Abstract: Recent progress in computer vision has led to new unresolved questions about their emergent properties. Understanding the emergent behaviour of computer vision algorithms can fuel the engineering of computer vision and help understand biological intelligence. In this talk, I will discuss three recent contributions that advance our understanding of the generalization capabilities of deep neural networks: i) a failure mode shared with humans, in which the network's object recognition accuracy in natural images sharply drops due to small changes of the visible region; ii) theoretical and empirical results on the generalization capabilities of state-of-the-art networks for image segmentation; and iii) the correspondence of individual units in deep neural networks with neurons in the brain.