Applications of a simplified multidimensional stochastic approximation algorithm

A stochastic analog of the Newton-Raphson gradient search method may be used under certain conditions to accelerate convergence of a stochastic approximation algorithm to a local minimum. The method depends upon an iterative technique for determining the Hessian matrix. This paper presents a simplified algorithm which in some design problems sharply reduces the calculations required to determine the matrix. An application of this method to an attitude controller is explored in detail.