Tag
FedFIbOS proposes a Fisher importance-based method for optimal submodel selection in heterogeneous federated learning, theoretically grounded and achieving about 10% higher accuracy than state-of-the-art methods under non-IID settings.