Response learning

摘要

An improved neural network which uses response learning and some of its application developments are reviewed. The improved neural network uses laws of physics expressed as performance functions to provide additional information to the network's response-driven learning procedure in order to achieve a desired response. As a consequence of response learning, a highly efficient computing mechanism is obtained, with a functional representation that replicates the physical law. Response learning is demonstrated with two application examples: learning inverse kinematic equations for robotic control and preliminary development of a neural network autopilot for high-performance aircraft. It is concluded that the improved neural network is superior to standard backpropagation for certain classes of problems