faderated_learning

Faderated Learning (FedAvg)

This example implements FedAvg with Redco. It supports assigning a dataset from torchvision.

Requirement

Install TorchVision

pip install torchvision==0.16.1

Install Redco

pip install redco==0.4.13

Usage

python main.py \
    --dataset_name MNIST \
    --n_clients 100 \
    --n_data_shards 200 \
    --n_rounds 100 \
    --n_clients_per_round 10 \
    --n_client_epochs_per_round 5 \
    --per_device_batch_size 64 \
    --do_iid_partition

See def main(...) in glue_main.py for all the tunable arguments.

Results

Setting	Command	Accuracy
IID Partition	python main.py --do_iid_partition	98.82%
Non-IID Partition	python main.py	97.53%