Put another way, the trained neural network is too specific to the training data and doesn't generalize well. The most common form of neural network dropout randomly selects half of a neural network's ...