-
Measuring Severity: What Error Rates Hide
Two models with the same accuracy can fail in very different ways. A Grand Challenges paper at AXIOM 2026 argues that expert-labeled severity is a missing evaluation primitive, most urgent when comparing efficient models for clinical prediction.
-
How I Got Here: From the Lab Bench to Hospital Data
A long answer to a short question. Basketball, a bad first year, optogenetics at SickKids, the overdose crisis, ten courses at UBC, and why I moved to Essen.
-
Federated Learning for Primary Care: Training Models Without Moving Data
Most medical AI is built in hospitals. FLIP-IT deliberately targets primary care: we train models inside general practices, where routine data actually lives.