Control Variate-based Estimation of Gradient and Its Application in Stochastic Optimization

This research focuses utilization of the control variates to estimate the gradients of performance with respect to input parameters and apply estimated gradients to stochastic optimization algorithm. The control variate-based method is advantageous in that it uses simulation results of a single design point of input parameters and does not need to describe the logical relationship among performance and perturbations of input parameters. This method is generally applicable across simulation models. Simulation experiments on queuing models show the promising results for application of proposed method. We consider the applications of this research to various simulation models and extension of parameter space as future researches.