跳到主要内容Skip to main content

联邦学习Federated learning

Federated learning

  • 数据与检索Data & retrieval
  • 第四站 · 看懂Station 4 · Under the Hood
  • 引用核验 2026-11-19refs checked 2026-11-19
解决什么问题What problem it solves

数据留在本地,只上传模型更新,在中心聚合出全局模型。Keep data local and upload only model updates, aggregating a global model centrally.

这是什么What it is

各参与方在本地数据上训练,得到梯度或模型更新。

服务器聚合这些更新(例如 FedAvg),再把新模型发回。

原始数据不出本地,但更新仍可能泄露信息,需要配合差分隐私/安全聚合。

Each participant trains on local data and produces a gradient or model update.

A server aggregates updates (for example FedAvg) and sends the new model back.

Raw data stays local, but updates can still leak; pair with differential privacy or secure aggregation.

为什么需要知道Why it matters

医疗、金融、手机输入法这类数据不能集中,联邦学习提供一种协作训练路径。Medical, financial and keyboard data cannot be centralised. Federated learning offers a collaborative path.

出处Sources
  • Communication-Efficient Learning of Deep Networks from Decentralized Data(arXiv)Communication-Efficient Learning of Deep Networks from Decentralized Data (arXiv 1602.05629) ↗