4 The Perceptron

The perceptron is the simplest type of neural network and it is typically used for classification.

Perceptron classifiers are trained with a supervised training algorithm. This means that the true class of the training data must be available so that they can be submitted to the training function. If a data sample is incorrectly classified then the algorithm modifies the network weights so that performance improves. This is an iterative training process that continues until the perceptron correctly classifies all data or the upper limit of training iterations is reached.

It is often possible to use a smart initialization of the weights of the perceptron so that the extent of the iterative training may be minimized or entirely circumvented.

A short tutorial about the perceptron is given in Section 2.4, The Perceptron. Section 4.1, Perceptron Network Functions and Options defines the commands and the options to work with perceptron networks, and in Section 4.2, Examples, you can find examples illustrating the commands. A small introductory example also appears in Section 3.4.1, Classification Problem Example.