FedAwS

Implementation of the Federated Averaging with Spreadout [FedAwS20] algorithm.

References

[FedAwS20]

Felix X. Yu, Ankit Singh Rawat, Aditya Krishna Menon, Sanjiv Kumar. Federated Learning with Only Positive Labels. In ICML (2020). URL: https://proceedings.mlr.press/v119/yu20f/yu20f.pdf

Classes included in fluke.algorithms.fedaws

SpreadModel
SpreadLoss
FedAwSServer
FedAwS

Classes

class fluke.algorithms.fedaws.SpreadModel(weights: Tensor)

Bases: Module

class fluke.algorithms.fedaws.SpreadLoss(margin: float, reduction: Literal['mean', 'sum'] = 'mean')

Bases: Module

class fluke.algorithms.fedaws.FedAwSServer(model: Module, test_set: FastDataLoader, clients: Iterable[Client], weighted: bool = False, aws_lr: float = 0.1, aws_steps: int = 100, margin: float = 0.5, last_layer_name: str = 'classifier', **kwargs: dict[str, Any])

Bases: Server

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

Bases: CentralizedFL