Does Your Neural Network Extrapolate? Feature Engineering as Identifiability Bias for OOD Generalization

概要

arXiv:2605.07483v1 Announce Type: cross Abstract: Successful deep neural networks discover salient features of data. We show when and why they fail to learn out-of-distribution (OOD)-relevant representations from an in-distribution (ID) training window. This requires decoupling feature learning fro…