Induction and polynomial networks

Induction plays a major role in a wide variety of application domains. Because of this broad range of applicability a variety of approaches have been suggested and employed to discover general models from data. A key goal in these approaches is to perform well on data not seen during the model construction process. This paper surveys the variety of techniques available for induction and categorizes them by their degree of automation. The authors then examine in more detail polynomial networks which are induction methods that grew out of cybernetics and early neural network research. The authors conclude the paper with suggested directions for continued work in polynomial networks.