OPTIMAL EXPERIMENTAL DESIGN CRITERIA FOR DATA-CONSISTENT INVERSION
Authors:
- 1 January 2026
- journal article
- Published by Begell House Inc. Publishers in International Journal for Uncertainty Quantification
- Vol. 16 (2) , 75-100
Abstract
The ability to design effective experiments is crucial for obtaining data that can substantially reduce the uncertainty in the predictions made using computational models. An optimal experimental design (OED) refers to the choice of a particular experiment that optimizes a particular design criterion, e.g., maximizing a utility function, which measures the information content of the data. However, traditional approaches for optimal experimental design typically require solving a large number of computationally intensive inverse problems to find the data that maximize the utility function. Here, we introduce two novel OED criteria that are specifically crafted for the data-consistent inversion (DCI) framework, but do not require solving inverse problems. DCI is a specific approach for solving a class of stochastic inverse problems by constructing a pullback measure on uncertain parameters from an observed probability measure on the outputs of a quantities of interest (QoI) map. While expected information gain (EIG) has been used for both DCI and Bayesian based OED, the characteristics and properties of DCI solutions differ from those of solutions to Bayesian inverse problems which should be reflected in the OED criteria. The new design criteria developed in this study, called the expected scaling effect and the expected skewness effect, leverage the geometric structure of pre-images associated with observable data sets, allowing for an intuitive and direct approach to OED. These criteria utilize singular value computations derived from sampled and approximated Jacobians of the experimental designs while avoiding the generally more computationally expensive step of solving the stochastic inverse problem. We present both simultaneous and sequential (greedy) formulations of OED based on these innovative criteria. Numerical results demonstrate the effectiveness in our approach for DCI solutions to stochastic inverse problems.Keywords
All Related Versions
This publication has 55 references indexed in Scilit:
- Solving Stochastic Inverse Problems for Property–Structure Linkages Using Data-Consistent Inversion and Machine LearningJOM, 2020
- Multiscale geometric methods for data sets I: Multiscale SVD, noise and curvatureApplied and Computational Harmonic Analysis, 2017
- Optimal Bayesian Experimental Design for Priors of Compact Support with Application to Shock‐Tube Experiments for Combustion KineticsInternational Journal for Numerical Methods in Engineering, 2016
- Definition and solution of a stochastic inverse problem for the Manning’s n parameter field in hydrodynamic modelsAdvances in Water Resources, 2015
- Parameter Estimation and Optimum Experimental Design for Differential Equation Models2012
- Numerical methods for A-optimal designs with a sparsity constraint for ill-posed inverse problemsComputational Optimization and Applications, 2011
- Optimal Experimental Design for the Large‐Scale Nonlinear Ill‐Posed Problem of Impedance Imaging2010
- Principal component analysisWIREs Computational Statistics, 2010
- Kernel principal component analysis1997
- An overview of robust Bayesian analysisTEST, 1994