Abstract
We investigate the idea of using the information obtained by observing a real system while simulating the real system to improve the accuracy of a prediction about the real system made based on the result of the simulation. Our approach runs multiple simulators simultaneously in parallel, where parameters of a simulation model are varied across the simulators. Based on the observation from a real system, some of the simulators are replicated, and others are terminated. We verify the effectiveness of our approach with numerical experiments. In addition, we provide a theoretical justification for our approach, using kernel density estimation. © 2011 IEEE.