Text this: Controlling update distance and enhancing fair trainable prototypes in federated learning under data and model heterogeneity