Connectionist models in which the input is the same as the target. These models have to learn to reproduce the input on the output layer. Often an intermediate layer has fewer processing units than either input and output layers and so forms a bottleneck. In order to learn the task, the model therefore needs to extract statistical information from the input and form compact internal representations.
See Backpropagation, Cognitive neuroscience, Computational models, Connectionism, Connectionist models, Distributed representation, Neural net, Neuroconstructivist networks