FedRS

Implementation of the [FedRS21] algorithm.

References

[FedRS21]

Xin-Chun Li and De-Chuan Zhan. FedRS: Federated Learning with Restricted Softmax for Label Distribution Non-IID Data. In KDD (2021). URL: https://doi.org/10.1145/3447548.3467254

Classes included in fluke.algorithms.fedrs

RSLoss	Restricted Softmax Loss function.
FedRSClient
FedRS

Classes

class fluke.algorithms.fedrs.RSLoss(class_scaling: Tensor, reduction: Literal['mean', 'sum'] = 'mean')

Bases: Module

Restricted Softmax Loss function.

Parameters:

• class_scaling (torch.Tensor) – Class scaling factor.

• reduction (Literal["mean", "sum"]) – Specifies the reduction to apply to the output.

See also

This loss function is very similar to the one used in FedLC, i.e., fluke.algorithms.fedlc.CalibratedLoss.

class fluke.algorithms.fedrs.FedRSClient(index: int, train_set: FastDataLoader, test_set: FastDataLoader, optimizer_cfg: OptimizerConfigurator, loss_fn: Module, local_epochs: int, alpha: float, count_as_missing: int = 2, fine_tuning_epochs: int = 0, **kwargs: dict[str, Any])

Bases: Client

class fluke.algorithms.fedrs.FedRS(n_clients: int, data_splitter: DataSplitter, hyper_params: DDict | dict[str, Any], **kwargs: dict[str, Any])

Bases: CentralizedFL