Privacy-Preserving AI: Federated Learning Simulator

This simulator visualizes how modern privacy-preserving machine learning trains a shared model without ever centralizing anyone's raw data. A ring of client devices, each holding its own private dataset, trains locally every round, perturbs its resulting update with differential-privacy noise scaled by a tunable privacy budget ε, and sends only that noised update to a central server. The server averages every update into an improved global model and broadcasts it back out — the core loop behind federated learning systems like Google's Gboard and cross-hospital diagnostic models. Tune the privacy budget to see the real trade-off between privacy and accuracy, and toggle the raw-data overlay to see exactly what federated learning avoids sending.