Abstract: Federated Learning (FL) enables multiple devices to collaboratively train models without sharing their raw data. Considering that clients may prefer to design their own models independently, ...