Evolutionary-based Federated Ensemble Learning on Face Recognition

Federated learning, as a distributed training framework of machine learning compared to the traditional training on a single, centralized dataset has been increasingly popular with advantages of addressing privacy, data ownership, data isolation and so on. One of the major challenges of federated learning framework is how to design model aggregation as data owners in the network may be different in data characteristics and communication delay which deceases training efficiency or even make the training invalid. To overcome these issues, we proposed a novel learning strategy based on evolutionary theory (EFEL) where we maintain a group of diversified models rather than a single, complex model and allow them to evolve independently. The proposed federated ensemble learning framework is evaluated on both benchmark and real-world databases and achieves better results than FedAvg and FedFS to train the state the art models in face recognition
¹