Federated learning is a machine learning training paradigm in which a model is updated collaboratively across many decentralised participants — phones, hospital servers, enterprise nodes — without requiring any participant to transmit their raw data to a central server. Instead, each participant trains on its local data, computes a gradient update or weight delta, and sends only that compressed numerical summary to an aggregator. The aggregator (often using a technique called FedAvg) combines the updates into a new global model, which is redistributed for the next round. Raw personal data never leaves the device.

Think of it like running a national cookbook competition where each contestant improves a shared base recipe in their own kitchen and mails only their handwritten modifications back to the organisers — who merge the suggestions into an improved master recipe — rather than everyone shipping their private fridge contents to a central warehouse.

Google's Gboard keyboard uses federated learning to improve next-word prediction across millions of Android phones without reading anyone's messages. Each phone trains locally on the user's typing patterns overnight, sends a small encrypted gradient update when plugged in and on Wi-Fi, and the global keyboard model improves across the fleet. The individual messages are never transmitted.

Federated learning intersects with Privacy-preserving AI, Gradient Descent (the update mechanism used locally), and the broader challenge of Overfitting — because each local dataset is small and non-representative, the aggregated model must generalise well from heterogeneous shards. It also relates to Knowledge Distillation in cases where communication bandwidth is reduced by distilling the local update before sending.

You will encounter federated learning when reading about on-device ML for consumer electronics, cross-silo learning in healthcare (hospitals training a shared diagnostic model without sharing patient records), and any production system where data-residency regulations prevent centralised collection.