Efficient Weight Estimation in Neural Networks for Adaptive Control

This paper considers the applicaton of neural networks in controlling a system with unknown process equations. To make such an approach practical, it is required that connection weights in the neural network be estimated efficiently. This paper considers the use of a new stochastic approximation algorithm for this weight estimation. It is shown that this algorithm can greatly reduce the computational burden that would be incurred if a more standard stochastic approximation algorithm, based on a finite-difference gradient approximation, were used.