Reproduce “The Effect of Intrinsic Dataset Properties on Generalization: Unraveling Learning Differences Between Natural and Medical Images” 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

Nicholas Konz, Maciej A. Mazurowski

The authors measure label sharpness, how often similar images have different labels, to explain why intrinsic dimension does not fully account for generalization differences between radiological and natural-image datasets. Across binary classification on seven radiology and four natural-image datasets, medical models show higher test loss at similar dimensions; sharpness improves fitted scaling-law likelihood, but causation is not established.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.