An autoencoder is a type of artificial neural network used to learn efficient data compression and decompression functions instead of having them encoded by humans.
An Autoencoder has 2 main components : an encoder that compresses some input data, and a decoder that recontructs data from compressed representation.
Autoencoders ends up being useful in number of cases. Autoencoders are used in traditional data compression sense, in that they can learn to reduce the dimensionality of any input. Then, anyone can use the compressed representation to share it, or view it and so on, faster than they could with the original data.