In machine learning there's need for a figure to model the features of the dataset. In deep learning, which is designed to be closer to how neurons work, features are found automatically by the classificator through a backpropagation and activation function mechanism.