Personalization in Distributed tinyML Applications via Adaptive Clustered Federated Learning

Personalization in Federated Learning (FL) aims to equip each participant with a robust task-specific (personalized) model that is also able to infer from other tasks within the network (generalization). In tinyML applications with model capacity constrains, this personalization vs generalization trade-off is of paramount importance. We showcase a framework with emphasis on personalization, by leveraging Adaptive Clustered Federated Learning, under the constrains imposed by the tinyML setting and based on related research.