Information-Theoretic Scores for Bayesian Model Selection and Similarity Analysis: Concept and Application to a Groundwater Problem DOI Creative Commons
Maria Fernanda Morales Oreamuno, Sergey Oladyshkin, Wolfgang Nowak

et al.

Published: Sept. 28, 2022

Bayesian model selection (BMS) and justifiability analysis (BMJ) provide a statistically rigorous framework to compare competing conceptual models through the use of evidence (BME).However, BME-based has two main limitations: (1) it's powerless when comparing with different data set sizes and/or types and(2) doesn't allow judge model's performance based on its posterior predictive capabilities.Thus, traditional approaches ignore useful or due issue disregards updating because (2).To address these limitations, we advocate include additional information-theoretic scores into BMS BMJ analysis: expected log-predictive density (ELPD), relative entropy (RE) information (IE).Exploring connection between inference theory, explicitly link BME ELPD together RE IE indicate flow in analysis.We show how compute interpret alongside BME, apply it similarity framework.We test methodology controlled 2D groundwater setup considering five accompanied sets.The results complement by providing more complete picture concerning process.Additionally, present both can be used objectively that feature sets.Overall, introduced helps avoid any potential loss leads an informed decision for similarity.

Language: Английский

Advancements in bacteria based self-healing concrete and the promise of modelling DOI Creative Commons

Manpreet Bagga,

Charlotte Hamley‐Bennett,

Aleena Alex

et al.

Construction and Building Materials, Journal Year: 2022, Volume and Issue: 358, P. 129412 - 129412

Published: Oct. 19, 2022

In the last two decades self-healing of concrete through microbial based carbonate precipitation has emerged as a promising technology for making structures more resilient and sustainable. Currently, progress in field is achieved mainly physical experiments, but their duration cost are barriers to innovation keep number large scale applications still very limited. Modelling simulation phenomena underlying healing may provide key complement experimental efforts, development its infancy. this review, we briefly present field, introduce some aspects from main ongoing developments modelling mineral systems, discuss how synergy be accomplished speed up near future.

Language: Английский

Citations

48
A fully Bayesian sparse polynomial chaos expansion approach with joint priors on the coefficients and global selection of terms DOI
Paul‐Christian Bürkner, Ilja Kröker, Sergey Oladyshkin

et al.

Journal of Computational Physics, Journal Year: 2023, Volume and Issue: 488, P. 112210 - 112210

Published: May 15, 2023

Language: Английский

Citations

8
Stability criteria for Bayesian calibration of reservoir sedimentation models DOI Creative Commons
Kilian Mouris, Eduardo Acuña Espinoza, Sebastian Schwindt

et al.

Modeling Earth Systems and Environment, Journal Year: 2023, Volume and Issue: 9(3), P. 3643 - 3661

Published: Feb. 3, 2023

Abstract Modeling reservoir sedimentation is particularly challenging due to the simultaneous simulation of shallow shores, tributary deltas, and deep waters. The upstream parts reservoirs, where deltaic avulsion erosion processes occur, compete with validity modeling assumptions used simulate deposition fine sediments in We investigate how complex numerical models can be calibrated accurately predict presence competing model simplifications identify importance calibration parameters for prioritization measurement campaigns. This study applies Bayesian calibration, a supervised learning technique using surrogate-assisted inversion Gaussian Process Emulator calibrate two-dimensional (2d) hydro-morphodynamic simulating Albania. Four were fitted obtain statistically best possible bed level changes between 2016 2019 through two differently constraining data scenarios. One scenario included measurements from entire half reservoir. Another only geospatially valid range model. Model accuracy parameters, evidence, variability four indicate that converges toward physically meaningful parameter combinations when nodes are approach also allowed comparison multiple found dry bulk density deposited most important factor calibration.

Language: Английский

Citations

7
A surrogate-assisted uncertainty-aware Bayesian validation framework and its application to coupling free flow and porous-medium flow DOI Creative Commons
Farid Mohammadi, Elissa Eggenweiler, Bernd Flemisch

et al.

Computational Geosciences, Journal Year: 2023, Volume and Issue: 27(4), P. 663 - 686

Published: July 13, 2023

Abstract Existing model validation studies in geoscience often disregard or partly account for uncertainties observations, choices, and input parameters. In this work, we develop a statistical framework that incorporates probabilistic modeling technique using fully Bayesian approach to perform quantitative uncertainty-aware validation. A perspective on task yields an optimal bias-variance trade-off against the reference data. It provides integrative metric parameter conceptual uncertainty. Additionally, surrogate technique, namely Sparse Polynomial Chaos Expansion, is employed accelerate computationally demanding calibration We apply comparative evaluation of models coupling free flow with porous-medium flow. The correct choice interface conditions proper parameters such coupled systems crucial physically consistent accurate numerical simulations applications. benchmark scenario uses Stokes equations describe considers different compartment at fluid–porous interface. These include Darcy’s law representative elementary volume scale classical generalized pore-network its related approach. study problems’ behaviors considering case, where pore-scale resolved solution. With suggested framework, sensitivity analysis, quantify parametric uncertainties, demonstrate each model’s predictive capabilities, make comparison.

Language: Английский

Citations

5
Contaminant source identification in an aquifer using a Bayesian framework with arbitrary polynomial chaos expansion DOI
Guodong Zhang, Teng Xu, Chunhui Lu

et al.

Stochastic Environmental Research and Risk Assessment, Journal Year: 2024, Volume and Issue: 38(5), P. 2007 - 2018

Published: Feb. 16, 2024

Language: Английский

Citations

1
Gaussian active learning on multi-resolution arbitrary polynomial chaos emulator: concept for bias correction, assessment of surrogate reliability and its application to the carbon dioxide benchmark DOI Creative Commons
Rebecca Kohlhaas, Ilja Kröker, Sergey Oladyshkin

et al.

Computational Geosciences, Journal Year: 2023, Volume and Issue: 27(3), P. 369 - 389

Published: April 14, 2023

Abstract Surrogate models are widely used to improve the computational efficiency in various geophysical simulation problems by reducing number of model runs. Conventional one-layer surrogate representations based on global (e.g. polynomial chaos expansion, PCE) or local kernels (e.g., Gaussian process emulator, GPE). Global omit some details, while require more The existing multi-resolution PCE is a promising hybrid: it representation with refinement. However, can not (yet) estimate uncertainty resulting surrogate, which techniques like GPE do. We propose join and s into joint framework get best out both worlds. By doing so, we correct bias assess remaining itself. emulator offers pathway for several active learning strategies at acceptable costs, compared PCE-kriging approach adds aspect. analyze performance plain using didactic test cases CO 2 benchmark, that representative many alike geosciences. Both approaches show similar improvements during learning, but our leads much stable results than GPE. Overall, suggested be seen as generalization concepts possibility learning.

Language: Английский

Citations

1
Information‐Theoretic Scores for Bayesian Model Selection and Similarity Analysis: Concept and Application to a Groundwater Problem DOI Creative Commons
Maria Fernanda Morales Oreamuno, Sergey Oladyshkin, Wolfgang Nowak

et al.

Water Resources Research, Journal Year: 2023, Volume and Issue: 59(7)

Published: July 1, 2023

Abstract Bayesian model selection (BMS) and justifiability analysis (BMJ) provide a statistically rigorous framework for comparing competing models through the use of evidence (BME). However, BME‐based has two main limitations: (a) it does not account model's posterior predictive performance after using data calibration (b) leads to biased results when that different subsets observations calibration. To address these limitations, we propose augmenting BMS BMJ analyses with additional information‐theoretic measures: expected log‐predictive density (ELPD), relative entropy (RE) information (IE). Exploring connection between inference theory, explicitly link BME ELPD together RE IE highlight flow in analyses. We show how compute interpret scores alongside BME, apply controlled 2D groundwater setup featuring five models, one which uses subset Our complement by providing more complete picture concerning updating process. Additionally, demonstrate both can be used objectively compare feature sets Overall, introduced lead better‐informed decision incorporating post‐calibration performance, allowing work considering usefulness

Language: Английский

Citations

0
Information-Theoretic Scores for Bayesian Model Selection and Similarity Analysis: Concept and Application to a Groundwater Problem DOI Creative Commons
Maria Fernanda Morales Oreamuno, Sergey Oladyshkin, Wolfgang Nowak

et al.

Published: Sept. 28, 2022

Bayesian model selection (BMS) and justifiability analysis (BMJ) provide a statistically rigorous framework to compare competing conceptual models through the use of evidence (BME).However, BME-based has two main limitations: (1) it's powerless when comparing with different data set sizes and/or types and(2) doesn't allow judge model's performance based on its posterior predictive capabilities.Thus, traditional approaches ignore useful or due issue disregards updating because (2).To address these limitations, we advocate include additional information-theoretic scores into BMS BMJ analysis: expected log-predictive density (ELPD), relative entropy (RE) information (IE).Exploring connection between inference theory, explicitly link BME ELPD together RE IE indicate flow in analysis.We show how compute interpret alongside BME, apply it similarity framework.We test methodology controlled 2D groundwater setup considering five accompanied sets.The results complement by providing more complete picture concerning process.Additionally, present both can be used objectively that feature sets.Overall, introduced helps avoid any potential loss leads an informed decision for similarity.

Language: Английский

Citations

0