One of the ways to obtain useful features from the autoencoder is done by constraining h to have a smaller dimension than input
x. An autoencoder with a code dimension less than the input dimension is called under-complete.
Learning a representation that is under-complete forces the autoencoder to capture the most salient features of the training data.
The learning process is described as minimizing a loss function,
L is a loss function penalizing
g(f (x)) for being dissimilar from
x, such as the mean squared error.