pith. machine review for the scientific record. sign in

arxiv: 2605.10249 · v1 · submitted 2026-05-11 · 📊 stat.ME

Recognition: no theorem link

Diffeomorphic registration distances for Bayesian calibration of infinite-dimensional computer models

Gwena\"el Salin, Paul Lartaud

Pith reviewed 2026-05-12 05:20 UTC · model grok-4.3

classification 📊 stat.ME
keywords Bayesian calibrationLDDMMdiffeomorphic registrationinfinite-dimensional modelscomputer modelsuncertainty quantificationRKHSpredictive posterior
0
0 comments X

The pith

Distances from the large deformation diffeomorphic metric matching framework can be used to perform Bayesian calibration of infinite-dimensional computer models.

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

The paper tries to establish that distances from the large deformation diffeomorphic metric matching framework provide a suitable metric for Bayesian calibration when computer model outputs are infinite-dimensional, such as scalar fields or function graphs. This metric is the minimal energy deformation needed to transform one shape into another, giving an interpretable comparison between model predictions and data. The diffeomorphism group is represented as an exponential transformation of a reproducing kernel Hilbert space, which fits directly into Bayesian inference. This setup allows a predictive posterior distribution to be defined over the infinite-dimensional output shapes. A sympathetic reader would care because standard distances often fail for functional outputs, and this approach extends reliable uncertainty quantification on calibrated parameters to more complex physical simulations.

Core claim

Bayesian calibration is performed using distances from the large deformation diffeomorphic metric matching (LDDMM) framework. LDDMM distances can provide a suitable metric for infinite-dimensional shapes such as scalar fields or function graphs. This metric can be interpreted as the minimal energy deformation required to transform one shape into another. As such, it provides a readily interpretable metric for Bayesian calibration. The representation of the diffeomorphism group as an exponential transformation of an RKHS is compatible with Bayesian inference and allows to define a predictive posterior distribution on the infinite-dimensional space shape.

What carries the argument

LDDMM distances, which quantify the minimal energy deformation between infinite-dimensional shapes by representing the diffeomorphism group as the exponential of a reproducing kernel Hilbert space.

If this is right

  • Bayesian calibration yields uncertainty quantification on physical and numerical parameters even when outputs are infinite-dimensional.
  • The metric supplies an interpretable deformation-energy basis for comparing model outputs to experimental data.
  • A predictive posterior distribution can be defined directly on the infinite-dimensional shape space.
  • The framework extends Bayesian methods to complex computer codes whose outputs are scalar fields or function graphs.

Where Pith is reading between the lines

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

  • The approach could be applied to spatial field outputs in physics simulations such as fluid flow or heat transfer to assess whether deformation-based calibration improves parameter recovery.
  • Comparison against Euclidean or other functional distances on benchmark datasets would clarify when the LDDMM metric adds value.
  • Kernel choice in the RKHS representation may require sensitivity checks to ensure the posterior remains stable across reasonable selections.

Load-bearing premise

That LDDMM distances provide a suitable, computationally feasible metric for the infinite-dimensional outputs without introducing intractable computations or unaccounted biases in the Bayesian procedure.

What would settle it

A controlled simulation with known true parameter values in which the posterior obtained via LDDMM calibration fails to concentrate around those values or the predictive posterior on shapes fails to cover the observed data.

Figures

Figures reproduced from arXiv: 2605.10249 by Gwena\"el Salin, Paul Lartaud.

Figure 1
Figure 1. Figure 1: Reference measurement (leftmost) and example of prior images. [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mean (center) and standard deviation (right) of the prior predictive distribution in image space, compared to [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Geodesic trajectory between a source image (upper left) and target image (lower right). [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Marginal posterior distributions of p(β|q (mes)). The red vertical line indicates the true value of β∗ used for the reference measurement [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Mean (center) and standard deviation (right) of the posterior predictive distribution in image space, compared [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Maximum a posteriori obtained during MCMC sampling of the posterior [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Experimental measurement and simulation curves [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Predictive posterior distribution and maximum a posteriori for the functional calibration of deformation speed [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Example of diffeomorphic registration for current-represented functional outputs. [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Averaged empirical 1D coverage as a function of nominal coverage for the GP-PCA surrogate model. [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Predictive posterior distribution and maximum a posteriori for the functional calibration of deformation [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Predictive posterior distribution and maximum a posteriori for the functional calibration of deformation [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Predictive posterior distribution and maximum a posteriori for the functional calibration of deformation [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison of all posterior densities p(β|q (mes)) for the different methods. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗
read the original abstract

The simulation of physical phenomena with computer models relies on the estimation of physical and/or numerical parameters calibrated to fit experimental data. The approximations within the computer model and the errors in the measurements lead to uncertainties in the calibrated parameters. Bayesian calibration offers a well-studied framework to provide reliable uncertainty quantification on the calibrated parameters. When dealing with complex computer codes whose outputs are infinite-dimensional, Bayesian calibration may be extended by providing a relevant distance in the output space. In this paper, Bayesian calibration is performed using distances from the large deformation diffeomorphic metric matching (LDDMM) framework. LDDMM distances can provide a suitable metric for infinite-dimensional shapes such as scalar fields (i.e. images) or function graphs. This metric can be interpreted as the minimal energy deformation required to transform one shape into another. As such, it provides a readily interpretable metric for Bayesian calibration. On top of this, the representation of the diffeomorphism group as an exponential transformation of an RKHS is compatible with Bayesian inference and allows to define a predictive posterior distribution on the infinite-dimensional space shape.

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

0 major / 3 minor

Summary. The manuscript proposes extending Bayesian calibration to computer models with infinite-dimensional outputs (e.g., scalar fields or function graphs) by replacing standard discrepancy measures with distances from the large deformation diffeomorphic metric matching (LDDMM) framework. Diffeomorphisms are represented as exponentials of vector fields in a reproducing kernel Hilbert space (RKHS), yielding an interpretable minimal-energy deformation metric that is embedded in a Gaussian-process-style likelihood; the resulting posterior on calibration parameters then induces a predictive posterior on the shape space via push-forward under the deterministic model.

Significance. If the technical construction holds, the work supplies a geometrically natural and energy-based metric for shape-valued outputs that is compatible with standard Bayesian sampling schemes, potentially improving uncertainty quantification in applications such as image registration or functional data calibration. The explicit provision of velocity-field optimization, kernel choices, and posterior sampling steps constitutes a concrete, reproducible contribution.

minor comments (3)
  1. The abstract states that the RKHS representation 'allows to define a predictive posterior distribution on the infinite-dimensional space shape,' but the precise measure-theoretic construction (e.g., how the push-forward is rigorously defined on the shape manifold) is only sketched; a short dedicated paragraph or appendix lemma would clarify this for readers unfamiliar with infinite-dimensional Bayesian nonparametrics.
  2. Notation for the velocity fields, kernel, and energy functional is introduced progressively; a single consolidated table or notation box in §2 would improve readability.
  3. The manuscript supplies the necessary technical steps for implementation, yet no explicit statement appears on the computational scaling of the velocity-field optimization with respect to output dimension; a brief complexity remark or reference to existing LDDMM solvers would strengthen the feasibility claim.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive and accurate summary of our manuscript, the assessment of its significance, and the recommendation for minor revision. No specific major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The derivation embeds an established LDDMM registration distance (from independent prior literature) as the metric inside a standard Bayesian likelihood for calibration parameters, then constructs the posterior predictive via push-forward under the deterministic model. No equation or step reduces a claimed result to a fitted quantity defined by the same data, nor does any load-bearing premise rest on a self-citation chain or ansatz smuggled from the authors' own prior work. The RKHS-exponential representation of diffeomorphisms is invoked only to make the distance and posterior well-defined, without self-definitional loops or renaming of known empirical patterns as novel derivations.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The proposal rests on standard mathematical properties of RKHS and diffeomorphism groups from the LDDMM literature; no new free parameters, ad-hoc axioms, or invented entities are introduced in the abstract.

axioms (1)
  • standard math Properties of reproducing kernel Hilbert spaces and exponential maps on diffeomorphism groups
    Invoked to ensure compatibility with Bayesian inference and to define the predictive posterior.

pith-pipeline@v0.9.0 · 5486 in / 1136 out tokens · 50076 ms · 2026-05-12T05:20:47.677564+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

40 extracted references · 40 canonical work pages · 1 internal anchor

  1. [1]

    Shape deformation analysis from the optimal control viewpoint

    Sylvain Arguillere et al. “Shape deformation analysis from the optimal control viewpoint”. In:Journal de mathématiques pures et appliquées104.1 (2015), pp. 139–178

    work page 2015

  2. [2]

    A supermartingale approach to Gaussian process based sequential design of experiments

    Julien Bect, François Bachoc, and David Ginsbourger. “A supermartingale approach to Gaussian process based sequential design of experiments”. In:Bernoulli25.4A (2019), pp. 2883–2919

    work page 2019

  3. [3]

    Computing large deformation metric mappings via geodesic flows of diffeomorphisms

    M Faisal Beg et al. “Computing large deformation metric mappings via geodesic flows of diffeomorphisms”. In: International journal of computer vision61.2 (2005), pp. 139–157

    work page 2005

  4. [4]

    Pyro: Deep Universal Probabilistic Programming

    Eli Bingham et al. “Pyro: Deep Universal Probabilistic Programming”. In:Journal of Machine Learning Research (2018)

    work page 2018

  5. [5]

    Multi-contrast large deformation diffeomorphic metric mapping for diffusion tensor imaging

    Can Ceritoglu et al. “Multi-contrast large deformation diffeomorphic metric mapping for diffusion tensor imaging”. In:Neuroimage47.2 (2009), pp. 618–627

    work page 2009

  6. [6]

    The fshape framework for the variability analysis of functional shapes

    Benjamin Charlier, Nicolas Charon, and Alain Trouvé. “The fshape framework for the variability analysis of functional shapes”. In:Foundations of Computational Mathematics17.2 (2017), pp. 287–357

    work page 2017

  7. [7]

    Functional currents: a new mathematical tool to model and analyse functional shapes

    Nicolas Charon and Alain Trouvé. “Functional currents: a new mathematical tool to model and analyse functional shapes”. In:Journal of mathematical imaging and vision48.3 (2014), pp. 413–431

    work page 2014

  8. [8]

    The varifold representation of nonoriented shapes for diffeomorphic registra- tion

    Nicolas Charon and Alain Trouvé. “The varifold representation of nonoriented shapes for diffeomorphic registra- tion”. In:SIAM journal on Imaging Sciences6.4 (2013), pp. 2547–2580

    work page 2013

  9. [9]

    Bayesian Registration of Functions and Curves

    Wen Cheng, Ian Dryden, and Xianzheng Huang. “Bayesian Registration of Functions and Curves”. In:Bayesian Analysis11 (June 2016).DOI:10.1214/15-BA957

    work page doi:10.1214/15-ba957 2016

  10. [10]

    MMD-Bayes: Robust Bayesian estimation via maximum mean discrepancy

    Badr-Eddine Chérief-Abdellatif and Pierre Alquier. “MMD-Bayes: Robust Bayesian estimation via maximum mean discrepancy”. In:Symposium on Advances in Approximate Bayesian Inference. PMLR. 2020, pp. 1–21

    work page 2020

  11. [11]

    On discontinuity of planar optimal transport maps

    Otis Chodosh et al. “On discontinuity of planar optimal transport maps”. In:Journal of Topology and Analysis 7.02 (2015), pp. 239–260

    work page 2015

  12. [12]

    John Wiley & Sons, 2005

    Peter Congdon.Bayesian models for categorical data. John Wiley & Sons, 2005

    work page 2005

  13. [13]

    Bayesian emulation of complex multi-output and dynamic computer models

    Stefano Conti and Anthony O’Hagan. “Bayesian emulation of complex multi-output and dynamic computer models”. In:Journal of statistical planning and inference140.3 (2010), pp. 640–651

    work page 2010

  14. [14]

    Hybrid monte carlo

    Simon Duane et al. “Hybrid monte carlo”. In:Physics letters B195.2 (1987), pp. 216–222

    work page 1987

  15. [15]

    Optimal data-driven sparse parameterization of diffeomorphisms for population analysis

    Sandy Durrleman et al. “Optimal data-driven sparse parameterization of diffeomorphisms for population analysis”. In:Biennial International Conference on Information Processing in Medical Imaging. Springer. 2011, pp. 123– 134

    work page 2011

  16. [16]

    Metamorphic image registration using a semi-Lagrangian scheme

    Anton François, Pietro Gori, and Joan Glaunès. “Metamorphic image registration using a semi-Lagrangian scheme”. In:International Conference on Geometric Science of Information. Springer. 2021, pp. 781–788

    work page 2021

  17. [17]

    Elastic Bayesian model calibration

    Devin Francom et al. “Elastic Bayesian model calibration”. In:SIAM/ASA Journal on Uncertainty Quantification 13.1 (2025), pp. 195–227

    work page 2025

  18. [18]

    Riemann manifold langevin and hamiltonian monte carlo methods

    Mark Girolami and Ben Calderhead. “Riemann manifold langevin and hamiltonian monte carlo methods”. In: Journal of the Royal Statistical Society Series B: Statistical Methodology73.2 (2011), pp. 123–214

    work page 2011

  19. [19]

    Computer model calibration using high-dimensional output

    Dave Higdon et al. “Computer model calibration using high-dimensional output”. In:Journal of the American Statistical Association103.482 (2008), pp. 570–583

    work page 2008

  20. [20]

    The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo

    Matthew D Hoffman, Andrew Gelman, et al. “The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo.” In:J. Mach. Learn. Res.15.1 (2014), pp. 1593–1623

    work page 2014

  21. [21]

    The Euler-Poincaré theory of metamorphosis

    Darryl Holm, Alain Trouvé, and Laurent Younes. “The Euler-Poincaré theory of metamorphosis”. In:Quarterly of Applied Mathematics67.4 (2009), pp. 661–685

    work page 2009

  22. [22]

    Bayesian calibration of computer models

    Marc C Kennedy and Anthony O’Hagan. “Bayesian calibration of computer models”. In:Journal of the Royal Statistical Society: Series B (Statistical Methodology)63.3 (2001), pp. 425–464

    work page 2001

  23. [23]

    Adam: A Method for Stochastic Optimization

    Diederik P Kingma and Jimmy Ba. “Adam: A method for stochastic optimization”. In:arXiv preprint arXiv:1412.6980(2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  24. [24]

    A Lagrangian Gauss–Newton–Krylov solver for mass-and intensity- preserving diffeomorphic image registration

    Andreas Mang and Lars Ruthotto. “A Lagrangian Gauss–Newton–Krylov solver for mass-and intensity- preserving diffeomorphic image registration”. In:SIAM Journal on Scientific Computing39.5 (2017), B860– B885

    work page 2017

  25. [25]

    Probabilistic surrogate modeling by Gaussian process: A new estimation algorithm for more robust prediction

    Amandine Marrel and Bertrand Iooss. “Probabilistic surrogate modeling by Gaussian process: A new estimation algorithm for more robust prediction”. In:Reliability Engineering & System Safety247 (2024), p. 110120

    work page 2024

  26. [26]

    Geodesic shooting for computational anatomy

    Michael I Miller, Alain Trouvé, and Laurent Younes. “Geodesic shooting for computational anatomy”. In: Journal of mathematical imaging and vision24.2 (2006), pp. 209–228. 22

    work page 2006

  27. [27]

    On the metrics and Euler-Lagrange equations of computa- tional anatomy

    Michael I Miller, Alain Trouvé, and Laurent Younes. “On the metrics and Euler-Lagrange equations of computa- tional anatomy”. In:Annual review of biomedical engineering4.1 (2002), pp. 375–405

    work page 2002

  28. [28]

    Bayesian learning via stochastic dynamics

    Radford Neal. “Bayesian learning via stochastic dynamics”. In:Advances in neural information processing systems5 (1992)

    work page 1992

  29. [29]

    Calibration of Johnson’s damage model by a Bayesian approach

    Guillaume Perrin and Laurianne Pillon. “Calibration of Johnson’s damage model by a Bayesian approach”. In: EPJ Web of Conferences. V ol. 183. EDP Sciences. 2018, p. 01037

    work page 2018

  30. [30]

    Change of measure for Bayesian field inversion with hierarchical hyperparameters sampling

    Nadège Polette et al. “Change of measure for Bayesian field inversion with hierarchical hyperparameters sampling”. In:Journal of Computational Physics529 (2025), p. 113888

    work page 2025

  31. [31]

    Thomas J Santner et al.The design and analysis of computer experiments. V ol. 1. Springer, 2003

    work page 2003

  32. [32]

    Calibration of computational models with categorical parameters and correlated outputs via Bayesian smoothing spline ANOV A

    Curtis B Storlie et al. “Calibration of computational models with categorical parameters and correlated outputs via Bayesian smoothing spline ANOV A”. In:Journal of the American Statistical Association110.509 (2015), pp. 68–82

    work page 2015

  33. [33]

    Inverse problems: a Bayesian perspective

    Andrew M Stuart. “Inverse problems: a Bayesian perspective”. In:Acta numerica19 (2010), pp. 451–559

    work page 2010

  34. [34]

    Diffeomorphisms groups and pattern matching in image analysis

    Alain Trouvé. “Diffeomorphisms groups and pattern matching in image analysis”. In:International journal of computer vision28.3 (1998), pp. 213–221

    work page 1998

  35. [35]

    Local geometry of deformable templates

    Alain Trouvé and Laurent Younes. “Local geometry of deformable templates”. In:SIAM journal on mathematical analysis37.1 (2005), pp. 17–59

    work page 2005

  36. [36]

    Metamorphoses through lie group action

    Alain Trouvé and Laurent Younes. “Metamorphoses through lie group action”. In:Foundations of computational mathematics5.2 (2005), pp. 173–198

    work page 2005

  37. [37]

    Diffeomorphic 3D image registration via geodesic shooting using an efficient adjoint calculation

    François-Xavier Vialard et al. “Diffeomorphic 3D image registration via geodesic shooting using an efficient adjoint calculation”. In:International Journal of Computer Vision97.2 (2012), pp. 229–241

    work page 2012

  38. [38]

    Analysis of AneuRisk65 data: Elastic shape registration of curves

    Qian Xie, Sebastian Kurtek, and Anuj Srivastava. “Analysis of AneuRisk65 data: Elastic shape registration of curves”. In:Electronic Journal of Statistics8.2 (2014), pp. 1920–1929.DOI: 10.1214/14- EJS938D .URL: https://doi.org/10.1214/14-EJS938D

    work page doi:10.1214/14- 2014

  39. [39]

    Laurent Younes.Shapes and diffeomorphisms. V ol. 171. Springer, 2010

    work page 2010

  40. [40]

    Springer Science & Business Media, 2013

    Eberhard Zeidler.Nonlinear functional analysis and its applications: III: variational methods and optimization. Springer Science & Business Media, 2013. 23

    work page 2013