Federated Learning Concepts
What you'll learn: How to train machine learning models across multiple decentralized data sources without ever moving the raw data to a central location.
What is Federated Learning?
Federated Learning is a machine learning approach where the model travels to the data, rather than the data traveling to the model. Instead of collecting everyone's data into one central database, you send the model out to where the data lives—on phones, hospitals, or partner organizations—train it locally on each device, and then bring back only the learned weights (the model updates), not the raw data itself.
How It Works
Think of it like a distributed study group where each student (data holder) works on practice problems at home with their own textbook (local data). Instead of photocopying everyone's textbooks to share, each student solves problems, writes down only the insights they learned (model gradients), and shares those at the next meeting. The group combines everyone's insights to improve the master answer key (global model), which then goes back to each student for the next round of practice.
The process:
- A central server sends the current model to all participants
- Each participant trains the model on their local data
- Participants send only the model updates (weights, gradients) back to the server
- The server aggregates these updates (typically by averaging)
- The cycle repeats with the improved global model
Privacy Advantages
Federated Learning naturally aligns with differential privacy and data anonymization strategies you've already learned. Since raw data never leaves its source, it reduces privacy risks while still enabling collaborative model training across sensitive datasets like medical records or financial information.
Key Takeaway: Federated Learning enables collaborative machine learning while keeping data decentralized, addressing privacy concerns by sharing only model updates instead of raw data.