A Real-world Privacy-preserving Mobile Application for Smart Campus via Federated Learning & Analytics
FedCampus is an open-source, cross-platform mobile application that brings federated learning (FL) and federated analytics (FA) into real-world smart campus scenarios. Developed and deployed at Duke Kunshan University, FedCampus empowers privacy-preserving applications like sleep tracking, physical activity monitoring, and personalized health recommendations — all without centralizing user data.
- 🔒 Privacy-first: Differential Privacy applied to both FL & FA workflows.
- 📱 Cross-platform: Native support for both Android (via TFLite) and iOS (via CoreML).
- 🧠 Federated Learning on-device: Train ML models collaboratively across personal smartphones.
- 📊 Federated Analytics: Perform statistical analysis with privacy guarantees.
- 🔁 MLOps-ready: Continuously deploy models and algorithms without updating the app.
FedCampus consists of:
- Flutter-based Mobile App (Android & iOS)
- Django-based Backend Server
- Huawei Health Kit integration for smartwatch data
- Custom FL/FA APIs for encrypted data processing and model lifecycle management
For a detailed breakdown, see our Demo Paper at MobiHoc 2024.
Deployed with 100+ volunteers, FedCampus supported:
| Task | Type | Description |
|---|---|---|
| 💤 Sleep Tracking | FL | Predict sleep efficiency using sensor and phone usage data |
| 🏃♂️ Physical Activity Monitoring | FL | Analyze fitness levels using steps, heart rate, etc. |
| 🎯 Personalized Recommendations | FA | Deliver user-specific health tips based on behaviors |
| 📈 Heavy Hitters Analysis | FA | Identify popular patterns across the student population |
Start building or customizing the mobile app: 📖 Client Developer Guide
Manage models, training, and FA pipelines: 📖 Backend Developer Guide
- Flutter (Cross-platform UI)
- TensorFlow Lite, CoreML (On-device inference/training)
- Django + PyTorch (Server-side backend)
- Differential Privacy APIs (Custom implementations for FL & FA)
- Huawei Health Kit (Wearable data integration)
Developed by:
- Jiaxiang Geng, Beilong Tang, Boyan Zhang, Jiaqi Shao, Bing Luo 📧 Contact: {jg645, bt132, bz106, js1139, bl291}@duke.edu
Special thanks to: Sichang He, Qingning Zeng, Luyao Wang, Renyuan Zhang
If you use FedCampus in your work, please cite our demo paper:
@inproceedings{geng2024fedcampus,
title={Demo: FedCampus: A Real-world Privacy-preserving Mobile Application for Smart Campus via Federated Learning & Analytics},
author={Geng, Jiaxiang and Tang, Beilong and Zhang, Boyan and Shao, Jiaqi and Luo, Bing},
booktitle={Proceedings of the 25th International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc)},
year={2024},
publisher={ACM}
}
This project is licensed under the MIT License. See the LICENSE file for details.
