A neural anisotropic view of underspecification in deep learning

Open in new window