Federated Learning Architecture
What you'll learn: How federated learning enables collaborative model training across distributed devices without ever centralizing raw user data.
The Core Idea
Imagine teaching a group of students spread across different cities to solve a problem together, but they can never share their personal notes—only their insights. That's federated learning.
In federated learning architecture, the model travels to where the data lives (on devices, hospitals, banks), learns locally, then only the learned updates (gradients or model weights) return to a central server. The raw data never leaves its original location.
How It Works
- Central server distributes a global model to participating nodes (phones, edge devices, institutions)
- Local training happens on each node using its private data
- Gradient aggregation: Nodes send back only model updates (not data)
- Global model update: Server averages updates and redistributes improved model
- Repeat until convergence
This is fundamentally different from traditional centralized training, where all data flows to one location.
Why It Matters for Privacy
Federated learning solves a critical problem: you can train powerful models on sensitive data (medical records, financial transactions, keyboard predictions) without violating privacy. Since raw data stays local, you minimize exposure risk, comply with regulations like GDPR, and respect user sovereignty.
It pairs naturally with differential privacy (lesson 1535-1539) by adding noise to the updates before sharing them, preventing even gradient-based reconstruction attacks.
Key Takeaway: Federated learning decentralizes the training process—models learn where data resides, and only aggregated insights travel, preserving privacy while enabling collaborative intelligence at scale.