TADILER: Continual Learning for Medical Imaging

Context: Developed at Emory Global Diabetes Research Center.

TADILER combines experience replay and zero-shot clustering with foundation model embeddings for strategy-agnostic domain incremental learning:

• Shortlisted for the Best Student Paper Award at VISAPP 2025
• Improves accuracy up to 3.1% on diverse patient populations
• 96.9% accuracy with 99% storage reduction in privacy-sensitive settings

Open source: tadiler
Publications: VISAPP 2025
BSC: Official software page