Efficient Privacy Auditing in Federated Learning (USENIX Security 2024)

This repository contains the official code for the paper "Efficient Privacy Auditing in Federated Learning", published at the 33rd USENIX Security Symposium (USENIX Security 2024). The code consists of two main parts:

1. Run Federated Learning (FL) tasks
2. Audit the privacy risks of the trained FL models

We leverage the FedML framework (a PyTorch-based federated learning library) to facilitate the FL process. Our implementation includes modifications to the FedML codebase to save intermediate training information for auditing purpose.

---

📁 Code Structure

The repository includes the following key scripts:

Script Name	Description
1_create_split.py	Prepares the federated learning data splits.
2_run_fl.py	Executes the federated learning training process.
3_run_audit.py	Audits the privacy risks of the trained model.

---

🚀 Requirements

Before running the code, ensure you have the following dependencies installed:

conda env create -f environment.yml

🖥️ Running the Code

To run the complete process (data preparation, federated learning, and privacy auditing), execute the provided bash script (which trains resnet56 model on CIFAR-10).

bash run.sh

For different models and dataset combinations, please refer to the Table 1 in the paper.

🔍 How to Cite

If you find this code useful in your research, please cite our paper:

@inproceedings {299655,
author = {Hongyan Chang and Brandon Edwards and Anindya S. Paul and Reza Shokri},
title = {Efficient Privacy Auditing in Federated Learning},
booktitle = {33rd USENIX Security Symposium (USENIX Security 24)},
year = {2024},
isbn = {978-1-939133-44-1},
address = {Philadelphia, PA},
pages = {307--323},
url = {https://www.usenix.org/conference/usenixsecurity24/presentation/chang},
publisher = {USENIX Association},
month = aug
}

📬 Contact

For questions or feedback, feel free to reach out to the Hongyan Chang via email.