Heuristic Dynamic Programming for Neural Networks Learning Part 1: Learning as a Control Problem

Abstract

The learning process of multilayer neural networks can be considered as a multistage optimal control process. We introduce a gain parameter into the models of neurons. Setting the parameter to a small value makes the neuron model “almost linear”, and the learning process problem can be solved using computational tools specified for linear-quadratic systems optimization. In Part 1 the continuation methodology is applied for changing the gain parameter in order to reach 1.0. In Part 2, by considering the gain parameter as an additional control variable, the optimal value of the parameter can be found. The methodology we propose to call the heuristic dynamic programming.