Rule Learning By Searching On Adapted Nets

Abstract

If the backpropagation network can produce an inference structure with high and robust performance, then it is sensible to extract rules from it. The KT algorithm is a novel algorithm for generating rules from an adapted net efficiently. The algorithm is able to deal with both single-layer and muti-layer networks, and can learn both confirming and disconfirming rules. Empirically, the algorithm is demonstrated in the domain of wind shear detection by infrared sensors with success.