5 Perspectives to Why Dropout Works So Well

In 5 minutes

Dropout works by randomly blocking off a fraction of neurons in a layer during training. Then, during prediction (after training), Dropout does not block any neurons. The results of this practice have been enormously successful — competition-winning networks almost always make Dropout an essential part of the architecture.