pith. sign in

arxiv: 2605.21717 · v1 · pith:NUXJUCNNnew · submitted 2026-05-20 · 📊 stat.CO · cs.NA· math.NA

Likelihood-informed dimension reduction across tempered Bayesian posteriors

Arne Bouillon , Oliver R. A. Dunbar This is my paper

Pith reviewed 2026-05-22 07:36 UTC · model grok-4.3

classification 📊 stat.CO cs.NAmath.NA
keywords likelihood-informed subspacesdimension reductiontempered posteriorsalpha-annealingBayesian inversioninformation loss boundsemulationposterior sampling
0
0 comments X

The pith

Generalizing likelihood-informed subspaces to alpha-tempered posteriors shows that alpha below one often produces near-optimal dimension reduction even with scarce noisy data.

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

This paper takes the existing approach of likelihood-informed subspaces that truncate parameter space to directions most relevant to the data and extends it so the same bounds on information loss can be optimized for tempered versions of the posterior. Tempering is controlled by an annealing parameter alpha between zero and one that gradually brings in the likelihood. The authors prove that the resulting alpha-LIS spaces remain well-defined and demonstrate through theory and experiments that values of alpha less than one frequently give spaces that perform as well as or better than the full posterior in practice. They further show how to accumulate information across a sequence of different alpha values and how to use simple gradient approximations so the method works for simulators that are chaotic or produce only noisy derivatives.

Core claim

The paper establishes that the likelihood-informed subspace construction, which optimizes bounds on information loss to identify informative directions, can be provably generalized from the standard posterior to alpha-tempered posteriors for any alpha in [0,1]. The resulting alpha-LIS spaces are partially informed and the authors show both theoretically and empirically that alpha less than one often yields spaces that are near-optimal for the original problem. Extensions that pool data across a sequence of alpha values and replace exact gradients with simple approximations allow the method to remain effective when data are limited and noisy and when the forward map is chaotic or stochastic.

What carries the argument

The alpha-LIS: a subspace obtained by optimizing information-loss bounds on an alpha-tempered posterior, which identifies directions that remain informative under tempered likelihoods.

If this is right

  • Dimension reduction via optimized information-loss bounds becomes applicable to annealed and power-posterior distributions used in sequential Monte Carlo and tempering schedules.
  • Choosing alpha below one can produce subspaces that are more robust than the full posterior when data are scarce or contaminated by noise.
  • Pooling evaluations across an increasing sequence of alpha values improves the quality of the identified subspace without requiring additional simulator runs at the target posterior.
  • Simple finite-difference or regression-based gradient approximations replace exact derivatives, enabling the method for simulators where analytic gradients are unavailable or uninformative.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same tempering idea might allow adaptive selection of alpha during the dimension-reduction step itself rather than fixing it in advance.
  • Because the method works with accumulated data across alpha levels, it could be combined with existing multi-fidelity or sequential design strategies for emulator construction.
  • In very high-dimensional problems the improvement from alpha less than one may compound with other reduction techniques such as active subspaces or projection-based emulators.

Load-bearing premise

The information-loss bounds optimized on tempered posteriors stay informative and near-optimal when data are severely limited and noisy, and simple gradient approximations suffice even for chaotic or stochastic forward maps.

What would settle it

Run a high-dimensional Bayesian inversion with very few noisy observations and a stochastic simulator, then compare posterior sampling efficiency and reconstruction error using alpha-LIS at alpha less than one versus the standard LIS at alpha equal to one.

Figures

Figures reproduced from arXiv: 2605.21717 by Arne Bouillon, Oliver R. A. Dunbar.

Figure 1
Figure 1. Figure 1: Wasserstein-2 posterior errors for different algorithms on the linear test problem [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Wasserstein-2 posterior errors for the linear test problem in function of [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Wasserstein-2 posterior errors for the linear test problem in function of the reduced input and output dimen [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Hellinger posterior errors for different algorithms on the linear-exponential test problem [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Wasserstein-2 posterior errors for the linear test problem in function of the reduced input and output dimen [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Wasserstein-2 posterior errors for the linear test problem in function of the reduced input and output dimen [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Hellinger posterior errors for the linear-exponential test problem in function of the reduced input and output [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Hellinger posterior errors for the linear-exponential test problem in function of [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Hellinger posterior errors for the Darcy test problem in function of [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Hellinger posterior errors for the Darcy test problem in function of [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Distance from the sample mean to the true parameter for the Lorenz ‘96 problem, with either PCA (for [PITH_FULL_IMAGE:figures/full_fig_p021_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Wasserstein-2 posterior errors as a function of the reduced output dimension [PITH_FULL_IMAGE:figures/full_fig_p030_12.png] view at source ↗
read the original abstract

Scientific computer simulations cannot represent all scales in realistic applications. To bridge this model-data gap, parameters are injected into models and constrained with noisy data using Bayesian inversion. To reduce the number of simulator evaluations, which can be 10^5 or more, modern approaches employ dimension reduction in conjunction with emulation of the forward map (that contains the simulator). Due to scarcity of model evaluations and data, this dimension reduction becomes very important for posterior sampling performance. Recent work on likelihood-informed subspaces (LIS) truncates to informative directions by optimizing bounds on information loss, and though mathematically well-adapted to sampling, they are often restrictive in practice. In this work, we provably generalize this methodology to facilitate application to $\alpha$-tempered (i.e., annealed, power-posterior) distributions for $\alpha$ in [0,1]. We provide theory to build partially-informed spaces termed $\alpha$-LIS. We show how $\alpha$ < 1 can often produce near-optimal spaces. In addition, we focus on applying $\alpha$-LIS to practical cases, where the available data is severely limited and noisy. We propose and test extensions for utilizing data from the entire sequence of distributions $\alpha$_0 < ... < $\alpha$_k, and use simple approximations of model gradients so that our approach can be used for emulation of forward maps for chaotic or stochastic systems where derivatives are unavailable or uninformative due to noise. In experiments, our accumulated approach is much more robust to these challenging circumstances than the theoretically optimal $\alpha$ = 1.

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

2 major / 2 minor

Summary. The manuscript generalizes likelihood-informed subspaces (LIS) to α-tempered posteriors (α in [0,1]), introducing partially-informed α-LIS spaces with theory showing that α < 1 can often be near-optimal. It further proposes practical extensions that accumulate data across a sequence of tempered distributions and replace exact gradients with simple approximations to handle limited noisy data and chaotic/stochastic forward maps, claiming greater robustness than α = 1 in experiments.

Significance. If the generalization and practical extensions hold, the work would meaningfully improve dimension reduction for Bayesian inversion of expensive simulators, enabling more efficient posterior sampling under data scarcity. The explicit theoretical treatment of tempered posteriors and the focus on gradient-free settings are strengths that could influence emulation-based inference pipelines.

major comments (2)
  1. [Abstract and α-LIS derivation] Abstract and the section deriving α-LIS: the claim that the accumulated heuristic with simple gradient approximations produces near-optimal spaces rests on the assertion that information-loss bounds remain informative, yet no derivation is supplied showing that these bounds are preserved (or remain tight) once exact gradients are replaced by approximations on stochastic or chaotic maps. This step is load-bearing for the practical robustness claim.
  2. [Practical extensions and experiments] The section on practical extensions and experiments: while the abstract states that the accumulated approach is 'much more robust' when data is severely limited and noisy, the manuscript provides no quantitative comparison of the achieved information-loss bounds or subspace optimality metrics between the approximated α-LIS and the exact α = 1 case, leaving the empirical support for the central practical assertion difficult to evaluate.
minor comments (2)
  1. [Notation] The notation α0 < … < αk for the tempering sequence is introduced without an accompanying table or explicit list of the specific α values and number of stages used in the numerical examples.
  2. [Figures] Figure captions and legends should explicitly indicate which curves or subspaces correspond to different α values and to the accumulated versus single-α constructions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments on the manuscript. We address each major comment point by point below, indicating where revisions will be made to clarify or strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract and α-LIS derivation] Abstract and the section deriving α-LIS: the claim that the accumulated heuristic with simple gradient approximations produces near-optimal spaces rests on the assertion that information-loss bounds remain informative, yet no derivation is supplied showing that these bounds are preserved (or remain tight) once exact gradients are replaced by approximations on stochastic or chaotic maps. This step is load-bearing for the practical robustness claim.

    Authors: We agree that the manuscript supplies no derivation establishing that the information-loss bounds remain informative or tight after replacing exact gradients with simple approximations on stochastic or chaotic maps. The theoretical development establishes the bounds and near-optimality results for α-LIS under exact gradients. The accumulated heuristic and gradient approximations are introduced as practical devices to enable use when derivatives are unavailable or uninformative; their justification in the current text rests on the numerical experiments rather than an extended theoretical guarantee. In the revision we will add an explicit statement in the practical-extensions section clarifying this distinction and qualifying the robustness claim accordingly, without asserting preservation of the bounds. revision: yes

  2. Referee: [Practical extensions and experiments] The section on practical extensions and experiments: while the abstract states that the accumulated approach is 'much more robust' when data is severely limited and noisy, the manuscript provides no quantitative comparison of the achieved information-loss bounds or subspace optimality metrics between the approximated α-LIS and the exact α = 1 case, leaving the empirical support for the central practical assertion difficult to evaluate.

    Authors: We acknowledge that the experiments section reports improved sampling efficiency and robustness for the accumulated approach under limited noisy data, but does not furnish direct quantitative comparisons of information-loss bounds or subspace optimality metrics between the approximated α-LIS and the exact α = 1 case. This omission makes the strength of the empirical support harder to assess. In the revised manuscript we will include additional quantitative comparisons—such as reported information-loss values or subspace-quality metrics where they can be computed—to provide clearer evidence for the robustness statement. revision: yes

Circularity Check

0 steps flagged

No circularity: generalization rests on external information-theoretic bounds

full rationale

The paper's central claim is a provable generalization of likelihood-informed subspace (LIS) bounds to α-tempered posteriors, presented as a mathematical extension of prior external theory on information loss. No derivation step reduces by the paper's own equations to a fitted parameter renamed as prediction, a self-citation chain, or an ansatz smuggled via citation. The theory is described as building on established information-theoretic results, with α<1 optimality shown via analysis rather than construction, and practical approximations tested separately in experiments without being asserted as theoretically guaranteed. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The approach rests on standard information-theoretic bounds for dimension reduction and the existence of useful low-dimensional structure in tempered posteriors; no new free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Likelihood-informed subspaces can be defined by optimizing bounds on information loss between full and reduced posterior measures.
    This is the foundational premise inherited from prior LIS work and extended to the tempered case.
invented entities (1)
  • α-LIS no independent evidence
    purpose: Partially-informed dimension reduction space for tempered posteriors
    New named construction that generalizes standard LIS; no independent evidence provided beyond the claimed theory.

pith-pipeline@v0.9.0 · 5815 in / 1215 out tokens · 34370 ms · 2026-05-22T07:36:46.008289+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

76 extracted references · 76 canonical work pages · 2 internal anchors

  1. [1]

    Ricardo Baptista, Youssef Marzouk, and Olivier Zahm.Gradient-based data and parameter dimension reduc- tion for Bayesian models: An information theoretic perspective. 2022. arXiv:2207.08670 [stat.CO]

    work page arXiv 2022

  2. [2]

    Manopt.jl: Optimization on Manifolds in Julia

    Ronny Bergmann. “Manopt.jl: Optimization on Manifolds in Julia”. In:Journal of Open Source Software7.70 (2022), p. 3866

    work page 2022

  3. [3]

    Adaptive Sampling with the Ensemble Trans- form Kalman Filter. Part I: Theoretical Aspects

    Craig H. Bishop, Brian J. Etherton, and Sharanya J. Majumdar. “Adaptive Sampling with the Ensemble Trans- form Kalman Filter. Part I: Theoretical Aspects”. In:Monthly Weather Review129.3 (2001), pp. 420–436

    work page 2001

  4. [4]

    Dimension reduction via score ratio matching

    Michael Brennan, Ricardo Baptista, and Youssef Marzouk. “Dimension reduction via score ratio matching”. In: NeurIPS 2022 Workshop on Score-Based Methods. 2022

    work page 2022

  5. [5]

    On the central limit theorem for dynamical systems

    Robert Burton and Manfred Denker. “On the central limit theorem for dynamical systems”. In:Transactions of the American Mathematical Society302.2 (1987), pp. 715–726

    work page 1987

  6. [6]

    Ensemble Kalman methods: A mean-field perspec- tive

    Edoardo Calvello, Sebastian Reich, and Andrew M Stuart. “Ensemble Kalman methods: A mean-field perspec- tive”. In:Acta Numerica34 (2025), pp. 123–291

    work page 2025

  7. [7]

    Chada, Yuming Chen, and Daniel Sanz-Alonso.Iterative Ensemble Kalman Methods: A Unified Per- spective with Some New Variants

    Neil K. Chada, Yuming Chen, and Daniel Sanz-Alonso.Iterative Ensemble Kalman Methods: A Unified Per- spective with Some New Variants. 2020. arXiv:2010.13299 [math.NA]

    work page arXiv 2020

  8. [8]

    Ensemble randomized maximum likelihood method as an iterative ensemble smoother

    Yan Chen and Dean S Oliver. “Ensemble randomized maximum likelihood method as an iterative ensemble smoother”. In:Mathematical Geosciences44.1 (2012), pp. 1–26. 30 Likelihood-informed dimension reduction across tempered Bayesian posteriorsA PREPRINT

    work page 2012

  9. [9]

    Nicolas Chopin, Omiros Papaspiliopoulos, et al.An introduction to sequential Monte Carlo. V ol. 4. Springer, 2020

    work page 2020

  10. [10]

    Markov chain Monte Carlo Using an Approximation

    J. Andr ´es Christen and Colin Fox. “Markov chain Monte Carlo Using an Approximation”. In:Journal of Com- putational and Graphical Statistics14.4 (2005), pp. 795–810

    work page 2005

  11. [11]

    Calibrate, emulate, sample

    Emmet Cleary et al. “Calibrate, emulate, sample”. In:Journal of Computational Physics424 (2021), p. 109716

    work page 2021

  12. [12]

    Paul Constantine and David Gleich.Computing active subspaces with Monte Carlo. 2015. arXiv:1408.0545 [math.NA]

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  13. [13]

    SIAM, 2015

    Paul G Constantine.Active subspaces: Emerging ideas for dimension reduction in parameter studies. SIAM, 2015

    work page 2015

  14. [14]

    Accelerating Markov Chain Monte Carlo with Active Subspaces

    Paul G. Constantine, Carson Kent, and Tan Bui-Thanh. “Accelerating Markov Chain Monte Carlo with Active Subspaces”. In:SIAM Journal on Scientific Computing38.5 (2016), A2779–A2805

    work page 2016

  15. [15]

    Efficient space-filling and non-collapsing sequential design strate- gies for simulation-based modeling

    K. Crombecq, E. Laermans, and T. Dhaene. “Efficient space-filling and non-collapsing sequential design strate- gies for simulation-based modeling”. In:European Journal of Operational Research214.3 (2011), pp. 683– 696

    work page 2011

  16. [16]

    Scalable posterior approximations for large-scale Bayesian inverse problems via likelihood-informed parameter and state reduction

    Tiangang Cui, Youssef Marzouk, and Karen Willcox. “Scalable posterior approximations for large-scale Bayesian inverse problems via likelihood-informed parameter and state reduction”. In:Journal of Computa- tional Physics315 (2016), pp. 363–387

    work page 2016

  17. [17]

    A unified performance analysis of likelihood-informed subspace methods

    Tiangang Cui and Xin T Tong. “A unified performance analysis of likelihood-informed subspace methods”. In: Bernoulli28.4 (2022), pp. 2788–2815

    work page 2022

  18. [18]

    Data-free likelihood-informed dimension reduction of Bayesian inverse prob- lems

    Tiangang Cui and Olivier Zahm. “Data-free likelihood-informed dimension reduction of Bayesian inverse prob- lems”. In:Inverse Problems37.4 (2021), p. 045009

    work page 2021

  19. [19]

    Likelihood-informed dimension reduction for nonlinear inverse problems

    Tiangang Cui et al. “Likelihood-informed dimension reduction for nonlinear inverse problems”. In:Inverse Problems30.11 (2014), p. 114015

    work page 2014

  20. [20]

    Sequential Monte Carlo samplers

    Pierre Del Moral, Arnaud Doucet, and Ajay Jasra. “Sequential Monte Carlo samplers”. In:Journal of the Royal Statistical Society Series B: Statistical Methodology68.3 (2006), pp. 411–436

    work page 2006

  21. [21]

    Simulated annealing: From basics to applica- tions

    Daniel Delahaye, Supatcha Chaimatanan, and Marcel Mongeau. “Simulated annealing: From basics to applica- tions”. In:Handbook of Metaheuristics. Springer, 2018, pp. 1–35

    work page 2018

  22. [22]

    A semiautomatic method for history matching using sequential Monte Carlo

    Christopher Drovandi, David J Nott, and Daniel E Pagendam. “A semiautomatic method for history matching using sequential Monte Carlo”. In:SIAM/ASA Journal on Uncertainty Quantification9.3 (2021), pp. 1034– 1063

    work page 2021

  23. [23]

    Hyperparameter optimization for randomized algorithms: a case study on random features

    Oliver R. A. Dunbar, Nicholas H. Nelsen, and Maya Mutic. “Hyperparameter optimization for randomized algorithms: a case study on random features”. In:Statistics and Computing35.3 (2025), p. 56

    work page 2025

  24. [24]

    CalibrateEmulateSample.jl: Accelerated parametric uncertainty quantification

    Oliver R. A. Dunbar et al. “CalibrateEmulateSample.jl: Accelerated parametric uncertainty quantification”. In: Journal of Open Source Software9.97 (2024), p. 6372

    work page 2024

  25. [25]

    Calibration and uncertainty quantification of convective parameters in an idealized GCM

    Oliver R. A. Dunbar et al. “Calibration and uncertainty quantification of convective parameters in an idealized GCM”. In:Journal of Advances in Modeling Earth Systems13.9 (2021), e2020MS002454

    work page 2021

  26. [26]

    Ensemble inference methods for models with noisy and expensive likelihoods

    Oliver R. A. Dunbar et al. “Ensemble inference methods for models with noisy and expensive likelihoods”. In: SIAM Journal on Applied Dynamical Systems21.2 (2022), pp. 1539–1572

    work page 2022

  27. [27]

    EnsembleKalmanProcesses.jl: Derivative-free ensemble-based model calibration

    Oliver R. A. Dunbar et al. “EnsembleKalmanProcesses.jl: Derivative-free ensemble-based model calibration”. In:Journal of Open Source Software7.80 (2022), p. 4869

    work page 2022

  28. [28]

    Parallel tempering: Theory, applications, and new perspectives

    David J Earl and Michael W Deem. “Parallel tempering: Theory, applications, and new perspectives”. In:Phys- ical Chemistry Chemical Physics7.23 (2005), pp. 3910–3916

    work page 2005

  29. [29]

    Fast algorithms for Bayesian uncertainty quantification in large-scale linear inverse problems based on low-rank partial Hessian approximations

    H. P. Flath et al. “Fast algorithms for Bayesian uncertainty quantification in large-scale linear inverse problems based on low-rank partial Hessian approximations”. In:SIAM Journal on Scientific Computing33.1 (2011), pp. 407–432

    work page 2011

  30. [30]

    Surrogate and reduced-order modeling: A comparison of approaches for large-scale sta- tistical inverse problems

    Michalis Frangos et al. “Surrogate and reduced-order modeling: A comparison of approaches for large-scale sta- tistical inverse problems”. In:Large-Scale Inverse Problems and Quantification of Uncertainty(2010), pp. 123– 149

    work page 2010

  31. [31]

    Marginal Likelihood Estimation via Power Posteriors

    N. Friel and A. N. Pettitt. “Marginal Likelihood Estimation via Power Posteriors”. In:Journal of the Royal Statistical Society Series B: Statistical Methodology70.3 (2008), pp. 589–607

    work page 2008

  32. [32]

    Introduction to Markov chain Monte Carlo

    Charles J Geyer. “Introduction to Markov chain Monte Carlo”. In:Handbook of Markov chain Monte Carlo 20116022.45 (2011), p. 22

    work page 2011

  33. [33]

    Rebecca Gjini et al.The Ensemble Kalman Inversion Race. 2025. arXiv:2511.15853 [physics.data-an]

    work page arXiv 2025

  34. [34]

    Accuracy improvement in ensemble Kalman inversion through data- informed ensemble selection

    Ruben Harris and Claudia Schillings. “Accuracy improvement in ensemble Kalman inversion through data- informed ensemble selection”. In:Inverse Problems41.10 (2025), p. 105010. 31 Likelihood-informed dimension reduction across tempered Bayesian posteriorsA PREPRINT

    work page 2025

  35. [35]

    Introduction to Gaussian Process Regression in Bayesian Inverse Problems, with New Results on Experimental Design for Weighted Error Measures

    Tapio Helin et al. “Introduction to Gaussian Process Regression in Bayesian Inverse Problems, with New Results on Experimental Design for Weighted Error Measures”. In:Monte Carlo and Quasi-Monte Carlo Methods. Ed. by Aicke Hinrichs, Peter Kritzer, and Friedrich Pillichshammer. Cham: Springer International Publishing, 2024, pp. 49–79

    work page 2024

  36. [36]

    Likelihood-free MCMC with amortized approximate ratio estimators

    Joeri Hermans, V olodimir Begy, and Gilles Louppe. “Likelihood-free MCMC with amortized approximate ratio estimators”. In:International conference on machine learning. PMLR. 2020, pp. 4239–4248

    work page 2020

  37. [37]

    A Bayesian approach to characterizing uncertainty in inverse problems using coarse and fine-scale information

    D. Higdon, H. Lee, and Zhuoxin Bi. “A Bayesian approach to characterizing uncertainty in inverse problems using coarse and fine-scale information”. In:IEEE Transactions on Signal Processing50.2 (2002), pp. 389– 399

    work page 2002

  38. [38]

    Combining field data and computer simulations for calibration and prediction

    Dave Higdon et al. “Combining field data and computer simulations for calibration and prediction”. In:SIAM Journal on Scientific Computing26.2 (2004), pp. 448–466

    work page 2004

  39. [39]

    Complexity analysis of accelerated MCMC methods for Bayesian inversion

    Viet Ha Hoang, Christoph Schwab, and Andrew M Stuart. “Complexity analysis of accelerated MCMC methods for Bayesian inversion”. In:Inverse Problems29.8 (2013), p. 085010

    work page 2013

  40. [40]

    Holthuijzen et al.Surrogate-based Bayesian calibration methods for chaotic systems: A comparison of traditional and non-traditional approaches

    Maike F. Holthuijzen et al.Surrogate-based Bayesian calibration methods for chaotic systems: A comparison of traditional and non-traditional approaches. 2026. arXiv:2508.13071 [stat.ME]

    work page arXiv 2026

  41. [41]

    Horn and Charles R

    Roger A. Horn and Charles R. Johnson.Matrix Analysis. 2nd ed. Cambridge University Press, 2012

    work page 2012

  42. [42]

    Parameter Uncertainty Quantification in an Idealized GCM With a Seasonal Cycle

    Michael F. Howland, Oliver R. A. Dunbar, and Tapio Schneider. “Parameter Uncertainty Quantification in an Idealized GCM With a Seasonal Cycle”. In:Journal of Advances in Modeling Earth Systems14.3 (2022), e2021MS002735

    work page 2022

  43. [43]

    Iterated Kalman Methodology for Inverse Problems

    Daniel Zhengyu Huang, Tapio Schneider, and Andrew M. Stuart. “Iterated Kalman Methodology for Inverse Problems”. In:Journal of Computational Physics463 (2022), p. 111262

    work page 2022

  44. [44]

    Efficient derivative-free Bayesian inference for large-scale inverse problems

    Daniel Zhengyu Huang et al. “Efficient derivative-free Bayesian inference for large-scale inverse problems”. In: Inverse Problems38.12 (2022), p. 125006

    work page 2022

  45. [45]

    Bayesian inversion in resin transfer molding

    Marco Iglesias, Minho Park, and Michael V Tretyakov. “Bayesian inversion in resin transfer molding”. In: Inverse Problems34.10 (2018), p. 105002

    work page 2018

  46. [46]

    Adaptive regularisation for ensemble Kalman inversion

    Marco Iglesias and Yuchen Yang. “Adaptive regularisation for ensemble Kalman inversion”. In:Inverse Prob- lems37.2 (2021), p. 025008

    work page 2021

  47. [47]

    Ensemble Kalman methods for inverse problems

    Marco A Iglesias, Kody JH Law, and Andrew M Stuart. “Ensemble Kalman methods for inverse problems”. In: Inverse Problems29.4 (2013), p. 045001

    work page 2013

  48. [48]

    Victor Janssens, Karl Meerbergen, and Wim Michiels.Linearizing a nonlinear eigenvalue problem with quadratic rational eigenvector nonlinearities. 2025. arXiv:2510.02900 [math.NA]

    work page arXiv 2025

  49. [49]

    An Inverse Iteration Method for Eigenvalue Problems with Eigenvector Nonlinearities

    Elias Jarlebring, Simen Kvaal, and Wim Michiels. “An Inverse Iteration Method for Eigenvalue Problems with Eigenvector Nonlinearities”. In:SIAM Journal on Scientific Computing36.4 (2014), A1978–A2001

    work page 2014

  50. [50]

    Fast Bayesian approach for parameter estimation

    Bangti Jin. “Fast Bayesian approach for parameter estimation”. In:International Journal for Numerical Meth- ods in Engineering76.2 (2008), pp. 230–252

    work page 2008

  51. [51]

    Bayesian history matching applied to the calibra- tion of a gravity wave parameterization

    Robert C King, Laura A Mansfield, and Aditi Sheshadri. “Bayesian history matching applied to the calibra- tion of a gravity wave parameterization”. In:Journal of Advances in Modeling Earth Systems16.4 (2024), e2023MS004163

    work page 2024

  52. [52]

    Ensemble Kalman inversion: A derivative-free technique for ma- chine learning tasks

    Nikola B Kovachki and Andrew M Stuart. “Ensemble Kalman inversion: A derivative-free technique for ma- chine learning tasks”. In:Inverse Problems35.9 (2019), p. 095005

    work page 2019

  53. [53]

    Computing Bayes factors using thermodynamic integration

    Nicolas Lartillot and Herv ´e Philippe. “Computing Bayes factors using thermodynamic integration”. In:System- atic Biology55.2 (2006), pp. 195–207

    work page 2006

  54. [54]

    Sampling strategies for computer experiments: Design and analysis

    DK Lin, Timothy W Simpson, and Wei Chen. “Sampling strategies for computer experiments: Design and analysis”. In:International Journal of Reliability and Applications2.3 (2001), pp. 209–240

    work page 2001

  55. [55]

    Predictability: A problem partly solved

    Edward N Lorenz. “Predictability: A problem partly solved”. In:Proc. Seminar on predictability. V ol. 1. 1. Reading. 1996, pp. 1–18

    work page 1996

  56. [56]

    Simulated tempering: A new Monte Carlo scheme

    Enzo Marinari and Giorgio Parisi. “Simulated tempering: A new Monte Carlo scheme”. In:EPL (Europhysics Letters)19.6 (1992), pp. 451–458

    work page 1992

  57. [57]

    A stochastic Newton MCMC method for large-scale statistical inverse problems with application to seismic inversion

    James Martin et al. “A stochastic Newton MCMC method for large-scale statistical inverse problems with application to seismic inversion”. In:SIAM Journal on Scientific Computing34.3 (2012), A1460–A1487

    work page 2012

  58. [58]

    A Stochastic Collocation Approach to Bayesian Inference in Inverse Problems

    Youssef Marzouk and Dongbin Xiu. “A Stochastic Collocation Approach to Bayesian Inference in Inverse Problems”. In:Communications in Computational Physics6.4 (2009), pp. 826–847

    work page 2009

  59. [59]

    Principal component analysis in linear systems: Controllability, observability, and model reduction

    B. Moore. “Principal component analysis in linear systems: Controllability, observability, and model reduction”. In:IEEE Transactions on Automatic Control26.1 (1981), pp. 17–32. 32 Likelihood-informed dimension reduction across tempered Bayesian posteriorsA PREPRINT

    work page 1981

  60. [60]

    Synthesis of minimum roundoff noise fixed point digital filters

    C. Mullis and R. Roberts. “Synthesis of minimum roundoff noise fixed point digital filters”. In:IEEE Transac- tions on Circuits and Systems23.9 (1976), pp. 551–562

    work page 1976

  61. [61]

    Annealed importance sampling

    Radford M Neal. “Annealed importance sampling”. In:Statistics and Computing11.2 (2001), pp. 125–139

    work page 2001

  62. [62]

    Sampling from multimodal distributions using tempered transitions

    Radford M Neal. “Sampling from multimodal distributions using tempered transitions”. In:Statistics and Com- puting6.4 (1996), pp. 353–366

    work page 1996

  63. [63]

    DIAS: A data-informed active subspace regularization framework for inverse problems

    Hai Nguyen, Jonathan Wittmer, and Tan Bui-Thanh. “DIAS: A data-informed active subspace regularization framework for inverse problems”. In:Computation10.3 (2022), p. 38

    work page 2022

  64. [64]

    Nad `ege Polette et al.Covariance-Informed Subspace: An Adaptive Gradient-Free Input Dimension Reduction Method for Bayesian Inference. 2026. arXiv:2604.13563 [math.NA]

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  65. [65]

    Model reduction of linear dynamical systems via balancing for Bayesian inference

    Elizabeth Qian et al. “Model reduction of linear dynamical systems via balancing for Bayesian inference”. In: Journal of Scientific Computing91.1 (2022), p. 29

    work page 2022

  66. [66]

    Random features for large-scale kernel machines

    Ali Rahimi and Benjamin Recht. “Random features for large-scale kernel machines”. In:Advances in neural information processing systems20 (2007)

    work page 2007

  67. [67]

    Analysis of the ensemble Kalman filter for inverse problems

    Claudia Schillings and Andrew M Stuart. “Analysis of the ensemble Kalman filter for inverse problems”. In: SIAM Journal on Numerical Analysis55.3 (2017), pp. 1264–1290

    work page 2017

  68. [68]

    Ensemble-Based Gradient Inference for Particle Methods in Optimization and Sampling

    Claudia Schillings, Claudia Totzeck, and Philipp Wacker. “Ensemble-Based Gradient Inference for Particle Methods in Optimization and Sampling”. In:SIAM/ASA Journal on Uncertainty Quantification11.3 (2023), pp. 757–787

    work page 2023

  69. [69]

    Sequential design of computer experiments for the solution of Bayesian inverse problems

    Michael Sinsbeck and Wolfgang Nowak. “Sequential design of computer experiments for the solution of Bayesian inverse problems”. In:SIAM/ASA Journal on Uncertainty Quantification5.1 (2017), pp. 640–664

    work page 2017

  70. [70]

    Posterior consistency for Gaussian process approximations of Bayesian posterior distributions

    Andrew Stuart and Aretha Teckentrup. “Posterior consistency for Gaussian process approximations of Bayesian posterior distributions”. In:Mathematics of Computation87.310 (2018), pp. 721–753

    work page 2018

  71. [71]

    Sequential Bayesian experimental design for calibration of expensive simulation models

    ¨Ozge S¨urer, Matthew Plumlee, and Stefan M Wild. “Sequential Bayesian experimental design for calibration of expensive simulation models”. In:Technometrics66.2 (2024), pp. 157–171

    work page 2024

  72. [72]

    On the Iterated Forms of Kalman Filters Using Statistical Linearization

    Sridhar Ungarala. “On the Iterated Forms of Kalman Filters Using Statistical Linearization”. In:Journal of Process Control22.5 (2012), pp. 935–943

    work page 2012

  73. [73]

    Perspectives on locally weighted ensemble Kalman methods

    Philipp Wacker. “Perspectives on locally weighted ensemble Kalman methods”. In:Journal of Computational and Applied Mathematics472 (2026), p. 116756

    work page 2026

  74. [74]

    Christopher KI Williams and Carl Edward Rasmussen.Gaussian processes for machine learning. V ol. 2. 3. MIT press Cambridge, MA, 2006

    work page 2006

  75. [75]

    Statistical properties of dynamical systems with some hyperbolicity

    Lai-Sang Young. “Statistical properties of dynamical systems with some hyperbolicity”. In:Annals of Mathe- matics147.3 (1998), pp. 585–650

    work page 1998

  76. [76]

    Certified dimension reduction in nonlinear Bayesian inverse problems

    Olivier Zahm et al. “Certified dimension reduction in nonlinear Bayesian inverse problems”. In:Mathematics of Computation91.336 (2022), pp. 1789–1835. 33

    work page 2022