Overview

This repository provides implementation for three case studies in our paper.

• MNIST digit handwritten dataset
• Free Spoken Digit Dataset (FSD)
• American Sign Language (ASL)

The demonstration of ObfNet is available here.

Getting Started

pip install -r requirements.txt # install dependencies
python mnist/main.py -T -IC -OM # first train cnn-based inference network and then mlp-based obfuscation network
python fsd/main.py  
python asl/main.py 

Note:

• FSD is located at releases. Download and unzip to current folder.
• For ASL dataset, it can be downloaded from here.

Citation

If you use this code for your research, please cite our papers:

@article{xu2019lightweight,
  title={Lightweight and Unobtrusive Privacy Preservation for Remote Inference via Edge Data Obfuscation},
  author={Xu, Dixing and Zheng, Mengyao and Jiang, Linshan and Gu, Chaojie and Tan, Rui and Cheng, Peng},
  journal={arXiv preprint arXiv:1912.09859},
  year={2019}
}

@inproceedings{zheng2019challenges,
  title={Challenges of Privacy-Preserving Machine Learning in IoT},
  author={Zheng, Mengyao and Xu, Dixing and Jiang, Linshan and Gu, Chaojie and Tan, Rui and Cheng, Peng},
  booktitle={Proceedings of the First International Workshop on Challenges in Artificial Intelligence and Machine Learning for Internet of Things},
  pages={1--7},
  year={2019}
}

Acknowledgements

• Code for MNIST & ASL is inspired by keras-mnist.
• Code for FSD is inspired by sound-mnist.