Dziugaite et al.

New benchmark shows deep learning generalization measures are regime-dependent

A new benchmark study revisits the challenge of predicting deep learning model generalization before target-test evaluation, focusing on settings with controlled covariate shift rather than independent and identically d…