Set this config (and recursively its subconfigs) to allow merging \ new keys from other configs.

Initialization data by ``cfg``.

Shut down all rooms and kill all subprocesses.

ROUGE requires texts split into sentences. In case the texts are not already split, this method can be used.

Split dataset to clients and build ``ClientData``. Returns: dict: dict of ``ClientData`` with ``client_idx`` as key.

Anisotropic TV.

calculate the total model size given the models hold by the \ worker/trainer Args models: torch.nn.Module or lis

Track the number of bytes uploaded.

To start the FL course when the expected number of clients have joined

Arguments: grads: grads of other clients to optimize the local model :returns: state_dict of generation model

To update the client information

To update the client information

Copy the model of the server as the discriminator Args: model (object): the model in the server Returns: the di

Update the policy. This implementation is borrowed from the open-sourced FedEx ( https://github.com/mkhodak/FedEx/blob/ \ 150f

View one specific FS task.

returns a list of character indices Arguments: word: string :returns: indices: int list with length len(word)

Build a `DittoTrainer` with a plug-in manner, by registering new functions into specific `BaseTrainer` The Ditto implementation, "Ditto:

Warp the trainer for gan_based class representative attack. Args: base_trainer: Type: core.trainers.GeneralTorchTrainer :return

wrap the gaussian attack trainer Args: base_trainer: Type: core.trainers.GeneralTorchTrainer :returns: The wrapped train

wrap the gradient_invert trainer Args: base_trainer: Type: core.trainers.GeneralTorchTrainer :returns: The wrapped trai

Build `SGDMFTrainer` with a plug-in manner, by registering new functions into specific `MFTrainer`

Wrap the trainer for attack client. Args: base_trainer (core.trainers.GeneralTorchTrainer): the trainer that will be wrapped;

Warp the benign trainer for backdoor attack: We just add the normalization operation. Args: base_trainer: Type: core.trainers.Gen

Use PHE to perform secure evaluation. For more details, please refer to the following papers: An Efficient and Robust System for Vert