You might liken each layer to the different areas found in the human brain’s cerebral cortex. The visual cortex, as an example, is responsible for not just seeing objects but has the ability to recognize them. This is the part of the brain that can identify images, recognize sequences, etc. In our brains, each of the sections has a very specific hierarchal order.
When our brain receives an image, it doesn’t just immediately label it, but it processes it through several different stages. First, it must detect the edges, then it perceives the shape, and then it deciphers the colors, etc. Of course, all of this happens at the lightning speed of 13 milliseconds; if you’re watching a movie it is about the speed of 75 frames a second. Neural networks, infused with deep learning, have a very similar hierarchal order. Even a simple network of three layers is now capable of distinguishing things in all kinds of environments and situations.
As data passes through the different layers it can select different aspects of the data and disregard those details that do not apply to the problem it is working on. There are several different types of deep learning networks: