An early type of single-layer neural network. An input array is covered by a set of feature detectors whose outputs are weighted, summed, and then thresholded to give a single binary output. A perceptron learning algorithm can be used to adjust the weights during training on examples from pattern classes so that new inputs may be correctly classified. Mathematical analyses of perceptrons in the 1970s exposed severe limitations and halted research on neural networks. Now, however, perceptrons have been superseded by multilayer neural networks that do not suffer from those limitations.