Bayesian inverse problems constitute an important methodological part of parameter studies in predictive modeling. Since high-dimensional parameter spaces often require a huge computational effort, approaches for cost reduction are of fundamental importance. For example, the costs can be substantially reduced by exploiting low-dimensional structure in the form of active subspaces. In this work, we derive new generalized bounds, develop an iterative algorithm, and demonstrate computational benefits with models from various applied disciplines.
«Bayesian inverse problems constitute an important methodological part of parameter studies in predictive modeling. Since high-dimensional parameter spaces often require a huge computational effort, approaches for cost reduction are of fundamental importance. For example, the costs can be substantially reduced by exploiting low-dimensional structure in the form of active subspaces. In this work, we derive new generalized bounds, develop an iterative algorithm, and demonstrate computational benefi...
»