Pith. sign in

REVIEW 3 major objections 4 minor 105 references

Target Shapley effects for high-dimensional correlated inputs can be estimated from one sample of failure points by rewriting closed Sobol indices as density ratios and learning those densities with normalizing flows.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-12 11:53 UTC pith:O4M3J2CT

load-bearing objection Solid methods paper that removes a real dimensionality barrier for reliability-oriented Shapley effects; the NF accuracy caveat is real but already shown and does not sink the contribution. the 3 major comments →

arxiv 2606.26826 v2 pith:O4M3J2CT submitted 2026-06-25 stat.ME

High-dimensional reliability-oriented Shapley effect estimation with Normalizing Flows

Lucas Monteiro , J\'er\^ome Morio , Julien Demange-Chryst , Fran\c{c}ois Bachoc This is my paper
classification stat.ME MSC 62G0762P3065C0590B25
keywords target Shapley effectsreliability-oriented sensitivity analysisnormalizing flowsclosed Sobol indiceshigh-dimensional density estimationfailure samplesuncertainty quantification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

When a black-box model can fail, engineers need to know which uncertain inputs drive the failure. Target Shapley effects answer that question even when the inputs are correlated, but existing estimators break down above roughly ten dimensions or demand many extra model runs. This paper shows that the closed target Sobol indices that underlie the Shapley effects can be rewritten as a simple function of the failure probability and an expectation of a density ratio under the conditional distribution of the inputs given failure. Those high-dimensional conditional densities are then estimated by normalizing flows trained on a single existing sample of failure points. The same sample also supplies a resampling-based error bar. The result is a complete estimation-plus-uncertainty pipeline that re-uses the rare-event sample already collected for reliability analysis and therefore requires no further calls to the expensive model.

Core claim

A single N-sample of failure points, together with an estimate of the failure probability, is sufficient to recover the full vector of target Shapley effects in dimensions greater than ten: rewrite every closed target Sobol index as (p_t/(1-p_t)) times (E[f_{X_u|F_t}/f_{X_u}]-1), estimate the conditional densities with normalizing flows, aggregate the indices by Monte-Carlo over permutations, and quantify the three sources of error by resampling that same failure sample.

What carries the argument

The density-ratio identity T-S^c_u = (p_t/(1-p_t))(E_{X_u|F_t}[f_{X_u|F_t}(X_u)/f_{X_u}(X_u)]-1), estimated by splitting the failure sample between a normalizing-flow density estimator and a Monte-Carlo average of the ratio, then aggregated via ApproShapley-style permutation sampling.

Load-bearing premise

A normalizing-flow architecture trained on roughly half the failure points produces density estimates accurate enough that every required ratio expectation yields a usable closed Sobol index, even though the paper itself notes the lack of consistency guarantees and shows that under-powered flows can converge to wrong values.

What would settle it

On a known high-dimensional Gaussian-linear model, replace the normalizing-flow density estimator by a deliberately under-parameterized architecture (or by a deliberately misspecified base measure) and check whether the recovered Shapley vector systematically deviates from the closed-form reference values while the reported error bars fail to cover them.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes a scheme to estimate reliability-oriented (target) Shapley effects for models with d ≳ 10 correlated inputs from a single N-sample of failing points obtained in a prior reliability analysis. It rewrites the closed target Sobol indices as T-S^c_u = (p_t/(1-p_t))(E_{X_u|F_t}[f_{X_u|F_t}(X_u)/f_{X_u}(X_u)]-1) (Eq. 13), estimates the (possibly high-dimensional) conditional densities with Normalizing Flows (NICE architecture), aggregates the indices via permutation sampling (ApproShapley), and quantifies estimation error from the same failing sample via nested resampling and a Gaussian-mixture approximation (Algorithms 1–4). The approach is illustrated on a 15-dimensional Gaussian-linear model with known closed-form targets and a 10-dimensional fire-spread model, with comparisons to KDE and GMM density estimators.

Significance. If the numerical performance generalizes, the method fills a genuine gap: existing target-Shapley estimators (MC or importance-sampling nearest-neighbor) become unreliable beyond roughly 8–9 dimensions, while industrial reliability models routinely exceed that size. The algebraic rewriting is clean and proved (Appendix A), the single-sample error procedure is a practical contribution that avoids extra model evaluations, and the accompanying code repository supports reproducibility. The work therefore has clear applied value for ROSA, even though it remains an empirical density-estimation pipeline rather than a fully analyzed estimator.

major comments (3)
  1. [§4.2–4.3, Fig. 1] §4.2–4.3 and Fig. 1(c): the central claim that NFs yield usable closed indices for every required u rests on an unproved premise. The paper itself states that consistency guarantees are lacking and shows that an under-powered architecture (K=5 layers, N=20 000) converges to a systematically wrong value for a high-order index while a richer architecture recovers the truth. Because ApproShapley aggregates many such high-order terms, a systematic density bias can shift the final Shapley vector; the manuscript needs either a more systematic study of architecture adequacy versus |u| or an adaptive selection rule before the high-d claim can be considered established.
  2. [§5, Algorithms 2–4] §5, Algorithms 2–4 and Fig. 2: the proposed error quantification re-uses the same finite failing sample and the same fixed NF architecture. Consequently it cannot detect model-misspecification bias of the flow. The comparison with 60 fully independent repetitions already indicates that the Gaussian-mixture intervals tend to understate the true variability; this limitation should be stated more prominently and, if possible, mitigated (e.g., by architecture perturbation or hold-out density diagnostics).
  3. [§6, Appendix D] §6 and Appendix D: the numerical evidence is limited to two examples. The Gaussian-linear case is favorable (conditional densities remain close to Gaussian), while the fire-spread case has only d=10. Appendix D further shows that good low-order pair-plots do not guarantee accurate high-order marginals—the very densities needed for |u| near d. Additional experiments with non-Gaussian failure regions and d>15, or quantitative diagnostics of high-order marginal fidelity, are required to support the “high-dimensional” claim.
minor comments (4)
  1. [Appendix C] The early-stopping criterion (Appendix C) is described only qualitatively; the precise patience and validation-loss threshold used for all reported runs should be stated so that the experiments are fully reproducible.
  2. [§4.3] Notation for the split proportion α is introduced in Algorithm 1 but never justified beyond the default 1/2; a short sensitivity check would be helpful.
  3. [Fig. 2] Figure 2 caption and surrounding text mix “theoretical Sobol values” with “NF+MC estimates” without clearly indicating that both share the same permutation sample; a clarifying sentence would avoid confusion.
  4. A few typographical inconsistencies appear (e.g., “sensivity” for “sensitivity”, occasional missing spaces around operators); a careful proof-reading pass is recommended.

Circularity Check

0 steps flagged

No load-bearing circularity: the closed-index rewriting is an identity derived from Bayes and change-of-measure, NF density estimates are trained and then evaluated against independent closed-form or large-budget references, and self-citations only supply the prior low-d ROSA-Shapley setting.

full rationale

The central derivation chain begins with the algebraic identity (11)–(13) that rewrites every closed target Sobol index T-S^c_u as a Monte-Carlo expectation of a density ratio under the failure-conditional law; the short proof in Appendix A uses only the definition of conditional probability, Bayes’ rule and the fact that 1_{F_t} is Bernoulli, none of which presuppose the numerical value of the index. The subsequent estimation steps (Algorithm 1) replace the unknown conditional density by a normalizing-flow approximant trained on a split of the given failure sample and evaluate the ratio on the held-out half; the resulting numbers are compared, in Section 6, to reference values obtained either from closed-form Gaussian formulae or from a double-Monte-Carlo estimator that uses a far larger independent budget. Those references are therefore external to the NF fit. The permutation aggregation (ApproShapley) and the resampling error procedure (Algorithms 2–4) are likewise standard Monte-Carlo devices applied to the already-computed indices; they do not redefine the target quantities. Self-citations to earlier ROSA-Shapley papers supply only the problem formulation and the low-dimensional baselines that the present work improves upon; they are not invoked as uniqueness theorems that force the NF architecture or the rewriting. Consequently the claimed high-dimensional estimator does not reduce by construction to its own inputs, and the circularity score remains at the minor-self-citation level.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central claim rests on standard measure-theoretic identities for conditional densities and Shapley values, the change-of-variables formula for diffeomorphisms, and the empirical premise that a sufficiently flexible NF trained on a few thousand failure points approximates the required high-dimensional conditionals well enough for the Monte-Carlo ratios to be useful. No new physical entities are postulated; free parameters are the usual neural-architecture and resampling hyperparameters.

free parameters (3)
  • NF architecture (K layers, NICE vs RealNVP, neurons, early-stopping patience)
    Chosen by hand; K=5 vs K=10 and early stopping materially affect bias/variance of high-order Sobol indices (Fig. 1, Appendix C).
  • sample split proportion α
    Fixed at 1/2; controls trade-off between density estimation and Monte-Carlo evaluation of the ratio expectation.
  • number of permutations M and resampling loops J,L,P
    Set to M=300, J=5, L=5, P=40; directly determine the width of the reported Gaussian-mixture uncertainty bands.
axioms (4)
  • standard math Change-of-variables formula for C1 diffeomorphisms yields an exact density for any invertible transport of a base Gaussian (Eq. 14–15).
    Invoked throughout §4.2 to justify that an NF can represent f_{X_u|F_t}.
  • standard math Closed target Sobol indices admit the density-ratio representation T-S^c_u = (p_t/(1-p_t))(E[f_{X_u|F_t}/f_{X_u}]-1) (Eq. 13).
    Proved in Appendix A from Bayes and the Bernoulli variance of the failure indicator; used as the estimation target.
  • domain assumption A finite sample of failure points drawn from f_{X|F_t} is already available from a prior reliability analysis, and f_X (hence its marginals) is known.
    Stated in §2.2 and §4; without it the whole pipeline cannot start.
  • ad hoc to paper NICE (or similar) flows with a modest number of layers are flexible enough to approximate the conditional densities that arise in the failure domain of the models considered.
    No universal approximation theorem is invoked for the finite-sample, finite-architecture regime actually used; performance is demonstrated only empirically.

pith-pipeline@v1.1.0-grok45 · 29600 in / 3080 out tokens · 30054 ms · 2026-07-12T11:53:22.945544+00:00 · methodology

0 comments
read the original abstract

This article presents a new estimation scheme for the reliability-oriented Shapley effects when there is a large number of correlated input variables in the model, using a unique sample of failure points. To do so, we first propose a new writing of the reliability-oriented closed Sobol indices involving the marginal densities conditionally to the failure, which may be high-dimensional. Then, we propose to estimate these densities with the available failing samples using Normalizing Flows, powerful tools from generative modeling that enable the estimation of complex high-dimensional densities. In addition, we provide an error estimation procedure relying on the same sample of failing points, which constitutes a new contribution for the estimation of target Shapley effects. Finally, we illustrate our methodology on numerical use-cases, discuss insightful features of our approach and provide prospects for the future.

Figures

Figures reproduced from arXiv: 2606.26826 by Fran\c{c}ois Bachoc, J\'er\^ome Morio, Julien Demange-Chryst, Lucas Monteiro.

Figure 1
Figure 1. Figure 1: The boxplots correspond the application of 50 repetitions of Algorithm 1 for three closed target Sobol indices. For each index, we [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: The boxplots correspond the application of [PITH_FULL_IMAGE:figures/full_fig_p014_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Gaussian Linear case results. Reference values vs. estimation of target Shapley effects in three settings : with theoretical values of Sobol indices, with estimations of Sobol indices with NF+MC, with 60 independent repetitions of the estimation scheme in Section 4 [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 2
Figure 2. Figure 2: Gaussian Linear case results. Reference values vs. estimation of target Shapley effects in three settings : with theoretical values of Sobol indices, with estimations of Sobol indices with NF+MC, with 60 independent repetitions of the estimation scheme in Section 4. be biased for some of them. Overall, the estimation of the target Shapley effects is good and never take absurd values such as negative values… view at source ↗
Figure 3
Figure 3. Figure 3: Gaussian Linear case results. Reference values vs. estimation of target Shapley effects with Normalizing Flows (+ our error estimation), with KDE (+ our error estimation), and with GMM (+ our error estimation). n° Input variable Symbol (unit) Distribution 1 Fuel depth δ (cm) log N(2.19, 0.517) 2 Fuel particle area-to-volume ratio σ (cm−1 ) log N(3.31, 0.294) 3 Fuel particle low heat content h (Kcal · kg−1 … view at source ↗
Figure 3
Figure 3. Figure 3: Gaussian Linear case results. Reference values vs. estimation of target Shapley effects with Normalizing Flows (+ our error estimation), with KDE (+ our error estimation), and with GMM (+ our error estimation). approaches and to discriminate the best one. However, it does allow us to verify that our approach using NFs can estimate the target Sobol indices fairly accurately overall, and thus provides a corr… view at source ↗
Figure 4
Figure 4. Figure 4: Fire-spread case results. Reference values vs. estimation of target Shapley effects with Normalizing Flows (+ our error estimation), with KDE (+ our error estimation), and with GMM (+ our error estimation). 7. Conclusion In this article, we tackled the challenge of estimating the reliability-oriented target Shapley effects for high-dimensional models when input variables are dependent. These sensitivity in… view at source ↗
Figure 4
Figure 4. Figure 4: Fire-spread case results. Reference values vs. estimation of target Shapley effects with Normalizing Flows (+ our error estimation), with KDE (+ our error estimation), and with GMM (+ our error estimation). reliability analysis. This is a major contribution for the practitioner as it avoids to repeat the estimation scheme like several other methods, hence avoiding additional calls to the model. Finally, we… view at source ↗

Review history (2 revisions) →

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

105 extracted references · 8 linked inside Pith

  1. [1]

    Estimating wildfire behavior and effects

    Albini, F.A., 1976. Estimating wildfire behavior and effects. Gen. Tech. Rep. INT-GTR-30. Ogden, UT: U.S. Department of Agriculture, Forest Service, Intermountain Forest and Range Experiment Station. 92 p

    1976

  2. [2]

    Probabilistic failure analysis by importance sampling Markov chain simulation

    Au., S.K., 2004. Probabilistic failure analysis by importance sampling Markov chain simulation. Journal of Engineering Mechanics 130, 303 – 311

    2004

  3. [3]

    Estimation of small failure probabilities in high dimensions by subset simulation

    Au, S.K., Beck, J.L., 2001. Estimation of small failure probabilities in high dimensions by subset simulation. Probabilistic Engineering Mechanics 16, 263–277

    2001

  4. [4]

    Sensitivity analysis to unobserved confounding with copula-based normalizing flows

    Balgi, S., Braun, M., Peña, J.M., Daoud, A., 2025. Sensitivity analysis to unobserved confounding with copula-based normalizing flows. International Journal of Approximate Reasoning 187, 109531

    2025

  5. [5]

    An approximation theory framework for measure-transport sampling algorithms

    Baptista, R., Hosseini, B., Kovachki, N., Marzouk, Y., Sagiv, A., 2025. An approximation theory framework for measure-transport sampling algorithms. Mathematics of Computation 94, 1863–1909

    2025

  6. [6]

    SHAFF: Fast and consistent SHApley eFfect estimates via random Forests, in: Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, PMLR

    Bénard, C., Biau, G., Da Veiga, S., Scornet, E., 2022. SHAFF: Fast and consistent SHApley eFfect estimates via random Forests, in: Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, PMLR. pp. 5563–5582. 25

    2022

  7. [7]

    Shapley effects for sensitivity analysis with dependent inputs: bootstrap and kriging-based algorithms

    Benoumechiara, N., Elie-Dit-Cosaque, K., 2019. Shapley effects for sensitivity analysis with dependent inputs: bootstrap and kriging-based algorithms. ESAIM: ProcS 65, 266–293

    2019

  8. [8]

    Triangular transformations of measures

    Bogachev, V.I., Kolesnikov, A.V., Medvedev, K.V., 2005. Triangular transformations of measures. Sbornik: Mathematics 196, 309–335

    2005

  9. [9]

    A new uncertainty importance measure

    Borgonovo, E., 2007. A new uncertainty importance measure. Reliability Engineering & System Safety 92, 771–784

    2007

  10. [10]

    ImportancesamplingforSobol’indicesestimation

    Boucharif, H., Morio, J., Rochet, P., 2026. ImportancesamplingforSobol’indicesestimation. Preprint: https://arxiv.org/abs/2507.05958

    arXiv 2026

  11. [11]

    Asymptotic approximations for multinormal integrals

    Breitung, K., 1984. Asymptotic approximations for multinormal integrals. Journal of Engineering Mechanics-asce 110, 357–366

    1984

  12. [12]

    Block-diagonal covariance estimation and application to the Shapley effects in sensitivity analysis

    Broto, B., Bachoc, F., Clouvel, L., Martinez, J.M., 2022. Block-diagonal covariance estimation and application to the Shapley effects in sensitivity analysis. SIAM/ASA Journal on Uncertainty Quan- tification 10, 379–403

    2022

  13. [13]

    Variance reduction for estimation of Shapley effects and adaptation to unknown input distribution

    Broto, B., Bachoc, F., Depecker, M., 2020. Variance reduction for estimation of Shapley effects and adaptation to unknown input distribution. SIAM/ASA Journal on Uncertainty Quantification 8, 693–716

    2020

  14. [14]

    Gaussian linear approximation for the estimation of the Shapley effects

    Broto, B., Bachoc, F., Depecker, M., Martinez, J.M., 2021. Gaussian linear approximation for the estimation of the Shapley effects. SIAM/ASA Journal on Uncertainty Quantification 9, 1132–1151

    2021

  15. [15]

    Introduction to Rare Event Simulation

    Bucklew, J., 2010. Introduction to Rare Event Simulation. 1st ed., Springer Publishing Company, Incorporated

    2010

  16. [16]

    PolynomialcalculationoftheShapleyvaluebasedonsampling

    Castro, J., Gómez, D., Tejada, J., 2009. PolynomialcalculationoftheShapleyvaluebasedonsampling. Computers & Operations Research 36, 1726–1730

    2009

  17. [17]

    Modelling moisture damping for fire spread in a mixture of live and dead fuels

    Catchpole, E.A., Catchpole, W.R., 1991. Modelling moisture damping for fire spread in a mixture of live and dead fuels. International Journal of Wildland Fire 1, 101–106

    1991

  18. [18]

    Reliability-oriented sensitivity analysis under probabilistic model uncertainty – Application to aerospace systems

    Chabridon, V., 2018. Reliability-oriented sensitivity analysis under probabilistic model uncertainty – Application to aerospace systems. Ph.D. thesis. Université Clermont Auvergne

    2018

  19. [19]

    Generalized Sobol sensitivity indices for dependent variables: numerical methods

    Chastaing, G., Gamboa, F., Prieur, C., 2015. Generalized Sobol sensitivity indices for dependent variables: numerical methods. Journal of Statistical Computation and Simulation 85, 1306–1333

    2015

  20. [20]

    Gaussianization, in: Advances in Neural Information Processing Sys- tems, MIT Press

    Chen, S., Gopinath, R., 2000. Gaussianization, in: Advances in Neural Information Processing Sys- tems, MIT Press. p. 423–429

    2000

  21. [21]

    Sensitivity analysis of a fire spread model in chaparral landscape

    Clark, R., Hope, A., Tarantola, S., Gatelli, D., Dennison, P., Moritz, M., 2008. Sensitivity analysis of a fire spread model in chaparral landscape. Fire Ecology 4, 1–13

    2008

  22. [22]

    Comparison of affine and rational quadratic spline coupling and autoregressive flows through robust statistical tests

    Coccaro, A., Letizia, M., Reyes-González, H., Torre, R., 2024. Comparison of affine and rational quadratic spline coupling and autoregressive flows through robust statistical tests. Symmetry 16, 942

    2024

  23. [23]

    Adaptive multilevel splitting for rare event analysis

    Cérou, F., Guyader, A., 2007. Adaptive multilevel splitting for rare event analysis. Stochastic Analysis and Applications 25, 417–443

    2007

  24. [24]

    Sequential Monte Carlo for rare event estimation

    Cérou, F., del Moral, P., Furon, T., Guyader, A., 2012. Sequential Monte Carlo for rare event estimation. Statistics and Computing 22, 795–808

    2012

  25. [25]

    Global sensitivity analysis with dependence measures

    Da Veiga, S., 2015. Global sensitivity analysis with dependence measures. Journal of Statistical Computation and Simulation 85, 1283–1305. 26

    2015

  26. [26]

    Efficient estimation of Sobol’ indices of any order from a single input/output sample

    Da Veiga, S., Gamboa, F., Lagnoux, A., Klein, T., Prieur, C., 2024. Efficient estimation of Sobol’ indices of any order from a single input/output sample. Preprint:https://hal.science/ hal-04052837

    2024

  27. [27]

    REIN: Reliability estimation via importance sampling with nor- malizing flows

    Dasgupta, A., Johnson, E.A., 2024. REIN: Reliability estimation via importance sampling with nor- malizing flows. Reliability Engineering & System Safety 242, 109729

    2024

  28. [28]

    Rare event modeling with self-regularized normalizing flows: what can we learn from a single failure? Preprint:https://arxiv.org/abs/2502.21110

    Dawson, C., Tran, V., Li, M.Z., Fan, C., 2025. Rare event modeling with self-regularized normalizing flows: what can we learn from a single failure? Preprint:https://arxiv.org/abs/2502.21110

    Pith/arXiv arXiv 2025

  29. [29]

    Nouvelles méthodes d’estimation par échantillonnage préférentiel pour l’analyse de sensibilité et l’estimation d’évènements rares

    Demange-Chryst, J., 2024. Nouvelles méthodes d’estimation par échantillonnage préférentiel pour l’analyse de sensibilité et l’estimation d’évènements rares. Ph.D. thesis. Université de Toulouse

    2024

  30. [30]

    Shapley effect estimation in reliability-oriented sen- sitivity analysis with correlated inputs by importance sampling

    Demange-Chryst, J., Bachoc, F., Morio, J., 2023. Shapley effect estimation in reliability-oriented sen- sitivity analysis with correlated inputs by importance sampling. International Journal for Uncertainty Quantification 13, 1–37

    2023

  31. [31]

    Variational autoencoder with weighted samples for high-dimensional non-parametric adaptive importance sampling

    Demange-Chryst, J., Bachoc, F., Morio, J., Krauth, T., 2024. Variational autoencoder with weighted samples for high-dimensional non-parametric adaptive importance sampling. Transactions on Machine Learning Research (TMLR)

    2024

  32. [32]

    Mesures de sensibilité de Borgonovo : estimation des indices d’ordre un et supérieur, et application à l’analyse de fiabilité

    Derennes, P., 2019. Mesures de sensibilité de Borgonovo : estimation des indices d’ordre un et supérieur, et application à l’analyse de fiabilité. Ph.D. thesis. Université Toulouse 3 Paul Sabatier (UT3 Paul Sabatier)

    2019

  33. [33]

    An introduction to variational autoencoders

    Diederik, P.K., Max, W., 2019. An introduction to variational autoencoders. Foundations and Trends®in Machine Learning 12, 307–392

    2019

  34. [34]

    NICE: Non-linear independent components estimation, in: ICLR 2015, Workshop Track Proceedings

    Dinh, L., Krueger, D., Bengio, Y., 2015. NICE: Non-linear independent components estimation, in: ICLR 2015, Workshop Track Proceedings

    2015

  35. [35]

    Density estimation using Real NVP, in: International Conference on Learning Representations

    Dinh, L., Sohl-Dickstein, J., Bengio, S., 2017. Density estimation using Real NVP, in: International Conference on Learning Representations

    2017

  36. [36]

    AK-MCS: An active learning reliability method combining Kriging and Monte Carlo simulation

    Echard, B., Gayton, N., Lemaire, M., 2011. AK-MCS: An active learning reliability method combining Kriging and Monte Carlo simulation. Structural Safety 33, 145–154

    2011

  37. [37]

    A combined importance sampling and kriging reliability method for small failure probabilities with time-demanding numerical models

    Echard, B., Gayton, N., Lemaire, M., Relun, N., 2013. A combined importance sampling and kriging reliability method for small failure probabilities with time-demanding numerical models. Reliability Engineering & System Safety 111, 232–240

    2013

  38. [38]

    Variance-based reliability sensitivity with dependent inputs using failure samples

    Ehre, M., Papaioannou, I., Straub, D., 2024. Variance-based reliability sensitivity with dependent inputs using failure samples. Structural Safety 106, 102396

    2024

  39. [39]

    Multivariate Bernoulli Ho- effding Decomposition: From Theory to Sensitivity Analysis

    Ferrere, B., Bousquet, N., Gamboa, F., Loubes, J.M., Muré, J., 2025. Multivariate Bernoulli Ho- effding Decomposition: From Theory to Sensitivity Analysis. Preprint:https://hal.science/ hal-05301780

    2025

  40. [40]

    Shapley effects and proportional marginal effects for global sensitivity analysis: application to computed tomography scan organ dose estimation

    Foucault, A., Il Idrissi, M., Iooss, B., Ancelet, S., 2023. Shapley effects and proportional marginal effects for global sensitivity analysis: application to computed tomography scan organ dose estimation. Preprint:https://arxiv.org/abs/2306.01353

    arXiv 2023

  41. [41]

    Rare event probability learning by normalizing flows

    Gao, Z., Zhang, D., Daniel, L., Boning, D.S., 2023. Rare event probability learning by normalizing flows. Preprint:https://arxiv.org/abs/2310.19167

    Pith/arXiv arXiv 2023

  42. [42]

    Handbook of Uncertainty Quantification

    Ghanem, R., Higdon, D., Owhadi, H., 2017. Handbook of Uncertainty Quantification. Springer Cham. 27

    2017

  43. [43]

    A simple algorithm for global sensitivity analysis with Shapley effects

    Goda, T., 2021. A simple algorithm for global sensitivity analysis with Shapley effects. Reliability Engineering & System Safety 213, 107702

    2021

  44. [44]

    Generative adversarial nets, in: Proceedings of the 28th International Conference on Neural Information Processing Systems, p

    Goodfellow, I.J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y., 2014. Generative adversarial nets, in: Proceedings of the 28th International Conference on Neural Information Processing Systems, p. 2672–2680

    2014

  45. [45]

    Exact and invariant second-moment code format

    Hasofer, A.M., Lind, N.C., 1974. Exact and invariant second-moment code format. Journal of the Engineering Mechanics Division 100, 111–121

    1974

  46. [46]

    Proportional marginal effects for global sensitivity analysis

    Herin, M., Il Idrissi, M., Chabridon, V., Iooss, B., 2024. Proportional marginal effects for global sensitivity analysis. SIAM/ASA Journal on Uncertainty Quantification 12, 667–692

    2024

  47. [47]

    A class of statistics with asymptotically normal distribution

    Hoeffding, W., 1948. A class of statistics with asymptotically normal distribution. Annals of Mathe- matical Statistics 19, 308–334

    1948

  48. [48]

    Manifold-based Shapley explanations for high dimen- sional correlated features

    Hu, X., Zhu, M., Feng, Z., Stanković, L., 2024. Manifold-based Shapley explanations for high dimen- sional correlated features. Neural Networks 180, 106634

    2024

  49. [49]

    Neural autoregressive flows, in: Proceed- ings of the 35th International Conference on Machine Learning, PMLR

    Huang, C.W., Krueger, D., Lacoste, A., Courville, A., 2018. Neural autoregressive flows, in: Proceed- ings of the 35th International Conference on Machine Learning, PMLR. pp. 2078–2087

    2018

  50. [50]

    Assessing small failure probabilities by AK–SS: An active learning method combining kriging and subset simulation

    Huang, X., Chen, J., Zhu, H., 2016. Assessing small failure probabilities by AK–SS: An active learning method combining kriging and subset simulation. Structural Safety 59, 86–95

    2016

  51. [51]

    Hoeffding decomposition of functions of random dependent variables

    Il Idrissi, M., Bousquet, N., Gamboa, F., Iooss, B., Loubes, J.M., 2025. Hoeffding decomposition of functions of random dependent variables. Journal of Multivariate Analysis 208, 105444

    2025

  52. [52]

    Developments and applications of Shapley effects to reliability-oriented sensitivity analysis with correlated inputs

    Il Idrissi, M., Chabridon, V., Iooss, B., 2021. Developments and applications of Shapley effects to reliability-oriented sensitivity analysis with correlated inputs. Environmental Modelling & Software 143, 105115

    2021

  53. [53]

    Shapley effects for sensitivity analysis with correlated inputs: Comparisons with Sobol’ indices, numerical estimation and applications

    Iooss, B., Prieur, C., 2019. Shapley effects for sensitivity analysis with correlated inputs: Comparisons with Sobol’ indices, numerical estimation and applications. International Journal for Uncertainty Quantification 9, 493–514

    2019

  54. [54]

    Variational diffusion models, in: Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P., Vaughan, J

    Kingma, D., Salimans, T., Poole, B., Ho, J., 2021. Variational diffusion models, in: Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P., Vaughan, J. (Eds.), Advances in Neural Information Pro- cessing Systems, pp. 21696–21707

    2021

  55. [55]

    Adam: A method for stochastic optimization

    Kingma, D.P., Ba, J., 2017. Adam: A method for stochastic optimization. Preprint:https://arxiv. org/abs/1412.6980

    Pith/arXiv arXiv 2017

  56. [56]

    Normalizing flows: An introduction and review of current methods

    Kobyzev, I., Prince, S., Brubaker, M.A., 2020. Normalizing flows: An introduction and review of current methods. IEEE Transactions on Pattern Analysis and Machine Intelligence 43, 3964–3979

    2020

  57. [57]

    The expressive power of a class of normalizing flow models, in: Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, PMLR

    Kong, Z., Chaudhuri, K., 2020. The expressive power of a class of normalizing flow models, in: Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, PMLR. pp. 3599–3609

    2020

  58. [58]

    The neural autoregressive distribution estimator, in: Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, pp

    Larochelle, H., Murray, I., 2011. The neural autoregressive distribution estimator, in: Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, pp. 29–37

    2011

  59. [59]

    Data-driven sensitivity indices for models with dependent inputs using polynomial chaos expansions

    Liu, Z., Choe, Y., 2021. Data-driven sensitivity indices for models with dependent inputs using polynomial chaos expansions. Structural Safety 88, 101984

    2021

  60. [60]

    A unified approach to interpreting model predictions, in: Proceedings of the 31st International Conference on Neural Information Processing Systems, p

    Lundberg, S.M., Lee, S.I., 2017. A unified approach to interpreting model predictions, in: Proceedings of the 31st International Conference on Neural Information Processing Systems, p. 4768–4777. 28

    2017

  61. [61]

    Bounding the estimation error of sampling-based Shapley value approximation with/without stratifying

    Maleki, S., Tran-Thanh, L., Hines, G., Rahwan, T., Rogers, A., 2014. Bounding the estimation error of sampling-based Shapley value approximation with/without stratifying. Preprint:https: //arxiv.org/abs/1306.4265

    Pith/arXiv arXiv 2014

  62. [62]

    Variance-based sensitivity indices for models with dependent inputs

    Mara, T., Tarantola, S., 2012. Variance-based sensitivity indices for models with dependent inputs. Reliability Engineering & System Safety 107, 115–121

    2012

  63. [63]

    Non-parametric methods for global sensitivity analysis of model output with dependent inputs

    Mara, T.A., Tarantola, S., Annoni, P., 2015. Non-parametric methods for global sensitivity analysis of model output with dependent inputs. Environmental Modelling & Software 72, 173–183

    2015

  64. [64]

    Uncertainty quantification for deep regression using contextualised normalizing flows

    Marco, A., Kirwan, J.D., Toumpa, A., Gerasimou, S., 2025. Uncertainty quantification for deep regression using contextualised normalizing flows. Preprint:https://arxiv.org/abs/2512.00835

    arXiv 2025

  65. [65]

    Statistical developments for target and conditional sensitivity anal- ysis: Application on safety studies for nuclear reactor

    Marrel, A., Chabridon, V., 2021. Statistical developments for target and conditional sensitivity anal- ysis: Application on safety studies for nuclear reactor. Reliability Engineering & System Safety 214, 107711

    2021

  66. [66]

    Randomized numerical linear algebra: Foundations & algorithms

    Martinsson, P.G., Tropp, J., 2021. Randomized numerical linear algebra: Foundations & algorithms. Preprint:https://arxiv.org/abs/2002.01387

    Pith/arXiv arXiv 2021

  67. [67]

    Sampling via Measure Transport: An Introduction

    Marzouk, Y., Moselhy, T., Parno, M., Spantini, A., 2016. Sampling via Measure Transport: An Introduction. Springer International Publishing. p. 1–41

    2016

  68. [68]

    A review of uncertainty quantification for density estimation

    McDonald, S., Campbell, D., 2021. A review of uncertainty quantification for density estimation. Statistics Surveys 15

    2021

  69. [69]

    Sampling permutations for Shapley value estimation

    Mitchell, R., Cooper, J., Frank, E., Holmes, G., 2022. Sampling permutations for Shapley value estimation. Journal of Machine Learning Research 23, 1–46

    2022

  70. [70]

    Estimation of Rare Event Probabilities in Complex Aerospace and Other Systems

    Morio, J., Balesdent, M., 2015. Estimation of Rare Event Probabilities in Complex Aerospace and Other Systems. Woodhead Publishing

    2015

  71. [71]

    ProvablyaccurateShapleyvalueestimationvialeveragescoresampling, in: The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025

    Musco, C., Witter, R.T., 2025. ProvablyaccurateShapleyvalueestimationvialeveragescoresampling, in: The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025

    2025

  72. [72]

    Improving the weighting strategy in KernelSHAP, in: Explainable Artificial Intelligence, Springer Nature Switzerland

    Olsen, L.H.B., Jullum, M., 2026. Improving the weighting strategy in KernelSHAP, in: Explainable Artificial Intelligence, Springer Nature Switzerland. pp. 194–218

    2026

  73. [73]

    Sobol’indicesandShapleyvalue

    Owen, A.B., 2014. Sobol’indicesandShapleyvalue. SIAM/ASAJournalonUncertaintyQuantification 2, 245–251

    2014

  74. [74]

    On Shapley value for measuring importance of dependent inputs

    Owen, A.B., Prieur, C., 2017. On Shapley value for measuring importance of dependent inputs. SIAM/ASA Journal on Uncertainty Quantification 5, 986–1002

    2017

  75. [75]

    Reliability-oriented sensitivity analysis using Shapley additive explanations and polynomial chaos expansion

    Palar, P., Renganathan, A., 2024. Reliability-oriented sensitivity analysis using Shapley additive explanations and polynomial chaos expansion. AIAA SCITECH 2024 Forum

    2024

  76. [76]

    Variance-based reliability sensitivity analysis and the FORMα- factors

    Papaioannou, I., Straub, D., 2021. Variance-based reliability sensitivity analysis and the FORMα- factors. Reliability Engineering & System Safety 210, 107496

    2021

  77. [77]

    FORM-based global reliability sensitivity analysis of systems with multiple failure modes

    Papaioannou, I., Straub, D., 2025. FORM-based global reliability sensitivity analysis of systems with multiple failure modes. Reliability Engineering & System Safety 260, 110974

    2025

  78. [78]

    Normalizing flows for probabilistic modeling and inference

    Papamakarios, G., Eric Nalisnick, E., Rezende, D.J., Mohamed, S., Lakshminarayanan, B., 2021. Normalizing flows for probabilistic modeling and inference. Journal of Machine Learning Research 22, 1–64. 29

    2021

  79. [79]

    Masked autoregressive flow for density estimation, in: Advances in Neural Information Processing Systems, p

    Papamakarios, G., Pavlakou, T., Murray, I., 2017. Masked autoregressive flow for density estimation, in: Advances in Neural Information Processing Systems, p. 2338–2347

    2017

  80. [80]

    Efficient evaluation of reliability-oriented sensitivity indices

    Perrin, G., Defaux, G., 2019. Efficient evaluation of reliability-oriented sensitivity indices. Journal of Scientific Computing 79, 1433–1455

    2019

Showing first 80 references.