yes, but why? Why does image learning require more data than text for deep networks? (if it does) What does that mean? Also amused by the idea of leaving out

yes, but why? Why does image learning require more data than text for deep networks? (if it does) What does that mean? Also amused by the idea of leaving out layers in deep models - perhaps we could call these new models "neural networks"?