Dropout in (Deep) Machine learnin
Learning Less to Learn Better — Dropout in (Deep) Machine learning
Amar Budhiraja

It is also something like ensemble learning where each network combination is a set of weak classifiers. We do this so that that as a whole the network becomes stronger and is not prone to overfit. We can train the set of ANN config with a subset of the sample.

In the end we combine the entire neural network when using on the test set