Blinder: End-to-end Privacy Protection in Sensing Systems via Personalized Federated Learning

This repository contains the implementation of the paper entitled "Blinder: End-to-end Privacy Protection in Sensing Systems via Personalized Federated Learning".

Directories:

• Blinder-Python: Source code of Blinder based on PyTorch
• Blinder-Android: Source code of Android deployment, open in Android Studio

Datasets

Blinder is evaluated on two Human Activity Recognition (HAR) datasets: MotionSense and MobiAct.

The datasets and the preprocessing script (required for MobiAct) are available at:

• MobiAct V2.0: Dataset

  • Preprocessing script: dataset_builder.py

• MotionSense: Dataset

Note: Pre-trained evaluation models can be found under Blinder-Python/eval_models/.

Dependencies

Package	Version
Python3	3.8.13
PyTorch	1.10.2
TensorFlow	2.8.0
imbalanced_learn	0.9.0
scikit-learn	1.1.2

Installation

Android App:

Blinder is deployed on Android platforms for real-time data anonymization. This deployment uses Blinder models pre-trained on MobiAct and MotionSense.

• Android Studio project: https://github.com/sustainable-computing/Blinder/tree/main/Blinder-Android

• Android apk file: https://github.com/sustainable-computing/Blinder/releases

Acknowledgement

• learn2learn: a software library for meta-learning research.

Citation

Xin Yang and Omid Ardakanian. 2023. Blinder: End-to-end Privacy Protection in Sensing Systems via Personalized Federated Learning. ACM Trans. Sen. Netw. 20, 1, Article 15 (January 2024), 32 pages.

@article{yang2023blinder,
  title={Blinder: End-to-end Privacy Protection in Sensing Systems via Personalized Federated Learning},
  author={Yang, Xin and Ardakanian, Omid},
  journal={ACM Transactions on Sensor Networks},
  volume={20},
  number={1},
  pages={1--32},
  year={2023},
  publisher={ACM New York, NY}
}