Search-based importance sampling

Importance sampling as a special technique in Monte Carlo probability integration has been shown to be a highly efficient and rather unrestricted method. Non-Gaussian and dependent random variables and nonlinear limit functions can be treated relatively easily and with reasonable rates of convergence. A major draw-back, however, is the need to identify so-called “interesting” or “important” regions for integration. Reference to first-order second-moment (FOSM) methods may help, as well as numerical maximization routines applied. Each involves certain difficulties. An alternative procedure, based on directing and correcting the importance sampling function as sampling is carried out, is presented herein. In particular it is possible to have a multi-modal sampling function.