A Bayesian encourages dropout

Shin-ichi Maeda

Dropout is one of the key techniques to prevent the learning from overfitting. It is explained that dropout works as a kind of modified L2 regularization. Here, we shed light on th…