Function optimization by simultaneous perturbation stochastic approximation with randomly varying truncations

A new recursive algorithm is proposed for finding the minimum of an objective function whose gradient is not obtainable directly but is approximated from the noisy observations of the function. The algorithm is based on the simultaneous perturbation stochastic approximation method (SPSA) combined with randomly varying truncations, and provides the estimate, which is convergent under weaker conditions than the conventional SPSA. Numerical simulation studies illustrate the applicability of the proposed algorithm.