Pith. sign in

REVIEW 3 major objections 6 minor 66 references

Surrogate modeling for uncertainty quantification in nonlinear dynamics

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The chapter classifies time-dependent UQ problems by input complexity and argues each class has a matching surrogate strategy.

desk verdict A readable review chapter whose simple-vs-complex input taxonomy is useful, but Section 5's 'dramatic decrease' claim is not supported by the evidence actually shown. read the letter →

arxiv 2507.12358 v1 pith:MEPY5TDT submitted 2025-07-16 stat.CO stat.APstat.ML

S. Marelli , S. Schär , B. Sudret This is my paper · ORCID
classification stat.COstat.APstat.ML
keywords uncertaintyquantificationsurrogatemodelingdynamicalsystemspolynomialchaosexpansionsNARXmodelstimewarpingmanifoldlearningcoupledoscillator
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Uncertainty quantification for nonlinear dynamical systems usually demands so many model evaluations that direct Monte Carlo is impractical, so the paper asks which surrogate model should be used when the quantity of interest is a whole time trajectory. The chapter's answer is a classification: the intrinsic complexity of the input excitation dictates the surrogate family. Excitations that can be encoded by roughly ten parameters, such as a few sinusoids, are fundamentally simple and are handled by sparse polynomial chaos expansions, often combined with principal component analysis and stochastic time warping. Excitations with rich, not-compactly-representable time content, such as earthquake ground motions or turbulent wind, are fundamentally complex and require autoregressive models with exogenous inputs, extended as polynomial-chaos NARX for uncertain systems and manifold-NARX for strongly nonlinear or high-dimensional maps. The coupled-oscillator application is meant to show that the two NARX extensions reduce approximation error substantially compared with classical NARX.

What carries the argument

The central machinery is the simple-versus-complex input classification in Section 3.1, which routes each problem to a different surrogate family. For simple inputs, the chapter builds on principal component analysis of the training trajectory matrix, retaining m eigenmodes that explain 95–99% of variance, followed by independent sparse polynomial chaos expansions of the score coefficients; when curves oscillate with different phase or frequency, stochastic time warping applies an invertible map $\mathrm{TW}(t,x)=\beta(x)f(t)$ (e.g., affine $t\mapsto kt+\varphi$) to align each trajectory to a reference before the PCA-PCE step. For complex inputs, the engine is the one-step-ahead NARX predictor $y(t+\delta t)=\mathcal{M}(\text{lags of }y\text{ and }x;c)$ with model orders $n_y$ and $n_x$; PC-NARX makes the coefficient vector $c(\xi)$ a sparse polynomial chaos expansion in the uncertain structural parameters $\xi$, and mNARX defines auxiliary quantities $z_i(t)$ through recursion or physics (e.g., the lower-mass response in the oscillator example) to assemble a low-dimensional manifold $\zeta(t)$ on which separate NARX models are trained.

What would settle it

Recompute the coupled-oscillator comparison on a random held-out set of, say, one hundred out-of-sample parameter-and-excitation realizations and report a single aggregate error measure (e.g., normalized root-mean-square error over all traces and all time steps) for classical NARX, PC-NARX, and mNARX; the claim of a 'dramatic decrease' would be confirmed only if the aggregate gap is large and consistent, and refuted if errors overlap.

Watch

Extended reading notes

Core claim

On its own terms, the chapter's discovery is a decision rule: time-dependent surrogate challenges are driven by the intrinsic dimensionality of the input excitation rather than by the fact that the response is time-dependent. For fundamentally simple inputs, a finite O(10)-parameter representation suffices, so time-frozen sparse polynomial chaos expansions work; principal component analysis makes this cheaper by fitting PCEs on a handful of score coefficients rather than on each time step, and stochastic time warping—an invertible reparametrization of each trajectory so it aligns with a reference curve—restores accuracy when oscillatory responses differ in phase and frequency. For fundamentally complex inputs, the one-step-ahead NARX map is the base tool, but classical NARX fails when system parameters are uncertain or the map is strongly nonlinear; polynomial-chaos NARX therefore replaces the NARX coefficients with sparse PCEs of the structural parameters, and manifold-NARX builds an auxiliary state manifold so that a sequence of simpler autoregressive models can replace one intractable map. The two-oscillator case study is presented as evidence that both extensions achieve a 'dramatic decrease in approximation error' relative to classical NARX on out-of-sample signals.

Load-bearing premise

The load-bearing premise behind the headline comparison is that the two visually selected out-of-sample traces shown for each method in Section 5 are representative of overall prediction accuracy, since no aggregate error statistics or sensitivity to the chosen lags and polynomial degrees are reported.

Editorial extensions

If this is right

  • A practitioner facing a time-dependent UQ problem can start from the input: if the excitation admits a compact parameterization, PCA-PCE or time warping should suffice; otherwise the fallback is an autoregressive strategy.
  • For fundamentally simple inputs, the PCA preprocessing cuts the number of sparse PCE fits from the number of time steps (often $10^3$–$10^4$) to a handful of score coefficients without changing the predictor's structure.
  • Time warping converts a problem in which frozen-time sparse PCE fails, such as a Bouc-Wen oscillator with random frequency and phase, into one where a standard surrogate trained on warped curves is accurate.
  • PC-NARX carries classical NARX over to parametric uncertainty by treating the regression coefficients themselves as functions of the uncertain system parameters, trained once on an experimental design spanning those parameters.
  • mNARX attacks strong nonlinearity by splitting one difficult autoregressive map into several simpler ones, as in the two-mass oscillator where the lower-mass response is predicted first and then used as an input to the upper-mass model.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension of the classification is a phase-transition experiment: sweep excitation complexity continuously on one oscillator and record when PCA-PCE accuracy degrades below NARX-based accuracy; the chapter does not perform this sweep.
  • The mNARX recipe currently relies on engineers to choose auxiliary quantities such as the lower-mass response; an algorithmic version that discovers those quantities from data would extend the method to systems without a clear physical hierarchy.
  • The classification suggests a decision boundary based on effective input dimensionality or spectral content; quantifying that boundary would turn the taxonomy into a practical workflow selector.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This book chapter reviews surrogate modeling techniques for uncertainty quantification of dynamical systems. It proposes a taxonomy of time-dependent problems based on the intrinsic complexity of the input excitation ("fundamentally simple" versus "fundamentally complex"), and maps each class to a surrogate strategy: PCA-based sparse polynomial chaos expansions for simple inputs, stochastic time warping for oscillatory responses, and autoregressive models with exogenous inputs (NARX) and their extensions (PC-NARX and mNARX) for complex inputs. The main methodological content is presented as a review, with Sections 2 and 4 giving the standard equations for sparse PCE and NARX, and Sections 3 and 5 illustrating the methods on dynamical system examples. The central practical claim is that PC-NARX and mNARX outperform classical NARX in the Section 5 case study, with the paper asserting an "overall dramatic decrease in approximation error."

Significance. If the presented classification and the Section 5 comparative claims are accepted, the chapter would provide practitioners with a useful decision rule for selecting surrogates for dynamical UQ. The review has clear organizational value: the taxonomy in Section 3.1 is intuitive, the standard equations for sparse PCE (Eqs. 4-13) and NARX (Eqs. 20-35) are correctly presented, and Section 4.3 honestly states the known limitations of classical NARX. The chapter also benefits from clear connections to the authors' prior work, which is appropriate for a review. However, the case study in Section 5 currently provides only visual evidence for the comparative claims, with no aggregate error metrics or sensitivity analysis. As the paper is a review rather than a novel methodological contribution, the standard derivations are not the main issue; the unsupported quantitative claims are the main weakness and are fixable within the manuscript's scope.

major comments (3)
  1. [Section 5 (PC-NARX comparison)] The comparison between classical NARX and PC-NARX is favorable to PC-NARX by construction. Classical NARX is trained on nED = 100 pooled trajectories with varying system parameters (Section 5, first configuration), yet Section 4.3 states that classical NARX "can only represent a stationary system" and that "a new model must be trained if the physical properties of the system are changed." Thus the baseline is a model whose assumptions are violated by the experimental design, and the observed improvement is expected regardless of the merits of PC-NARX. The text claims a "significant improvement in accuracy" without reporting even a single aggregate error metric (e.g., normalized RMSE over the out-of-sample set). Please add quantitative error statistics over the full out-of-sample set, and either train the classical NARX baseline in a setting it can represent or explicitly frame the comparison as a stress test that classical NARX is known to fail.
  2. [Section 5 (mNARX comparison)] The mNARX comparison does not control for model complexity. Classical NARX uses polynomial degree d = 7 with nx = 3 and ny = 4, while mNARX uses d = 4 for yu and d = 6 for ys, both with nx = 2 and ny = 3 (Table 5). The text states that these settings were chosen because "we obtained the best results" for the classical model, but no selection procedure or sensitivity analysis is reported for either method. As a result, the "overall dramatic decrease in approximation error" attributed to the manifold approach could be due to the different hyperparameter budgets rather than to the method itself. Please report aggregate error metrics and a sensitivity study over the model orders (ny, nx, d), or use a common hyperparameter budget for both methods.
  3. [Section 5 (overall evidence)] The central comparative claim rests on two visually selected out-of-sample traces per method (Figures 3 and 4). No aggregate error statistics, confidence intervals, or quantitative comparison across the out-of-sample set are provided, so the displayed traces could be the most favorable subset. The abstract and conclusion assert a classification-driven surrogate selection rule; the Section 5 evidence currently does not demonstrate that rule quantitatively. Please add summary error metrics (e.g., mean and quantiles of trajectory-wise normalized error over the out-of-sample set) for both configurations, and state the number of traces used for assessment.
minor comments (6)
  1. [Section 3.3.2] The text contains a typo: "Bouc-Wenn" should be "Bouc-Wen".
  2. [Section 5 (weak damping paragraph)] There is a typo in "weak damping causees the top mass response" — "causees" should be "causes".
  3. [Section 5 (mNARX construction)] The sentence "we take advantage of the of Eq. (41)" is grammatically incomplete; it should read "we take advantage of the structure of Eq. (41)" or similar.
  4. [Figures 3 and 4 captions] The captions state that in all plots the black dashed line is the reference, the blue line is the NARX approximation, the magenta line is the PC-NARX/mNARX approximation, and the red line is the error. Since the left column shows only the classical NARX and the right column only the PC-NARX/mNARX, the caption is confusing; please clarify which line appears in which column.
  5. [Section 2.2] The text uses both "experimental design (ED)" and "DOE" for the same concept; please standardize the abbreviation for clarity.
  6. [References] Meles et al. (2025) is listed as "Submitted"; for a book chapter, please update the reference if a published version is available.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the chapter reviews established surrogate methods and validates them on out-of-sample traces; the taxonomy is organizational, and the self-citations are not load-bearing.

full rationale

The chapter is a review rather than a derivation. Its section-level claims (PCA-PCE for simple inputs, time warping for phase-shifted oscillators, PC-NARX and mNARX for complex inputs) are presented as methodological summaries with references to prior peer-reviewed work, including Blatman and Sudret (2013), Mai and Sudret (2017), Spiridonakos and Chatzi (2015), and Schär et al. (2024). The Section 5 comparisons use out-of-sample realizations: the surrogate is trained on n_ED = 100 realizations and the figures show traces not in the training set. No equation in the paper reduces a reported prediction to a fitted parameter: the warping coefficients and NARX coefficients are fitted from training trajectories and then emulated by PCE, but the validation is against fresh trajectories. The distinction between 'fundamentally simple' and 'fundamentally complex' inputs is a taxonomic framing credited to Meles et al. (2025) and Schär (2025), which are author-affiliated works, but the chapter's technical content does not depend on a uniqueness theorem or on a result derived solely from those citations. The Section 5 'dramatic decrease' claim rests on weak evidence (two selected traces, no aggregate errors, different model orders), but that is a statistical-support and correctness concern, not circularity. The clearest self-citation is organizational rather than load-bearing, so the appropriate score is 2.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The chapter's central content is a review, so the ledger focuses on the modeling assumptions of the presented methods and the hand-selected hyperparameters of the illustrative case study. The four free parameters listed are all chosen by the authors without reported sensitivity analysis; they do affect the strength of the case study's comparative claims. No new physical or mathematical entities are postulated; the autoregressive manifold of mNARX is from prior work (Schär et al. 2024), not introduced here.

free parameters (4)
  • NARX polynomial degree d = 3 for PC-NARX; 7 for classical NARX in weak damping; 4 and 6 for the two mNARX stages
    Chosen without reported cross-validation; directly affects the size of the regressor basis and thus the comparison (Sec. 5).
  • Model orders ny and nx = ny=4, nx=5 for PC-NARX; ny=3, nx=2 or 3 for mNARX
    Hand-selected; no grid search or sensitivity analysis is shown (Sec. 5).
  • Maximum sparse PCE degree dPCE = 10
    Adaptive selection is claimed, but the candidate basis bound is a modeling choice that influences coefficient surrogates (Sec. 5).
  • PCA variance retention threshold epsilon = 1% to 5% (Sec. 3.2)
    General guidance; not used in the Section 5 case study, but governs PCA-PCE methods elsewhere in the chapter.
assumptions (5)
  • domain assumption Input random variables X are independent for PCE construction.
    Eqs. (5)-(7) build tensor-product orthogonal polynomials assuming independent components; stated in Sec. 2.3.
  • standard math The model response Y has finite variance for the PCE expansion.
    Required for the convergence of the polynomial chaos expansion in Eq. (4); a standard Hilbert space result.
  • domain assumption The time warping mapping TW is invertible for each realization.
    Eqs. (14)-(15) require inverting TW to map predictions back to physical time; stated in Sec. 3.3.2.
  • domain assumption The system is stationary for classical NARX models.
    Acknowledged as a limitation in Sec. 4.3; PC-NARX handles parametric variability by making coefficients depend on structural parameters.
  • domain assumption Response at time t depends only on past inputs and states, not future ones (causality).
    Eqs. (20)-(22) formalize this causal structure; it underpins all autoregressive approaches in Sec. 4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Surrogate modeling for uncertainty quantification in nonlinear dynamics." pith.science (2026). https://pith.science/paper/MEPY5TDT

@misc{pith2026250712358,
  author       = {Pith},
  title        = {Pith review of: Surrogate modeling for uncertainty quantification in nonlinear dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MEPY5TDT}},
  note         = {Machine review of arXiv:2507.12358}
}
read the original abstract

Predicting the behavior of complex systems in engineering often involves significant uncertainty about operating conditions, such as external loads, environmental effects, and manufacturing variability. As a result, uncertainty quantification (UQ) has become a critical tool in modeling-based engineering, providing methods to identify, characterize, and propagate uncertainty through computational models. However, the stochastic nature of UQ typically requires numerous evaluations of these models, which can be computationally expensive and limit the scope of feasible analyses. To address this, surrogate models, i.e., efficient functional approximations trained on a limited set of simulations, have become central in modern UQ practice. This book chapter presents a concise review of surrogate modeling techniques for UQ, with a focus on the particularly challenging task of capturing the full time-dependent response of dynamical systems. It introduces a classification of time-dependent problems based on the complexity of input excitation and discusses corresponding surrogate approaches, including combinations of principal component analysis with polynomial chaos expansions, time warping techniques, and nonlinear autoregressive models with exogenous inputs (NARX models). Each method is illustrated with simple application examples to clarify the underlying ideas and practical use.

Figures

Figures reproduced from arXiv: 2507.12358 by the authors.

Figure 1
Figure 1. Bouc-Wen oscillator with random input parameters under sinusoidal forcing (after [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Surrogate model of Bouc-Wen oscillator trajectories with random parameters- and [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Comparison of classical NARX and PC-NARX surrogates on the strong-damping, [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparison of classical NARX and mNARX surrogates on the weak-damping case [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 46 canonical work pages

  1. [1]

    Ramirez, and M

    Acuña, G., C. Ramirez, and M. Curilem (2012). Comparing NARX and NARMAX models using ANN and SVM for cash demand forecasting for ATM . In The 2012 International Joint Conference on Neural Networks (IJCNN) , pp.\ 1--6

    work page 2012

  2. [2]

    Aguirre, L. and S. Billings (1993). Relationship between the structure and performance of identified nonlinear polynomial models. Research report, The University of Sheffield

    work page 1993

  3. [3]

    Bellemans, Z

    Aversano, G., A. Bellemans, Z. Li, A. Coussement, O. Gicquel, and A. Parente (2019). Application of reduced-order models based on PCA & Kriging for the development of digital twins of reacting flow applications . Computers & Chemical Engineering\/ 121 , 422--441

    work page 2019

  4. [4]

    Billings, S. A. (2013). Nonlinear system identification: NARMAX methods in the time, frequency, and spatio-temporal domains . Chichester, West Sussex, United Kingdom: John Wiley & Sons, Inc

    2013

  5. [5]

    Blatman, G. and B. Sudret (2011). Adaptive sparse polynomial chaos expansion based on least angle regression. J. Comput. Phys.\/ 230\/ (6), 2345–2367

    work page 2011

  6. [6]

    Blatman, G. and B. Sudret (2013). Sparse polynomial chaos expansions of vector-valued response quantities. In G. Deodatis (Ed.), Proc. 11th Int. Conf. Struct. Safety and Reliability (ICOSSAR'2013) , New York, USA , pp.\ 1--8

    work page 2013

  7. [7]

    Chen, S., X. X. Wang, and C. J. Harris (2008). NARX -based nonlinear system identification using orthogonal least-squares basis hunting. IEEE Transactions on Control Systems Technology\/ 16\/ (1), 78--84

    work page 2008

  8. [8]

    Lebrun, A

    Chevreuil, M., R. Lebrun, A. Nouy, and P. Rai (2015). A least-squares method for sparse low rank approximation of multivariate functions. SIAM/ASA Journal Uncertainty Quantification\/ 3\/ (1), 897--921

    work page 2015

Show all 66 references
  1. [9]

    Choi, M., T. P. Sapsis, and G. E. Karniadakis (2014). On the equivalence of dynamically orthogonal and bi-orthogonal methods: Theory and numerical simulations. Journal of Computational Physics\/ 270 , 1--20

    2014

  2. [10]

    Coca, D. and S. A. Billings (2001). Non-linear system identification using wavelet multiresolution models. International Journal of Control\/ 74\/ (18), 1718--1736

    2001

  3. [11]

    Deshmukh, A. P. and J. T. Allison (2017). Design of dynamic systems using surrogate models of derivative functions. Journal of Mechanical Design\/ 139\/ (10), 101402

    2017

  4. [12]

    Validi, and G

    Doostan, A., A. Validi, and G. Iaccarino (2013). Non-intrusive low-rank separated approximation of high-dimensional stochastic models. Computer Methods in Applied Mechanics and Engineering\/ 263 , 42--55

    2013

  5. [13]

    Hastie, I

    Efron, B., T. Hastie, I. Johnstone, and R. Tibshirani (2004). Least angle regression . The Annals of Statistics\/ 32\/ (2), 407 -- 499

    2004

  6. [14]

    Gao, Y., S. Liu, F. Li, and Z. Liu (2016). Fault detection and diagnosis method for cooling dehumidifier based on LS-SVM NARX model. International Journal of Refrigeration\/ 61 , 69--81

    2016

  7. [15]

    Gupta, and S

    Garg, S., H. Gupta, and S. Chakraborty (2022). Assessment of DeepONet for reliability analysis of stochastic nonlinear dynamical systems. arXiv:2201.13145

    2022 arXiv

  8. [16]

    van der Steen , P

    Gerritsma, M., J.-B. van der Steen , P. Vos, and G. Karniadakis (2010). Time-dependent generalized polynomial chaos. Journal of Computational Physics\/ 229\/ (22), 8333--8363

    2010

  9. [17]

    Ghanem, R. and P. Spanos (2003). Stochastic F inite E lements: A S pectral A pproach\/ (2nd ed.). Courier D over P ublications, Mineola

    2003

  10. [18]

    Ghanem, R. G. and P. D. Spanos (1991). Stochastic finite elements -- A spectral approach . Springer Verlag, New York. (Reedited by Dover Publications, Mineola, 2003)

    1991

  11. [19]

    Bengio, and A

    Goodfellow, I., Y. Bengio, and A. Courville (2016). Deep learning . MIT Press Ltd

    2016

  12. [20]

    Mahadevan, S

    Guo, Y., S. Mahadevan, S. Matsumoto, S. Taba, and D. Watanabe (2023). Investigation of surrogate modeling options with high-dimensional input and output. AIAA Journal\/ 61\/ (3), 1334--1348

    2023

  13. [21]

    Hu, Z., J. Fang, R. Zheng, M. Li, B. Gao, and L. Zhang (2024). Efficient model predictive control of boiler coal combustion based on NARX neutral network. Journal of Process Control\/ 134 , 103158

    2024

  14. [22]

    Jolliffe, I. T. (2002). Principal component analysis\/ (2 ed.). Springer Series in Statistics. New York: Springer Verlag (Second Edition)

    2002

  15. [23]

    Jonkman, J. (2009). TurbSim User's Guide: Version 1.50 . Technical report, National Renewable Energy Laboratory

    2009

  16. [24]

    Kocijan, J. (2012). Plenary lecture 1: Dynamic GP models: An overview and recent developments. In Proceedings of the 6th International Conference on Applied Mathematics, Simulation, Modelling , ASM'12, Stevens Point, Wisconsin, USA, pp.\ 12. World Scientific and Engineering Ac...

    2012

  17. [25]

    Konakli, K. and B. Sudret (2016a). Global sensitivity analysis using low-rank tensor approximations. Reliability Engineering & System Safety\/ 156 , 64--83

    2016

  18. [26]

    Konakli, K. and B. Sudret (2016b). Polynomial meta-models with canonical low-rank approximations: Numerical insights and comparison to sparse polynomial chaos expansions. Journal of Computational Physics\/ 321 , 1144--1169

    2016

  19. [27]

    Langeron, Y., K. T. Huynh, and A. Grall (2021). A root location-based framework for degradation modeling of dynamic systems with predictive maintenance perspective. Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability\/ 235\/ (2), 253--267

    2021

  20. [28]

    Le Ma \^ i tre, O. P., L. Mathelin, O. M. Knio, and M. Y. Hussaini (2010). Asynchronous time integration for polynomial chaos expansion of uncertain periodic dynamics . Discrete and Continuous Dynamical Systems\/ 28\/ (1), 199--226

    2010

  21. [29]

    Levin, A. and K. Narendra (1996). Control of nonlinear dynamical systems using neural networks. II . Observability , identification, and control. IEEE Transactions on Neural Networks\/ 7\/ (1), 30--42

    1996

  22. [30]

    Zhou, and Y

    Li, D., J. Zhou, and Y. Liu (2021). Recurrent-neural-network-based unscented Kalman filter for estimating and compensating the random drift of mems gyroscopes in real time. Mechanical Systems and Signal Processing\/ 147 , 107057

    2021

  23. [31]

    Luchtenburg, D. M., S. L. Brunton, and C. W. Rowley (2014). Long-time uncertainty propagation using generalized polynomial chaos and flow map composition. Journal of Computational Physics\/ 274 , 783--802

    2014

  24. [32]

    Marelli, and B

    L\"uthen, N., S. Marelli, and B. Sudret (2021). Sparse polynomial chaos expansions: Literature survey and benchmark. SIAM/ASA Journal on Uncertainty Quantification\/ 9\/ (2), 593--649

    2021

  25. [33]

    Marelli, and B

    L\"uthen, N., S. Marelli, and B. Sudret (2022). Automatic selection of basis-adaptive sparse polynomial chaos expansions for engineering applications. Int. J. Uncertainty Quantification\/ 12\/ (3), 49--74

    2022

  26. [34]

    Mai, C. and B. Sudret (2017). Surrogate models for oscillatory systems using sparse polynomial chaos expansions and stochastic time warping. SIAM/ASA Journal on Uncertainty Quantification\/ 5\/ (1), 540--571

    2017

  27. [35]

    Mai, C.-V. (2016). Polynomial chaos expansions for uncertain dynamical systems. Applications in earthquake engineering . Ph.\ D. thesis, ETH Z \"u rich, Switzerland

    2016

  28. [36]

    Mai, C.-V., M. D. Spiridonakos, E. Chatzi, and B. Sudret (2016). Surrogate modeling for stochastic dynamical systems by combining nonlinear autoregressive with exogeneous input models and polynomial chaos expansions. International Journal of Uncertainty Quantification\/ 6\/ (4...

    2016

  29. [37]

    Mattson, S. G. and S. M. Pandit (2006). Statistical moments of autoregressive model residuals for damage localisation. Mechanical Systems and Signal Processing\/ 20\/ (3), 627--645

    2006

  30. [38]

    Meles, G. A., S. Marelli, and N. Linde (2025). Bayesian full waveform inversion with sequential surrogate model refinement . Geophysical Journal International\/ . Submitted

    2025

  31. [39]

    Murphy, K. P. (2012). Machine Learning: A Probabilistic Perspective . Adaptive Computation and Machine Learning Series. Cambridge, Massachusetts, USA: The MIT Press

    2012

  32. [40]

    Murray-Smith, R., T. A. Johansen, and R. Shorten (1999). On transient dynamics, off-equilibrium behaviour and identification in blended multiple model structures. In 1999 European Control Conference , pp.\ 3569--3574. IEEE

    1999

  33. [41]

    Rieckermann, and B

    Nagel, J., J. Rieckermann, and B. Sudret (2020). Principal component analysis and sparse polynomial chaos expansions for global sensitivity analysis and model calibration: application to urban drainage simulation. Reliability Engineering & System Safety\/ 195\/ (106737)

    2020

  34. [42]

    Grujovi\'c, D

    Rankovi\'c, V., N. Grujovi\'c, D. Divac, and N. Milivojevi\'c (2014). Development of support vector regression identification model for prediction of dam structural behaviour. Structural Safety\/ 48 , 33--39

    2014

  35. [43]

    Rasmussen, C. E. and C. K. I. Williams (2006). Gaussian processes for machine learning\/ ( Internet ed.). Adaptive computation and machine learning. Cambridge, Massachusetts: MIT Press

    2006

  36. [44]

    Rezaeian, S. and A. Der Kiureghian (2010). Simulation of synthetic ground motions for specified earthquake and site characteristics . Earthquake Engineering & Structural Dynamics\/ 39\/ (10), 1155--1180

    2010

  37. [45]

    Samsuri, N. A., S. A. Raman, and T. M. Y. S. Tuan Ya (2023). Evaluation of NARX network performance on the maintenance application of rotating machines. In F. Ahmad, H. H. Al-Kayiem, and W. P. King Soon (Eds.), ICPER 2020 , pp.\ 593--609. Springer Nature Singapore

    2023

  38. [46]

    Santner, T. J., B. J. Williams, and W. I. Notz (2003). The D esign and A nalysis of C omputer E xperiments . Springer, New York

    2003

  39. [47]

    Sapsis, T. and P. Lermusiaux (2009). Dynamically orthogonal field equations for continuous stochastic dynamical systems. Physica D: Nonlinear Phenomena\/ 238\/ (23), 2347--2360

    2009

  40. [48]

    Sch\"ar, S. (2025). Autoregressive surrogate modeling for dynamical systems . Ph.\ D. thesis, ETH Zurich (Switzerland)

    2025

  41. [49]

    Marelli, and B

    Sch\"ar, S., S. Marelli, and B. Sudret (2024). Emulating the dynamics of complex systems using autoregressive models on manifolds (mNARX) . Mechanical Systems and Signal Processing\/ 208 , 110956

    2024

  42. [50]

    Marelli, and B

    Sch \"a r, S., S. Marelli, and B. Sudret (2025). Surrogate modeling with functional nonlinear autoregressive models (F-NARX) . Reliability Engineering & System Safety\/ , 111276

    2025

  43. [51]

    Horne, and C

    Siegelmann, H., B. Horne, and C. Giles (1997). Computational capabilities of recurrent NARX neural networks. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics)\/ 27\/ (2), 208--215

    1997

  44. [52]

    Smola, A. J. and B. Sch\"olkopf (2004). A tutorial on support vector regression. Statistics and Computing\/ 14 , 199--222

    2004

  45. [53]

    Song, H., X. Shan, L. Zhang, G. Wang, and J. Fan (2022). Research on identification and active vibration control of cantilever structure based on NARX neural network. Mechanical Systems and Signal Processing\/ 171 , 108872

    2022

  46. [54]

    Spiridonakos, M. and E. Chatzi (2015a). Metamodeling of dynamic nonlinear structural systems through polynomial chaos NARX models. Computers & Structures\/ 157 , 99--113

    2015

  47. [55]

    Spiridonakos, M. D. and E. N. Chatzi (2015b). Metamodeling of nonlinear structural systems with parametric uncertainty subject to stochastic dynamic excitation. Earthquakes and Structures\/ 8\/ (4), 915--934

    2015

  48. [56]

    Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society. Series B (Methodological)\/ 58\/ (1), 267--288

    1996

  49. [57]

    Fahrni, M

    Wagner, P.-R., R. Fahrni, M. Klippel, A. Frangi, and B. Sudret (2020). Bayesian calibration and sensitivity analysis of heat transfer models for fire insulation panels. Engineering Structures\/ 205\/ (110063)

    2020

  50. [58]

    Wan, X. and G. E. Karniadakis (2005). An adaptive multi-element generalized polynomial chaos method for stochastic differential equations . Journal of Computational Physics\/ 209\/ (2), 617--642

    2005

  51. [59]

    Becker, T

    Worden, K., W. Becker, T. Rogers, and E. Cross (2018). On the confidence bounds of Gaussian process NARX models and their higher-order frequency response functions. Mechanical Systems and Signal Processing\/ 104 , 188--223

    2018

  52. [60]

    Xiu, D. (2010). Numerical Methods for Stochastic Computations: A Spectral Method Approach . Princeton, New Jersey, USA: Princeton University Press

    2010

  53. [61]

    Xiu, D. and G. E. Karniadakis (2002, January). The Wiener-Askey polynomial chaos for stochastic differential equations . SIAM Journal on Scientific Computing\/ 24\/ (2), 619--644

    2002

  54. [62]

    Marelli, B

    Yaghoubi, V., S. Marelli, B. Sudret, and T. Abrahamsson (2017). Sparse polynomial chaos expansions of frequency response functions using stochastic frequency transformation. Probabilistic Engineering Mechanics\/ 48 , 39--58

    2017

  55. [63]

    Yu, C., Y.-P. Zhu, H. Luo, Z. Luo, and L. Lei (2023). Design assessments of complex systems based on design oriented modelling and uncertainty analysis. Mechanical Systems and Signal Processing\/ 188\/ (109988)

    2023

  56. [64]

    Yin, and R

    Zhang, J., Z. Yin, and R. Wang (2017). Nonlinear dynamic classification of momentary mental workload using physiological features and NARX -model-based least-squares support vector machines. IEEE Transactions on Human-Machine Systems\/ 47\/ (4), 536--549

    2017

  57. [65]

    Dong, and M

    Zhang, Y., Y. Dong, and M. Beer (2024). rLSTM-AE for dimension reduction and its application to active learning-based dynamic reliability analysis. Mechanical Systems and Signal Processing\/ 215 , 111426

    2024

  58. [66]

    Zhou, J. and J. Li (2023). An efficient time-variant reliability analysis strategy embedding the NARX neural network of response characteristics prediction into probability density evolution method. Mechanical Systems and Signal Processing\/ 200 , 110516

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.