This repository contains the code implementation for the paper "Global Group Fairness in Federated Learning via Function Tracking." The implementation is structured to facilitate experimentation with the proposed method and includes scripts for running experiments on multiple datasets.

• requirements.txt - Lists the dependencies required to run the code.
• setup.py - Script to create the necessary folder structure for the project.
• run_{***}.py - Various scripts to execute different experiments.
• src/ - Contains the core implementation of the method.
• datasets/ - Folder where datasets should be placed (created by setup.py).

Ensure you have Python installed (we used Python 3.10.12). Then, install the required packages:

pip install -r requirements.txt

Run the setup script to create the necessary directories:

python setup.py

This will create the datasets/ folder where required datasets should be placed.

The following datasets need to be manually downloaded and placed in the datasets/ folder:

• COMPAS Dataset
• Communities and Crime Dataset

Ensure the dataset files are correctly formatted and located in datasets/ before running experiments.

Experiments can be run using the run_{***}.py scripts. Each script corresponds to a different experiment. Settings are passed as keyword arguments. For example:

python run_synthetic.py --method ours --home .

We use Weights & Biases (wandb) for tracking experiments. Our project names for logging are fairFLConvergence and fairFL. Ensure you have a Weights & Biases account and are logged in before running experiments:

wandb login

Refer to individual scripts for details on hyperparameters and configurations.

For citing our work, you may use:

@inproceedings{global_fair_fl,
  title={Global Group Fairness in Federated Learning via Function Tracking},
  author={Yves Rychener, Daniel Kuhn,  Yifan Hu},
  booktitle={International Conference on Artificial Intelligence and Statistics},
  year={2025},
  organization={PMLR}
}

For any issues or questions, please contact Yves Rychener.