How does dropout regularization works in deep learning?

Dropout regularization works by removing a random selection of a fixed number of the units in a network layer for a single gradient step.