What is Federated Learning? - Definition & Meaning

Definition

Federated learning is a machine learning approach where models are trained on data that remains distributed across multiple devices or organizations. Only model updates (gradients or weights) are shared, not the raw data itself.

Technical explanation

A central server coordinates training: clients train locally on their data and send only aggregated updates. Federated Averaging (FedAvg) aggregates weight updates. Challenges: communication costs, client heterogeneity (different data distributions), and security (Byzantine clients). Differential privacy can be applied to updates for extra privacy. Federated learning is used in mobile keyboards, hospitals, and the financial sector where data cannot be centralized.

How AVARC Solutions applies this

AVARC Solutions advises and implements federated learning for clients with distributed, privacy-sensitive data — e.g., across branches, partners, or devices. We help with architecture choices and compliance with GDPR and sector regulations.

Practical examples

• A hospital group jointly training a diagnosis model without sharing patient data between locations.
• A bank training a fraud model across multiple countries without centralizing transaction data.
• A mobile app improving spell suggestions through federated learning on user devices.

Related terms