Regularization

Dropout as Implicit Bagging

Chapter 7 clarified that dropout works so well because it approximates bagging over many thinned networks with shared parameters.