Federated Learning

Gold definitionUpdated Apr 2, 2026

Definition

Federated Learning (FL) is a distributed machine learning paradigm that enables collaborative model training across multiple decentralized clients, such as mobile devices or organizations, without requiring them to share their raw private data.

At a glance

Executive summary

Federated Learning allows multiple parties to collaboratively train an AI model without sharing their private data, enhancing privacy and data sovereignty. It works by having each participant train a local model and then sending only the model updates to a central server for aggregation, which then improves a global model.

TL;DR

Federated Learning lets many computers work together to build a better AI model by only sharing what they learned, not their private data, keeping everything confidential.

Key points

Clients train models locally on private data and send only model updates (gradients, parameters, logits) for central aggregation.
Solves the problem of training AI models on decentralized, sensitive data while preserving privacy and data sovereignty.
Used by healthcare institutions, LEO satellite operators, threat intelligence consortia, and mobile/edge computing environments.
Unlike centralized training, FL eliminates the need to pool raw data, significantly reducing privacy risks.
Current research focuses on addressing non-IID data, communication efficiency, quantum security, and integrating parameter-efficient fine-tuning.

Use cases

Collaborative medical diagnosis model training across hospitals without sharing patient records (e.g., Med-MMFL, FedKDX).
Securely sharing threat intelligence indicators among organizations using quantum-secure FL to detect cyberattacks.
Optimizing beam selection and management in LEO satellite constellations for efficient non-terrestrial networks.
Enabling AI-native Radio Access Networks (RAN) and Multi-Access Edge Computing (MEC) systems for improved network efficiency.
Cross-device deployments for personalized services on mobile phones, where models are trained on user data without leaving the device.

Also known as

FL, FD, HFL, MMFL, SFed-LoRA, FLoRG, FedKDX, FedLECC, MP-FedKD