Federated learning is a way to train a machine learning model without collecting all raw data in one central location. Instead, the model is sent to many devices or servers. Each device learns from its own local data and sends back model updates. The central system combines those updates into an improved shared model.
The main reason to use federated learning is privacy and data control. A phone keyboard, for example, may learn from typing patterns on many devices without every typed message being uploaded to one database. Hospitals or banks may also want to collaborate on a model while keeping sensitive records within their own systems.
Federated learning does not mean no information is shared. Model updates still move between systems, and those updates need protection. Techniques such as secure aggregation, careful access rules, and privacy safeguards may be used to reduce risk.
This approach can be harder than ordinary centralized training. Devices may be offline, slow, or using different kinds of data. Some data may be biased toward one region or user group. Engineers must design the training process so the model improves without being distorted by unreliable or uneven updates.
In simple terms, federated learning lets many participants help train an AI model while keeping the original data closer to where it was created. It is useful when data is valuable, private, distributed, or difficult to move.