Federated Meta Learning for Health Monitoring

Privacy-preserving personalized health monitoring system using federated meta-learning on wearable device data.

Overview

This project combines Federated Learning and Meta-Learning to create a system that can generalize across multiple users while adapting rapidly to individual physiological patterns, all without sharing raw health data.

Key Features

• Federated learning architecture with privacy preservation
• Meta-learning (MAML) for fast personalization
• Non-IID data partitioning across simulated clients
• Wearable device health data analysis
• Differential privacy integration
• Interactive visualization dashboard

Tech Stack

• PyTorch 2.0+
• Flower (Federated Learning)
• learn2learn (Meta-Learning) ← We use learn2learn, NOT higher
• Opacus (Differential Privacy)
• TensorBoard (Training Visualization)
• Hugging Face Datasets

Note: This project specifically uses learn2learn for MAML implementation. If you encounter installation issues, see docs/learn2learn_setup.md.

Installation

# Quick install (all dependencies)
pip install -r requirements.txt

Quick Start

1. Explore the dataset:

jupyter notebook notebooks/phase2_data_exploration.ipynb

2. Train federated model (Phase 3+):

python -m src.federated.server

3. Launch TensorBoard (Phase 4+):

tensorboard --logdir=results/tensorboard

Running Tests

Run the consolidated test suite (recommended):

python -m test.run_all_tests

Or run pytest directly:

pytest

Rapid Prototyping Notice

Most of this code (except for the machine learning modules) was rapidly prototyped and may lead to unexpected errors. Please check the CONTRIBUTING.md file for guidelines, and report any issues via the GitHub Issues tab.

Project Structure

Federated Meta Learning/
├── notebooks/           # Jupyter notebooks for analysis
├── src/
│   ├── data/           # Data loading and preprocessing
│   ├── models/         # Neural network architectures
│   ├── federated/      # Flower client/server
│   └── utils/          # Metrics and visualization
├── configs/            # Configuration files
└── results/            # Experiment outputs

Dataset

Source: SahilBhatane/Federated_Meta-learning_on_wearable_devices (Hugging Face)

Specifications:

• 200 samples: 140 train, 60 test
• 4 users with heterogeneous health patterns
• Features: Heart Rate, Blood Pressure, Temperature, SpO2, Respiratory Rate, Battery
• Target: Health Status (Healthy/Unhealthy)
• Non-IID distribution by user (natural data heterogeneity)

Documentation

See GUIDE.md for detailed setup and usage instructions.

See Planning.md for architecture and research documentation.

Development Status

• Phase 1: Planning (done)
• Phase 2: Dataset & Foundation (done)
• Phase 3: Implementation (done) (63.26% accuracy achieved)
• Phase 4: Evaluation ← Currently working
• Phase 5: Optimization

License

GNU AFFERO GENERAL PUBLIC LICENSE