FedROD

Implementation of the FedROD [FedROD22] algorithm.

References

[FedROD22]

Hong-You Chen and Wei-Lun Chao. On Bridging Generic and Personalized Federated Learning for Image Classification. In ICLR (2022). URL: https://openreview.net/pdf?id=I1hQbx10Kxn

Classes included in fluke.algorithms.fedrod

RODModel

Model that combines a global model and a local head.

BalancedSoftmaxLoss

Compute the Balanced Softmax Loss.

FedRODClient

FedROD

Classes

class fluke.algorithms.fedrod.RODModel(global_model: EncoderHeadNet, local_head: EncoderHeadNet)[source]

Bases: Module

Model that combines a global model and a local head. During the forward pass, the global model, formed by an encoder and a head, is used to extract the representation of the input (using the encoder). The representation is then passed to the local head and the global head. The output of the local head is added to the output of the global head and returned as the final output.

Parameters:
class fluke.algorithms.fedrod.BalancedSoftmaxLoss(sample_per_class: Tensor, reduction: Literal['mean', 'sum'] = 'mean')[source]

Bases: Module

Compute the Balanced Softmax Loss.

Parameters:

sample_per_class (torch.Tensor) – Number of samples per class.

class fluke.algorithms.fedrod.FedRODClient(index: int, train_set: FastDataLoader, test_set: FastDataLoader, optimizer_cfg: OptimizerConfigurator, loss_fn: Module, local_epochs: int, fine_tuning_epochs: int = 0, clipping: float = 0, **kwargs: dict[str, Any])[source]

Bases: Client

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

Bases: CentralizedFL