Heuristic Dynamic Programming for Neural Networks Learning Part 2: I-order Differential Dynamic Programming

Summary

The learning process of multilayer neural networks is considered as a multistage optimal control process. For small values of the gain parameter of used neurons for the learning differential dynamic programming of first order can be applied. Adjustment of the gain parameters can be done by continuation methodology as was described in [1]. In this paper by considering the gain parameter as an additional control variable„ starting form a small value of the parameter, the optimal value of the parameter is found. The methodology we propose to call the heuristic dynamic programming.