What’s the difference between autoencoder and NNPosted: January 20, 2015
When you have small dataset (labeled), you will want to do unsupervised learning on large unlabeled data(same kind) using autoencoder, and learn the features of the dataset (this is called pre-training). And the train the entire network again using labeled data (fine-tunning).
When you have a LOT of data, you can build the NN entirely using backpropagation. No need to do autoencoder way.