pith. sign in

arxiv: 2604.04315 · v1 · submitted 2026-04-05 · 📊 stat.ME · stat.CO

Mean--Variance Risk-Aware Bayesian Optimal Experimental Design for Nonlinear Models

Wanggang Shen , Xun Huan This is my paper

Pith reviewed 2026-05-14 21:50 UTC · model grok-4.3

classification 📊 stat.ME stat.CO
keywords Bayesian optimal experimental designmean-variance optimizationrisk-aware designnonlinear modelsMonte Carlo estimationdelta methodBayesian optimizationrobust experimental design
0
0 comments X p. Extension

The pith

Bayesian optimal experimental design for nonlinear models can achieve robust performance by optimizing a mean-variance utility criterion.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a risk-aware extension of Bayesian optimal experimental design that augments the standard expected utility with a penalty on utility variance. This mean-variance objective is evaluated using Monte Carlo sampling from the prior distribution to compute the necessary statistics without requiring posterior sampling. Bias and variance expressions are derived via the conditional delta method to support accurate optimization. Bayesian optimization with common random samples is used to find the best design under this criterion. The method is tested on a linear Gaussian case, a nonlinear problem, and a contaminant source inversion task, showing designs with much lower utility variability at similar average utility levels.

Core claim

The proposed variance-penalized formulation of Bayesian optimal experimental design yields designs with substantially reduced utility variability while maintaining competitive expected utility values for nonlinear models.

What carries the argument

The mean-variance objective estimated by Monte Carlo from prior samples together with conditional delta-method approximations for bias and variance.

Load-bearing premise

The Monte Carlo estimators using prior sampling and the conditional delta-method expressions give accurate enough estimates of the utility variance for the optimization to succeed.

What would settle it

Compute the actual sample variance of the utility over a large number of independent draws from the posterior for the optimized mean-variance designs and compare it to that of standard expected-utility designs; lower variance would confirm the result.

Figures

Figures reproduced from arXiv: 2604.04315 by Wanggang Shen, Xun Huan.

Figure 1
Figure 1. Figure 1: Utility distributions at two example designs. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Case 1: Convergence of the MC estimator for the utility variance [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Case 1: Estimated utility variance with (top row) and without (bottom row) CRS. [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Case 1: Estimated and exact expected utility. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Case 2: Estimated expected utility and utility variance. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Case 2: Estimated mean–variance objective [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Case 2: Histogram distributions of utility [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Case 2: Scatter plots of uKL(ξ, y) versus y. 0.0 0.2 0.4 0.6 0.8 1.0 0 20 40 60 80 100 p ( |y, ) y=0.03 y=1 (a) ξ = 0.2 0.0 0.2 0.4 0.6 0.8 1.0 0 20 40 60 80 100 p ( |y, ) y=0.03 y=1 (b) ξ = 1 [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Case 2: Posterior densities p(θ|y, ξ) for y = 0.03 and y = 1. This behavior can be understood from the forward model itself in (43). At ξ = 0.2, the model is dominated by the approximately linear term in θ, whereas at ξ = 1 the cubic term becomes dominant. As shown in [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Case 2: Forward model G(θ, ξ) as a function of θ for ξ = 0.2 and ξ = 1. We next examine the behavior of BO with λ = 1 [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Case 2: BO for λ = 1 with (top row) and without (bottom row) CRS. 0.0 0.2 0.4 0.6 0.8 1.0 1 0.0 0.2 0.4 0.6 0.8 1.0 2 3.29 3.34 3.38 3.43 3.48 3.53 3.58 3.62 3.67 0.0 0.2 0.4 0.6 0.8 1.0 1 0.0 0.2 0.4 0.6 0.8 1.0 2 3.28 3.33 3.38 3.43 3.48 3.53 3.58 3.63 3.68 (a) Ub(ξ) 0.0 0.2 0.4 0.6 0.8 1.0 1 0.0 0.2 0.4 0.6 0.8 1.0 2 0.00 0.06 0.12 0.18 0.24 0.30 0.36 0.42 0.48 0.0 0.2 0.4 0.6 0.8 1.0 1 0.0 0.2 0.4 0.6… view at source ↗
Figure 12
Figure 12. Figure 12: Case 2: Estimated expected utility, utility variance, and mean–variance objective with (top row) [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Case 2: BO for λ = 1 with (top row) and without (bottom row) CRS. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Case 3: Comparison of the concentration field [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Case 3: Estimated expected utility, utility variance, and their trade-off for the one-sensor problem. [PITH_FULL_IMAGE:figures/full_fig_p019_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Case 3: Estimated mean–variance objective with different values of [PITH_FULL_IMAGE:figures/full_fig_p020_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Case 3: Histogram distributions of uKL(ξ ∗ , Y ) and uKL(ξ ∗ λ , Y ) for the one-sensor problem. The values before and after ± are the mean and standard deviation, respectively. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Case 3: BO with λ = 0.5 for the one-sensor problem. resulting KL divergences, and then select the five lowest-utility cases for each design. The cor￾responding posterior distributions are shown in [PITH_FULL_IMAGE:figures/full_fig_p021_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Case 3: Representative lowest-utility posterior distributions for the one-sensor problem. The bb [PITH_FULL_IMAGE:figures/full_fig_p021_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Case 3: Estimated expected utility, utility variance, and their trade-off for the two-sensor problem. [PITH_FULL_IMAGE:figures/full_fig_p022_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Case 3: Estimated mean–variance objective with different values of [PITH_FULL_IMAGE:figures/full_fig_p022_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Case 3: Histogram distributions of uKL(ξ ∗ , Y ) and uKL(ξ ∗ λ , Y ) for the two-sensor problem. The values before and after ± are the mean and standard deviation, respectively. 0 5 10 15 20 25 30 35 40 45 50 update 1.8 2.0 2.2 2.4 2.6 J ( ) BO init BO search BO opt (a) λ = 0 0 5 10 15 20 25 30 35 40 45 50 update 1.2 1.4 1.6 1.8 2.0 2.2 2.4 J ( ) BO init BO search BO opt (b) λ = 0.5 [PITH_FULL_IMAGE:figu… view at source ↗
Figure 23
Figure 23. Figure 23: Case 3: BO histories for the two-sensor problem. [PITH_FULL_IMAGE:figures/full_fig_p023_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Case 3: Representative lowest-utility posterior distributions for the two-sensor problem. The bb [PITH_FULL_IMAGE:figures/full_fig_p024_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Case 4: Estimated expected utility, utility variance, and their trade-off under seven obstacle [PITH_FULL_IMAGE:figures/full_fig_p025_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Case 4: Estimated mean–variance objective with different values of [PITH_FULL_IMAGE:figures/full_fig_p025_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Case 4: Histogram distributions of uKL(ξ ∗ , Y ) and uKL(ξ ∗ λ , Y ) for the one-sensor problem with building #4. The values before and after ± are the mean and standard deviation, respectively. 0.0 0.2 0.4 0.6 0.8 1.0 zx 0.0 0.2 0.4 0.6 0.8 1.0 zy 0.60 0.68 0.75 0.83 0.90 0.98 1.05 1.12 1.20 (a) Evaluation locations overlaid on the ob￾jective contour 0 4 8 12 16 20 24 28 32 36 40 update 0.85 0.90 0.95 1.… view at source ↗
Figure 28
Figure 28. Figure 28: Case 4: BO with λ = 0.5 for the one-sensor problem with building #4. 26 [PITH_FULL_IMAGE:figures/full_fig_p026_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: Case 4: Representative lowest-utility posterior distributions for the one-sensor problem with bb [PITH_FULL_IMAGE:figures/full_fig_p027_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: Case 4: Histogram distributions of u(ξ ∗ , Y ) and u(ξ ∗ λ , Y ) for the two-sensor problem with build￾ing #5. The values before and after ± are the mean and standard deviation, respectively [PITH_FULL_IMAGE:figures/full_fig_p028_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: Case 4: Representative lowest-utility posterior distributions for the two-sensor problem with bb [PITH_FULL_IMAGE:figures/full_fig_p028_31.png] view at source ↗
read the original abstract

We propose a variance-penalized formulation of Bayesian optimal experimental design for nonlinear models that augments the classical expected utility criterion with a penalty on utility variability, yielding a mean--variance objective that promotes robust experimental performance. To evaluate this objective, we develop Monte Carlo estimators for the expected utility, its second moment, and the resulting utility variance using prior sampling, thereby avoiding explicit posterior sampling. We then derive leading-order bias and variance expressions using conditional delta-method arguments. The objective is optimized using Bayesian optimization with common random samples to reduce noise. Numerical examples, including a linear-Gaussian benchmark, a nonlinear test problem, and contaminant source inversion in diffusion fields, demonstrate that the proposed approach identifies designs with substantially reduced variability while maintaining competitive expected utility.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 3 minor

Summary. The paper proposes a mean-variance risk-aware formulation of Bayesian optimal experimental design for nonlinear models. It augments the classical expected-utility criterion with a penalty on utility variability, derives Monte Carlo estimators for the mean and second moment of the utility from prior sampling (avoiding explicit posterior draws), obtains leading-order bias and variance corrections via conditional delta-method expansions, optimizes the resulting objective with Bayesian optimization that re-uses common random numbers, and illustrates the method on a linear-Gaussian benchmark, a nonlinear test problem, and a contaminant-source inversion example in diffusion fields.

Significance. If the delta-method approximations remain accurate across the design space, the work supplies a computationally attractive route to robust experimental designs that explicitly trade off expected utility against its variability. The prior-sampling strategy and common-random-number optimization are practical strengths that could be adopted in settings where full posterior sampling is prohibitive.

major comments (3)
  1. [§3.3] §3.3 (conditional delta-method derivations): the first-order Taylor expansions for bias and variance of the Monte Carlo utility estimators are presented without uniform remainder bounds or higher-order error controls. In nonlinear models the utility surface can exhibit strong curvature or non-differentiability with respect to the design variables; if the neglected terms are comparable in magnitude to the variance penalty itself, the optimizer may select designs whose reported robustness is an artifact of the approximation rather than a genuine reduction in utility variability.
  2. [§5] Numerical examples (§5): the reported reductions in utility variance are shown via point estimates only; no Monte Carlo standard errors, sensitivity plots versus sample size N, or cross-validation against exact posterior sampling are provided. Without these diagnostics it is impossible to determine whether the observed gains exceed the approximation error of the delta-method estimators.
  3. [§4] Optimization procedure (§4): the claim that common random numbers suffice to make the mean-variance objective sufficiently smooth for Bayesian optimization is stated without a convergence analysis or comparison to gradient-based alternatives that could exploit the explicit delta-method gradients.
minor comments (3)
  1. Notation for the risk-aversion weight λ is introduced without an explicit statement of its admissible range or scaling relative to the utility units.
  2. [§5] Figure captions in §5 should include the Monte Carlo sample size N used for each panel to allow direct reproducibility.
  3. The linear-Gaussian benchmark could usefully report the exact closed-form mean-variance optimum for direct comparison with the numerical optimizer.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us identify areas where the manuscript can be strengthened. We address each major comment below and outline the planned revisions.

read point-by-point responses

  1. Referee: [§3.3] §3.3 (conditional delta-method derivations): the first-order Taylor expansions for bias and variance of the Monte Carlo utility estimators are presented without uniform remainder bounds or higher-order error controls. In nonlinear models the utility surface can exhibit strong curvature or non-differentiability with respect to the design variables; if the neglected terms are comparable in magnitude to the variance penalty itself, the optimizer may select designs whose reported robustness is an artifact of the approximation rather than a genuine reduction in utility variability.

    Authors: We agree that the conditional delta-method yields only leading-order approximations and that uniform remainder bounds are not provided. This is a genuine limitation of the current theoretical development. In the revised manuscript we will expand the discussion in §3.3 to state the smoothness assumptions required for the expansion to be valid, note the absence of higher-order controls, and caution that the approximation may degrade for utilities with strong curvature or non-differentiability. We will also add a short numerical check comparing the delta-method estimates against direct Monte Carlo estimates of higher moments in the examples to illustrate practical accuracy. revision: partial

  2. Referee: [§5] Numerical examples (§5): the reported reductions in utility variance are shown via point estimates only; no Monte Carlo standard errors, sensitivity plots versus sample size N, or cross-validation against exact posterior sampling are provided. Without these diagnostics it is impossible to determine whether the observed gains exceed the approximation error of the delta-method estimators.

    Authors: We accept that the numerical results in §5 are reported only as point estimates and lack the requested statistical diagnostics. In the revision we will augment all tables and figures with Monte Carlo standard errors, add sensitivity plots that vary the Monte Carlo sample size N, and, for the linear-Gaussian benchmark where exact posterior quantities are available, include a direct comparison against exact sampling to verify that the observed variance reductions exceed the delta-method approximation error. revision: yes

  3. Referee: [§4] Optimization procedure (§4): the claim that common random numbers suffice to make the mean-variance objective sufficiently smooth for Bayesian optimization is stated without a convergence analysis or comparison to gradient-based alternatives that could exploit the explicit delta-method gradients.

    Authors: Common random numbers are employed to induce positive correlation across design evaluations and thereby reduce the variance of the estimated mean-variance objective, which empirically improves the stability of the Bayesian optimization surrogate. We do not supply a formal convergence analysis of this stochastic optimization procedure, as that would require substantial additional theory. In the revision we will add a brief discussion citing relevant variance-reduction literature and include an empirical comparison (in an appendix) of optimization trajectories obtained with and without common random numbers. While the delta-method supplies explicit gradients, the overall objective remains noisy; Bayesian optimization was selected for its robustness in this setting. A systematic comparison with gradient-based methods is feasible but lies outside the present scope and will be noted as future work. revision: partial

Circularity Check

0 steps flagged

No circularity in mean-variance BOED derivation chain

full rationale

The paper defines the mean-variance objective explicitly as E[U] minus lambda times Var(U), then constructs Monte Carlo estimators for E[U] and E[U^2] directly from prior samples of the utility function without any parameter fitting to the target quantities or self-referential definitions. The conditional delta-method bias and variance expressions are derived as standard asymptotic approximations applied to those estimators. No equations reduce to tautologies, no fitted inputs are renamed as predictions, and no load-bearing uniqueness claims or ansatzes are imported via self-citation. The derivation remains self-contained against external statistical primitives.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard Bayesian modeling assumptions and the validity of Monte Carlo integration plus delta-method approximations; no new free parameters beyond the implicit risk-weighting coefficient or invented entities are introduced.

free parameters (1)
  • risk aversion weight
    Scalar that trades off expected utility against utility variance in the mean-variance objective; its value is chosen by the user or tuned during optimization.
axioms (1)
  • domain assumption Existence of a well-defined prior distribution over model parameters from which samples can be drawn
    Required for the prior-sampling Monte Carlo estimators described in the abstract.

pith-pipeline@v0.9.0 · 5415 in / 1159 out tokens · 44684 ms · 2026-05-14T21:50:22.568639+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 1 internal anchor

  1. [1]

    Alexanderian

    A. Alexanderian. Optimal experimental design for infinite-dimensional Bayesian inverse prob- lems governed by PDEs: A review.Inverse Problems, 37(4):043001, 2021

    work page 2021

  2. [2]

    J. Beck, B. M. Dia, L. F. Espath, Q. Long, and R. Tempone. Fast Bayesian experimental design: Laplace-based importance sampling for the expected information gain.Computer Methods in Applied Mechanics and Engineering, 334:523–553, 2018

    work page 2018

  3. [3]

    Chaloner and I

    K. Chaloner and I. Verdinelli. Bayesian experimental design: A review.Statistical Science, 10(3), 1995

    work page 1995

  4. [4]

    Duong, T

    D.-L. Duong, T. Helin, and J. R. Rojo-Garcia. Stability estimates for the expected utility in Bayesian optimal experimental design.Inverse Problems, 39(12):125008, 2023

    work page 2023

  5. [5]

    Englezou, T

    Y. Englezou, T. W. Waite, and D. C. Woods. Approximate Laplace importance sampling for the estimation of expected Shannon information gain in high-dimensional Bayesian design for nonlinear models.Statistics and Computing, 32(5):82, 2022

    work page 2022

  6. [6]

    A layered multiple importance sampling scheme for focused optimal Bayesian experimental design

    C. Feng and Y. M. Marzouk. A layered multiple importance sampling scheme for focused optimal Bayesian experimental design. arXiv:1903.11187, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  7. [7]

    Foster, M

    A. Foster, M. Jankowiak, E. Bingham, P. Horsfall, Y. W. Teh, T. Rainforth, and N. Goodman. Variational Bayesian optimal experimental design. In H. Wallach, H. Larochelle, A. Beygelz- imer, F. d'Alch´ e-Buc, E. Fox, and R. Garnett, editors,Advances in Neural Information Pro- cessing Systems, volume 32. Curran Associates, Inc., 2019

    work page 2019

  8. [8]

    Foster, M

    A. Foster, M. Jankowiak, M. O’Meara, Y. W. Teh, and T. Rainforth. A unified stochastic gradient approach to designing Bayesian-optimal experiments. In S. Chiappa and R. Calandra, editors,Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, volume 108 ofProceedings of Machine Learning Research, pages 2959–2969...

    work page 2020

  9. [9]

    P. I. Frazier.Bayesian Optimization, page 255–278. INFORMS, 2018

    work page 2018

  10. [10]

    X. Huan, J. Jagalur, and Y. Marzouk. Optimal experimental design: Formulations and com- putations.Acta Numerica, 33:715–840, 2024

    work page 2024

  11. [11]

    Huan and Y

    X. Huan and Y. Marzouk. Gradient-based stochastic optimization methods in Bayesian ex- perimental design.International Journal for Uncertainty Quantification, 4(6):479–510, 2014

    work page 2014

  12. [12]

    Huan and Y

    X. Huan and Y. M. Marzouk. Simulation-based optimal Bayesian experimental design for nonlinear systems.Journal of Computational Physics, 232(1):288–317, 2013. 34

    work page 2013

  13. [13]

    J. D. Jakeman, R. D. White, B. G. van Bloemen Waanders, A. Alexandarian, and D. P. Kouri. Risk-averse goal-oriented optimal experimental design using non-linear models. Sandia National Laboratories (SNL-NM), Albuquerque, NM (United States), 2023

    work page 2023

  14. [14]

    D. R. Jones, M. Schonlau, and W. J. Welch. Efficient global optimization of expensive black- box functions.Journal of Global Optimization, 13(4):455–492, 1998

    work page 1998

  15. [15]

    Kleinegesse and M

    S. Kleinegesse and M. U. Gutmann. Bayesian experimental design for implicit models by mutual information neural estimation. In H. D. III and A. Singh, editors,Proceedings of the 37th International Conference on Machine Learning, volume 119 ofProceedings of Machine Learning Research, pages 5316–5326. PMLR, 2020

    work page 2020

  16. [16]

    D. P. Kouri, J. D. Jakeman, and J. Gabriel Huerta. Risk-adapted optimal experimental design. SIAM/ASA Journal on Uncertainty Quantification, 10(2):687–716, 2022

    work page 2022

  17. [17]

    K. P. Kusumo, K. Kuriyan, S. Vaidyaraman, S. Garc´ ıa-Mu˜ noz, N. Shah, and B. Chachuat. Risk mitigation in model-based experiment design: A continuous-effort approach to optimal campaigns.Computers & Chemical Engineering, 159:107680, 2022

    work page 2022

  18. [18]

    D. V. Lindley. On a measure of the information provided by an experiment.The Annals of Mathematical Statistics, 27(4):986–1005, 1956

    work page 1956

  19. [19]

    Q. Long, M. Scavino, R. Tempone, and S. Wang. Fast estimation of expected information gains for Bayesian experimental designs based on Laplace approximations.Computer Methods in Applied Mechanics and Engineering, 259:24–39, 2013

    work page 2013

  20. [20]

    Markowitz

    H. Markowitz. Portfolio selection.The Journal of Finance, 7(1):77–91, 1952

    work page 1952

  21. [21]

    Moˇ ckus.On Bayesian methods for seeking the extremum, page 400–404

    J. Moˇ ckus.On Bayesian methods for seeking the extremum, page 400–404. Springer Berlin Heidelberg, 1975

    work page 1975

  22. [22]

    Nogueira

    F. Nogueira. Bayesian Optimization: Open source constrained global optimization tool for Python, 2014

    work page 2014

  23. [23]

    A. M. Overstall, J. M. McGree, and C. C. Drovandi. An approach for finding fully Bayesian op- timal designs using normal-based approximations to loss functions.Statistics and Computing, 28(2):343–358, 2017

    work page 2017

  24. [24]

    Rainforth, R

    T. Rainforth, R. Cornish, H. Yang, A. Warrington, and F. Wood. On nesting Monte Carlo estimators. In J. Dy and A. Krause, editors,Proceedings of the 35th International Conference on Machine Learning, volume 80 ofProceedings of Machine Learning Research, pages 4267–

    work page

  25. [25]

    Rainforth, A

    T. Rainforth, A. Foster, D. R. Ivanova, and F. Bickford Smith. Modern Bayesian experimental design.Statistical Science, 39(1), 2024

    work page 2024

  26. [26]

    R. T. Rockafellar and S. Uryasev. The fundamental risk quadrangle in risk management, optimization and statistical estimation.Surveys in Operations Research and Management Science, 18(1–2):33–53, 2013

    work page 2013

  27. [27]

    J. O. Royset. Risk-adaptive approaches to stochastic optimization: A survey.SIAM Review, 67(1):3–70, 2025. 35

    work page 2025

  28. [28]

    E. G. Ryan, C. C. Drovandi, J. M. McGree, and A. N. Pettitt. A review of modern computa- tional algorithms for Bayesian optimal design.International Statistical Review, 84(1):128–154, 2015

    work page 2015

  29. [29]

    K. J. Ryan. Estimating expected information gains for experimental designs with applica- tion to the random fatigue-limit model.Journal of Computational and Graphical Statistics, 12(3):585–603, 2003

    work page 2003

  30. [30]

    Shahriari, K

    B. Shahriari, K. Swersky, Z. Wang, R. P. Adams, and N. de Freitas. Taking the human out of the loop: A review of Bayesian optimization.Proceedings of the IEEE, 104(1):148–175, 2016

    work page 2016

  31. [31]

    Shapiro, D

    A. Shapiro, D. Dentcheva, and A. Ruszczynski.Lectures on Stochastic Programming: Modeling and Theory, Third Edition. Society for Industrial and Applied Mathematics, 2021

    work page 2021

  32. [32]

    Strutz and A

    D. Strutz and A. Curtis. Variational Bayesian experimental design for geophysical applications: Seismic source location, amplitude versus offset inversion, and estimating co2 saturations in a subsurface reservoir.Geophysical Journal International, 236(3):1309–1331, 2023

    work page 2023

  33. [33]

    P. E. Valenzuela, C. R. Rojas, and H. Hjalmarsson. Uncertainty in system identification: Learning from the theory of risk.IFAC-PapersOnLine, 48(28):1053–1058, 2015

    work page 2015

  34. [34]

    X. Wang, Y. Jin, S. Schmitt, and M. Olhofer. Recent advances in Bayesian optimization.ACM Computing Surveys, 55(13s):1–36, 2023

    work page 2023

  35. [35]

    White, J

    R. White, J. Jakeman, B. van Bloemen Waanders, D. Kouri, and A. Alexanderian. Exploring risk-averse design criteria for sequential optimal experimental design in a Bayesian setting. Sandia National Laboratories (SNL-NM), Albuquerque, NM (United States), 2021

    work page 2021

  36. [36]

    White, B

    R. White, B. van Bloemen Waanders, J. Jakeman, D. Kouri, and A. Alexanderian. Risk-aware Bayesian goal-oriented optimal experimental design. Sandia National Laboratories (SNL-NM), Albuquerque, NM (United States), 2022

    work page 2022

  37. [37]

    R. D. White, J. D. Jakeman, B. G. van Bloemen Waanders, D. P. Kouri, and A. Alexandar- ian. A Bayesian approach to designing experiments that account for risk. Sandia National Laboratories (SNL-NM), Albuquerque, NM (United States), 2023. 36

    work page 2023