regulation-sakak/cwFedAvg

cwFedAvg for Personalized Federated Learning

Introduction

---

This repository is the original implementation of our paper Regularizing and Aggregating Clients with Class Distribution for Personalized Federated Learning. Link)

The code in this repository was developed based on Zhang, J.; Liu, Y.; Hua, Y.; Wang, H.; Song, T.; Xue, Z.; Ma, R.; and Cao, J. 2023b. Pfllib: Personalized federated learning algorithm library. arXiv preprint arXiv:2312.04992.

Citation

---

@misc{Lee2024cwfedavg,
  title={Regularizing and Aggregating Clients with Class Distribution For Personalized Federated Learning},
  author={Lee, Gyuejeong and Choi, Daeyoung},
  year={2024},
  eprint={2406.07800},
  archivePrefix={arXiv},
  primaryClass={cs.LG}
}

Dataset setting

---

cd dataset
python generate_cifar10.py noniid - dir

PFL training using cwFedAvg algorithm

---

# cwFedAvg with WDR
cd system
python main.py -lbs 10 -nc 20 -jr 1 -nb 10 -data Cifar10 -m cnn -algo cwFedAvg -gr 1000 -cw -wdr -wd 10 -did 0 -go cnn

# cwFedAvg without WDR
cd system
python main.py -lbs 10 -nc 20 -jr 1 -nb 10 -data Cifar10 -m cnn -algo cwFedAvg -gr 1000 -cw -did 0 -go cnn

# cwFedAvg with Local Class Distribution
cd system
python main.py -lbs 10 -nc 20 -jr 1 -nb 10 -data Cifar10 -m cnn -algo cwFedAvg -gr 1000 -cw -gt -did 0 -go cnn