinelearning framework for a crowd of smart devices, which can solve a wide rangeof learning problems for crowdsensing data with differential privacyguarantees. Crowd-ML endows a crowdsensing system with an ability to learnclassifiers or predictors online from crowdsensing data privately with minimalcomputational overheads on devices and servers, suitable for a practical andlarge-scale employment of the framework. We analyze the performance and thescalability of Crowd-ML, and implement the system with off-the-shelfsmartphones as a proof of concept. We demonstrate the advantages of Crowd-MLwith real and simulated experiments under various conditions.
展开▼
