Autoencoders are models that compress the input data (often images) into a lower dimensional plane and then try to recreate the input data. In the process, they learn the essential information about the data distribution.