C2FL: Clustered Continual Federated Learning under Spatial and Temporal Drift

The paper introduces C2FL, a novel fully distributed Federated Learning (FL) framework designed to address challenges posed by spatial and temporal drift in mobile sensor networks. C2FL utilizes spatial clustering for node organization and employs a dwell-time-aware adaptive averaging mechanism combined with experience replay to mitigate the effects of temporal drift on local models. This approach demonstrates significant improvements in collective adaptation over traditional federated strategies in scenarios with evolving data distributions, making it relevant for practitioners dealing with privacy-sensitive, mobile data environments.