Phase 4: Privacy-Preserving AI · 40 min · Python · scikit-learn · numpy

Federated Learning — Training Without Centralizing Data

Don't bring the data to the model. Bring the model to the data.

Hiring signal: Federated learning is increasingly required in healthcare and finance RAI roles. Understanding the architecture, its privacy benefits, and its limitations (gradient leakage) demonstrates systems-level thinking.

What you will learn

Explain the federated learning architecture: local training + gradient aggregation
Implement a simple federated averaging (FedAvg) simulation
Articulate the privacy benefits: data never leaves the device
Explain the limitations: gradient leakage attacks and communication costs

The Problem

Traditional ML requires all training data in one place — a central server. For healthcare, this means every hospital must send patient records to a central location. For finance, every bank must share customer transactions. This creates massive privacy and regulatory challenges:

GDPR restricts cross-border data transfer
HIPAA restricts sharing of protected health information
Competitive concerns: banks don't want to share customer data with each other
Single point of failure: a central data repository is a high-value target for attackers

Federated learning solves this by flipping the architecture: the model travels to the data, not the data to the model. Each participant trains locally and only shares model updates (gradients/weights) — never raw data.

How Federated Learning Works

Round 1:
  Server → sends global model to all clients
  Client A → trains locally on their data → sends updated weights to server
  Client B → trains locally on their data → sends updated weights to server
  Client C → trains locally on their data → sends updated weights to server
  Server → averages all weights (FedAvg) → new global model

Round 2:
  Server → sends new global model to all clients
  ... (repeat)

Key properties:

Raw data never leaves the client device
Only model weights/gradients are shared
The server aggregates updates but never sees individual data
Each client can control how many rounds they participate in

Unlock the full lesson

You've read the first 2 sections. The rest of this lesson covers FedAvg (Federated Averaging), Limitations, Building a Federated Learning Simulation, What's Next — plus a hands-on lab, quiz, and project artifact.

Create a free account to unlock Phase 0 and Phase 1 of every course — no credit card.