martFL

The prototype of ACM CCS 2023 paper "martFL: Enabling Utility-Driven Data Marketplace with a Robust and Verifiable Federated Learning Architecture". This work features a robust and verifiable Federated Learning architecture underpinned by a secure model evaluation protocol to remove malicious model updates and a zero-knowledge proving system that ensures fair data trading. Arxiv

WARNING: This is an academic proof-of-concept prototype, and in particular has not received careful code review. This implementation is NOT ready for production use.

Overview

We implemented our martFL framework with Pytorch, Zokrates, and Solidity. We've outlined the organization of this repository:

• src: It contains the Python code for martFL.
• datasets: It contains the necessary dataset files.
• circuit: It contains the circuit of the verifiable aggregation, which is implemented using ZoKrates.
• smart_contract: It contains the trading smart contract, which is implemented using Solidity.
• SEAL-Python: The SEAL library for homomorphic encryption. Repository link: https://github.com/Huelse/SEAL-Python.git
• quantized_data: The quantized data that was generated during the model training process.
• save_result & models: It contains the training results and models, respectively.

Dependencies

The script has been tested running under Python 3.7.13. The dependencies are summarised in requirements.txt.

Reference

If you take advantage of martFL in your research, please cite the following in your manuscript:

@inproceedings{li2023martfl,
  title={martFL: Enabling Utility-Driven Data Marketplace with a Robust and Verifiable Federated Learning Architecture},
  author={Li, Qi and Liu, Zhuotao and Li, Qi and Xu, Ke},
  booktitle={ACM SIGSAC Conference on Computer and Communications Security (CCS)},
  pages={1496--1510},
  year={2023}
}