The FedMedICL framework supports a variety of algorithms. Depending on the algorithm specified, different settings will be applied automatically.
The --algorithm parameter supports the following baselines:
ERM: Empirical Risk Minimization, standard training without federated learning.fedavg: Federated Averaging, aggregates model updates from multiple clients.mixup: F-MixUp, applies mixup data augmentation in a federated setting.SWAD: F-SWAD, applies stochastic weight averaging in a federated setting.ER: F-ER, uses experience replay in a federated learning context.resampling: F-GB, applies group balanced resampling in federated learning.CRT: F-CRT, employs class rebalanced training in a federated learning setup.CB: F-CB, utilizes class balanced sampling in federated learning.