We show that dropout improves the performance of neural networks on supervised learning tasks in vision, speech recognition, document classification and computational biology, obtaining state-of-the-art results on many benchmark data sets.…

This significantly reduces overfitting and gives major improvements over other regularization methods. Nitish Srivastava, et al., "Dropout: A Simple Way to Prevent Neural Networks from Overfitting" http://jmlr.org/papers/volume15/srivastav…