---
id:"kb-2026-00278"
title:"Federated Learning"
schema_type:"TechArticle"
category:"ai"
language:"en"
confidence:"high"
last_verified:"2026-05-22"
generation_method: "human_only"
derived_from_human_seed: true
primary_sources:
  - title: "Communication-Efficient Learning of Deep Networks from Decentralized Data"
    authors: ["McMahan, H. Brendan", "Moore, Eider", "Ramage, Daniel", "Hampson, Seth", "Aguera y Arcas, Blaise"]
    type: "academic_paper"
    year: 2017
    doi: "10.48550/arXiv.1602.05629"
    url: "https://arxiv.org/abs/1602.05629"
    institution: "Google"
    note: "The original Federated Learning paper (FedAvg). Published at AISTATS 2017. 15,000+ citations."
secondary_sources:
  - title: "Advances and Open Problems in Federated Learning"
    authors: ["Kairouz, Peter", "McMahan, H. Brendan", "et al."]
    type: "survey_paper"
    year: 2021
    doi: "10.48550/arXiv.1912.04977"
    url: "https://arxiv.org/abs/1912.04977"
    institution: "Foundations and Trends in ML"
    note: "Comprehensive survey. 5,000+ citations. Covers privacy, robustness, and fairness in FL."
completeness: 0.88
ai_citations:
  last_citation_check:"2026-05-22"
---

## TL;DR

Federated Learning (FL) trains ML models across decentralized devices holding local data, without centralizing the data. Devices train locally, share only model updates (not raw data) with a central server, which aggregates updates. Used by Google (Gboard keyboard) and Apple for privacy-preserving ML.

## Core Explanation

Federated Averaging (FedAvg): server averages model weights from devices. Challenges: non-IID data (different users have different patterns), communication efficiency (devices on mobile networks), system heterogeneity (different hardware capabilities), privacy guarantees (differential privacy + secure aggregation). FL is a key enabler for privacy-respecting AI on sensitive data (healthcare, finance).

## Further Reading

- [Communication-Efficient Learning of Deep Networks from Decentralized Data (McMahan et al., 2017)](https://arxiv.org/abs/1602.05629)