An autoencoder is a type of artificial neural network used for unsupervised learning of efficient data codings. The aim of an autoencoder is to learn a representation (encoding) for a set of data, typically for dimensionality reduction, feature learning, or data denoising, without supervision.