Traditional ML centralizes data in the cloud. Federated Learning brings the model to the data. It trains algorithms across multiple decentralized edge devices or servers holding local data samples, without exchanging them.
Module 1: The Federated Averaging (FedAvg) Algorithm
A central server distributes a global model to clients. Clients train locally on their private data, compute weight updates, and send ONLY the updates (not the data) back. The server securely aggregates these updates to improve the global model.