Reproduce “Memorisation bias in medical AI” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Moritz A. Knolle, Martin J. Menten, Laurin Lux, Mélanie Roschewitz, Emma A.M. Stanley, Georgios Kaissis, Daniel Rueckert, Ben Glocker
Knolle et al. test whether training on patients’ historical records changes predictions on their future medical records. They find prediction shifts whose rates vary by architecture. In simulated deployment, sensitivity is lower for returning patients with de novo conditions, while sensitivity and specificity rise for unchanged health states. Prospective clinical outcomes were not measured.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.