FedUV: Uniformity and Variance for Heterogeneous Federated Learning

Federated learning performance can be significantly improved by promoting IID emulation through variance regularization and hyperspherical uniformity.