BIMC: The Bayesian Inverse Monte Carlo method for goal-oriented uncertainty quantification. Part I
share
We consider the problem of estimating rare event probabilities, focusing on systems whose evolution is governed by differential equations with uncertain input parameters. If the system dynamics is expensive to compute, standard sampling algorithms such as the Monte Carlo method may require infeasible running times to accurately evaluate these probabilities. We propose an importance sampling scheme (which we call BIMC) that relies on solving an auxiliary, fictitious Bayesian inverse problem. The solution of the inverse problem yields a posterior PDF, a local Gaussian approximation to which serves as the importance sampling density. We apply BIMC to several problems and demonstrate that it can lead to computational savings of several orders of magnitude over the Monte Carlo method. We delineate conditions under which BIMC is optimal, as well as conditions when it can fail to yield an effective IS density.
READ FULL TEXT
