Reduced-order model tracking and interpolation to solve PDE-based Bayesian inverse problems

Abstract

This work presents a computationally efficient probabilistic framework that enables the identification of model parameters from noisy measurements of the response. We consider transient PDE-based models, where the parameters correspond to physical properties. An efficient and reliable procedure for estimation of those unknown parameters is pursued. The proposed framework uses a Bayesian approach, an efficient sequential Monte Carlo sampling scheme, and adaptive reduced-order models (ROMs). The Bayesian approach has several advantages including the ability to provide not only point estimates of the quantities of interest, but also measures of credibility and robustness concerning those estimates. The associated sequential Monte Carlo method sampling scheme is embarrassingly parallelizable, as well as efficient in terms of the number of calls to the forward solver (e.g. a finite element code) used in evaluating the likelihood function. We propose to use a ROM adaptation procedure where projection-based ROMs are seen as points on a certain Riemannian manifold and are 'tracked' and interpolated during the sampling process using a database of precomputed ROMs. This approach ensures that an appropriate ROM is used for the likelihood evaluation in every region of the parameter space, thus leading to computational savings while conserving sufficient accuracy. Using numerical examples, we illustrate the capabilities of the proposed framework, and show that it leads to quality estimates with a quantified predictive uncertainty.

Duct heights inferred from radar sea clutter using proper orthogonal bases

“…In this work, we use the weighted average of the closest surface layer parameter, θ i , neighbors: the weights, a i , are given by where κ is a scaling parameter. Although this is a less accurate procedure than direct interpolation, it is much faster and the loss of accuracy is reasonable [ Sternfels and Earls , ]. The scale parameter, κ , is taken as the average of the two smallest distances , where θ i correspond to surface layer environmental parameters contained explicitly within our library.…”

“…Both the exponential and the logarithmic maps use thin SVD procedures [Begelfor and Werman, 2006] that are not computationally expensive (with complexity O(m 2 n)) [Sternfels and Earls, 2013], so the whole manifold interpolation procedure is computationally cheap and fast, thus useful within the context of the proposed RFC inversion framework.…”

“…where is a scaling parameter. Although this is a less accurate procedure than direct interpolation, it is much faster and the loss of accuracy is reasonable [Sternfels and Earls, 2013]. The scale parameter, , is taken as the average of the two smallest distances | i − * |, where i correspond to surface layer environmental parameters contained explicitly within our library.…”

A Bayesian Approach for Localization of Acoustic Emission Source in Plate-Like Structures

“…In the present study, it is solved from a probabilistic point of view by employing Bayesian identification. Rather than pinpointing a single solution by deterministic approaches, the Bayesian approach can provide probabilistic distributions of the unknown parameters, giving both point and interval estimates [29][30][31][32][33][34]. The basic idea of Bayesian identification is that it treats the parameters, currently denoted by a vector , as random variables with joint distribution ( ).…”

A new network approach to Bayesian inference in partial differential equations

“…Improvements use surrogate forward models based on polynomial chaos and Karhunen–Loève expansions . In other approaches to Bayesian inference, surrogate models are also constructed by model reduction based on Gaussian processes and on order reduction . However, for expensive forward models, e.g., based on PDEs, it is still a big challenge to make Bayesian inference computationally tractable.…”