What you'll learn
- Describe how differentially private training (DP-SGD) protects individual records
- Estimate the accuracy cost on small or under-represented patient groups
- Decide when to combine differential privacy with federated learning
Who it's for
Data protection officers, compliance leads, CTOs, CISOs, data engineers, health ministry teams and researchers who hold or publish patient data.
Hands-on
Train the same simple risk model with and without differential privacy and compare the results.
How the session runs
35 minutes of teaching, a 15-minute hands-on demonstration, then 10 minutes of questions. No programming is needed to follow along; technical attendees can run the optional notebook afterwards.
Part of the Data Privacy & AI pillar