Opportunistic Federated Learning

Opportunistic Federated Learning (OFL) is an encounter-based learning paradigm in pervasive environments. We explore how privacy-preserving model training and fine-tuning can be achieved by leveraging neighbor discovery and device-to-device communication in egocentric networks. In OFL, devices opportunistically incorporate intelligence from others in proximity, to obtain the most relevant information as efficiently as possible in terms of communication cost. Our work shows the feasibility of OFL even with extreme skewness in devices' data distributions and diverse personalized objectives.

Project Participants: Sangsu (Seth) Lee (Ph.D. student), Haoxiang (Steven) Yu (Ph.D. student), Dr. Christine Julien (MPC director)

Publications:

  1. Sangsu (Seth) Lee, James (Xi) Zheng, Jie Hua, Haris Vikalo, Christine Julien. "Opportunistic Federated Learning: An Exploration of Egocentric Collaboration for Pervasive Computing Applications," Proceedings of the IEEE 19th International Conference on Pervasive Computing and Communications. 2021.