The current deep learning approaches allow us to take raw sensory information and automatically build up hierarchies of concepts. It helps to recognize situations like people riding bikes holding a leash and the dog running alongside.