Reproduce “Evading the Simplicity Bias: Training a Diverse Set of Models Discovers Solutions with Superior OOD Generalization” 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

Damien Teney, Ehsan Abbasnejad, Simon Lucey, Anton van den Hengel

The authors train collections of predictive models with a penalty that discourages alignment between their input gradients, encouraging them to rely on different cues. In four-block collage tests, the best of 32 proposed models reaches 61.1% accuracy on the CIFAR-predictive test, versus 50.1% for a baseline. These selected-model results do not establish a practical OOD selection strategy.

Summary by alphaXiv.

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