Skip to main content
BelizeChain Docs
GitHub

Privacy-Preserving Federated Learning

Nawal uses the Flower (flwr) framework for distributed training across national institutions (e.g., hospitals, schools, and government offices).

How It Works

  • No Centralized Data: Nawal does not collect central data. Models are trained on local devices where the data lives.
  • Differential Privacy: Uses Opacus DP-SGD with strong privacy settings (ε=1.0) to ensure individual data points cannot be reverse-engineered from model updates.
  • Byzantine Fault Tolerance: Employs Krum aggregation to automatically reject malicious or anomalous updates from bad actors in the network.
  • Blockchain Integration: Validators acting as federated learning nodes submit Proof of Useful Work (PoUW) to the Staking Pallet on BelizeChain. They earn DALLA rewards based on the quality (accuracy), timeliness, and privacy compliance of their contribution.