The perceptron training rule is based on the idea that weight modification is best determined by some fraction of the difference between target and output.