Translations:Dropout A Simple Way to Prevent Overfitting/21/en
The paper also analyzed the features learned by networks trained with dropout, finding that hidden units developed more distinct and individually meaningful features compared to networks without dropout, which tended to learn redundant co-adapted features.