Current computer-vision systems do a decent job at classifying images and localizing objects in photos, when they're trained on enough examples. But at their core, the deep-learning algorithms that ...