Title: Acceleration of Monte Carlo forward and Bayesian inverse solvers with approximate models

Speaker: Ian Langmore, Columbia University

Date/Time: Tuesday, February 15, 2011, 10:00A.M. to 11:00A.M.       

Location: CSRI Building/Room 90 (Sandia NM)

Brief Abstract: Monte Carlo forward solvers (e.g. simulation of photons paths) offer high accuracy and relatively straightforward implementation.  We use importance sampling techniques and an approximate forward model to dramatically increase speed.  Bayesian inverse solvers often use Markov Chain Monte Carlo methods to explore a posterior density.  We use approximate models to accelerate this exploration.  Taking a step back, we see that the probabilistic nature of Monte Carlo simulations allows one incorporate fast but inaccurate models as part of a larger, accurate scheme.  This allows for a nice mating of deterministic and probabilistic mathematics.

CSRI POC: