In artificial neural networks, these are ranges of appropriate data, obtained from experimental measurements or from a rigorous mathematical model of a system, which are then used to “train” the neural network to represent a specific problem.