SC²S Colloquium - October 9, 2013

From Sccswiki
Jump to navigation Jump to search
Date: October 9, 2013
Room: 02.05.037
Time: 3 pm, s.t.


Emily Mo-Hellenbrand: Sparse grid interpolants as surrogate models in statistical inverse problems

Inverse problems, also called inference problems, are one of the most important and well-studied mathematical model. In the past few decades, statistical inverse problems have raised in many branches and fields of mathematics, science and engineering. In most practical cases, they involve dealing with large data sets and high-dimensional problem spaces, which makes them intractable to solve with high-fidelity forward models. For this reason, different techniques for reducing computational costs are required, such as employing more efficient sampling methods, employing surrogate models that approximate the high-fidelity models at much lower computational costs. Driven by this motivation, this thesis targets on the analysis and experiments of sparse grid interpolants (SGI) as surrogate models in the Bayesian inference framework. For assessing the quality of the SGI as surrogate models, this thesis presents three experiments, which are inverse problems based on three different classes of systems. The results show that the SGI surrogate models are suitable for statistical inverse problems. Indeed, they demonstrate good capability of inferring parameters with sparse observed data containing large noise.