hiyuchang / mimic_codes

MimiC

This is the code of paper MimiC: Combating Client Dropouts in Federated Learning by Mimicking Central Updates.

This repo contains the codes for our paper MimiC published in Transactions on Mobile Computing, with an introduction in link.

Usage

Prepare

• Create the enviroment conda create -n mimic python=3.7.

• Install neccessary packages pip install requirement.txt.

• Login in your wandb account (for visualization). Or just comment all codes related to wandb.

• Download data to ./data.

Run!

Here is one example to run this code:

python experiments.py \
	--model=cnn \
	--dataset=fmnist \
	--alg=mimic \
    	--lr=0.01 \
    	--lr_decay=0.95 \
	--n_clients=30 \
	--comm_iter=300 \
	--partition='noniid-#label2' \
	--device='cuda:0' \
	--active_mode='tau' \
	--tau_max=20

Client dropout setup

We simulate the following setups for the

• Bounded consecutive dropout iterations: Each client becomes active every $\tau_{\max}(i)$ iterations. Use the parameter --active_mode='tau' and --tau_max=20.
• Static active probabilities: All clients have the same and fixed active probability. Use --active_mode='individual' and --sample=0.8.
• Time-varying active probabilities: The server waits for $P \times N$ active clients by sampling clients without replacement. Use --active_mode='random' and --sample=0.1.

Other parameters

Parameter	Description
model	The model architecture. Options: cnn, vgg11. Default = cnn.
dataset	Dataset to use. Options: fmnist, cifar10. Default = fmnist.
alg	The training algorithm. Options: fedavg, mifa, mimic. Default = mimic.
epochs	Number of local training epochs, default = 5.
n_clients	Number of clients, default = 30.
comm_iter	Number of communication rounds to use, default = 300.
partition	The partition way. Options: noniid-labeldir, noniid-#label1 (or 2, 3, ..., which means the fixed number of labels each party owns). Default = noniid-#label2
beta	The concentration parameter of the Dirichlet distribution for heterogeneous partition, default = 0.5.
device	Specify the device to run the program, default = cuda:0.

Citation

If you find this repository useful, please cite our paper:

@article{sun2023mimic,
  title={Mimic: Combating client dropouts in federated learning by mimicking central updates},
  author={Sun, Yuchang and Mao, Yuyi and Zhang, Jun},
  journal={IEEE Transactions on Mobile Computing},
  year={2023},
  publisher={IEEE}
}

Acknowledgement

The codes are revised based on the benchmark repo: NIID-Bench.