Facilitating Decentralized and Opportunistic Learning in Pervasive Computing
Sangsu Lee, Christine Julien, Xi Zheng · 2022 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops) · 2022
We explore algorithms and develop tools to support the collaboration between devices to build personalized deep learning models for pervasive computing applications. In our work, we first investigate whether individual devices belonging to different users can learn robust models that are personalized to their user’s own experiences while preserving the privacy of the data that resides on the devices. We term this approach Opportunistic Collaborative Learning (OppCL), and we build a resilient learning algorithm that ameliorates the negative effects when coping with unpredictable mobility patterns. Moreover, we extend OppCL from networks of homogeneous devices to heterogeneous ones by employing lossy compression techniques for deep learning models. Furthermore, we develop a programming environment for developers to test decentralized learning algorithms in scale.