In deep learning, each layer learns

In deep learning, each layer learns to transform its input data into a slightly abstracted composite representation. In an image recognition application, the original input might be a matrix of pixels; the first representation layer could abstract pixels and encode edges; the second layer could synthesise and encode the arrangement of edges; the third layer could encode the nose and eyes; and the fourth layer could recognise that the image contains a face. Importantly, the deep learning process can learn for itself which features to best place in which layer. This does not completely eliminate the need for manual adjustment; for example, different layers and layer sizes can provide different levels of abstraction.