31. Reducing MCMC Computational Cost with a Two Layered Bayesian Approach

One of the major challenges in the implementation sampling techniques in the Bayesian approaches is the computational cost involved with the estimation of the likelihood and/or posterior, especially in problems where the models being updated are computationally expensive. This paper proposes the use of surrogate models in a two-layer Bayesian approach to reduce the computational cost of estimating these PDF. In the first layer, the posterior is written in a traditional manner. The second layer attempts to estimate the PDF of the first layer with a surrogate model. Only a few runs of the structural model are needed to create the required samples. Preliminary results are shown with a numerical example to identify the stiffness of a structural system. Only ten simulations of the structural model are used to estimate the posterior PDF.