AI
Federated Learning Privacy in Healthcare Data
Code: ANX-RP-002
Design a federated learning framework that enables cross-hospital model training while provably preserving patient privacy through differential privacy guarantees.
Why it matters
Hospitals cannot share patient data due to regulations, yet rare diseases require large datasets for accurate diagnosis. Federated learning offers a path but current implementations leak gradient information.
Research gap
Most FL frameworks either sacrifice too much utility for privacy or provide insufficient privacy guarantees against gradient inversion attacks.
Research questions
- 01 What is the optimal noise calibration for medical imaging FL?
- 02 Can we prove differential privacy bounds for transformer architectures?
- 03 How do we handle non-IID medical data distributions across hospitals?
Suggested technologies
Required skills
Team and supervisor
3-5 researchers, supervised by Muhammad Qasim.
Suitable for: AI, DS, CS.
Expected deliverables
- FL framework with formal privacy guarantees for medical data
- Evaluation on MIMIC-CXR and CheXpert datasets
- Privacy-utility Pareto frontier analysis
Publication
Privacy-Preserving Federated Learning for Medical Imaging
Preprint · 2026-06-20
Work on this problem
Apply to the lab and this problem becomes your supervised research project.