Federated Learning
Training TechniquesA machine learning approach where a model is trained across multiple decentralized devices or servers holding local data, without ever exchanging the raw data itself - only model updates are shared.
Each hospital trains a diagnostic model on its own patient data, then shares only what the model learned - not the patient records. The central server combines the lessons into a smarter model and sends it back. No hospital ever sees another hospital's data.
Federated learning is a training technique where a shared model is improved collaboratively across many participants without centralizing their data. Instead of sending raw data to a central server, each participant trains the model locally on their own data and sends only the model updates (gradients or weight changes) back to a coordinating server. The server aggregates these updates to produce an improved global model, then sends it back to participants for the next round.
Google introduced federated learning in 2016 to improve the Gboard keyboard predictions on Android phones. Each phone trained a local model on the user's typing patterns, sent only the model update to Google's servers, and the aggregated model improved predictions for all users without Google ever seeing individual keystrokes.
The core algorithm is Federated Averaging (FedAvg). Each participant trains for several local epochs, then the server averages the resulting weights proportional to each participant's dataset size. This reduces communication rounds significantly compared to sending gradients after every batch.
Federated learning addresses three practical problems: privacy (raw data never leaves the device), regulation (data stays within jurisdictional boundaries, relevant for GDPR and HIPAA), and bandwidth (sending model updates is far cheaper than transmitting raw datasets).
Challenges include non-IID data (participants have different data distributions, which can cause the global model to diverge), communication overhead (even compressed updates are expensive at scale), and security (malicious participants can poison the model by sending adversarial updates). Techniques like secure aggregation and differential privacy are used to mitigate these risks.
Federated learning is used in healthcare (hospitals collaborating on diagnostic models without sharing patient records), finance (banks building fraud detection models without sharing transaction data), and mobile devices (improving autocomplete, voice recognition, and recommendations locally).
References & Resources
Related Terms
Last updated: March 11, 2026