A resurgence of research in artificial neural networks has sparked interest in applying these networks to difficult computational tasks such as inversion. Artificial neural networks are composed of simple processing elements, richly interconnected. These networks can be trained to perform arbitrary mappings between sets of input-output pairs by adjusting the weights of interconnections. They require no a priori information or built-in rules; rather, they acquire knowledge through the presentation of examples. This characteristic allows neural networks to approximate mappings for functions which do not appear to have a clearly defined algorithm or theory. Neural network performance has proven robust when faced with incomplete, fuzzy, or novel data.