Parameter-Freezing-based Federated Dynamic Sparse Training (PFFDST)

Dependencies

• Python 3.6 or greater
• PyTorch, torchvision
• tqdm

Examples

Experiment	Command line
FedAvg on CIFAR-10	python3 dst.py --dataset cifar10 --sparsity 0.0
FedDST on CIFAR-10 (S=0.9)	python3 dst.py --dataset cifar10 --sparsity 0.9 --rounds 400 --epochs 3
FedDST+FedProx on CIFAR-10 (S=0.9, mu=1)	python3 dst.py --dataset cifar10 --sparsity 0.9 --prox 1 --rounds 400 --epochs 3
PFFDST on CIFAR-10 (S=0.95)	python3 dst.py --dataset cifar10 --sparsity 0.95 --rounds 700 --freeze --epochs 2 --server-readjustment
PFFDST w/o PF on CIFAR-10 (S=0.95)	python3 dst.py --dataset cifar10 --sparsity 0.95 --rounds 400 --epochs 3 --server-readjustment
PFFDST w/o SMR on CIFAR-10 (S=0.95)	python3 dst.py --dataset cifar10 --sparsity 0.95 --rounds 700 --epochs 3 --freeze

Acknowledgements

This work is developed based on the FedDST .