Pith. sign in

REVIEW 3 major objections 5 minor 62 references

A two-stage pushforward turns closed-form linear posteriors into Bayesian uncertainty for nonlinear PDE imaging without MCMC.

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

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

T0 review · grok-4.5

2026-07-14 09:01 UTC pith:5C3XII5M

load-bearing objection Solid MCMC-free pushforward UQ for QPAT/EIT that correctly extends Koers et al.; QPAT theory is clean, EIT theory and cost claims are more conditional but still usable. the 3 major comments →

arxiv 2607.10817 v1 pith:5C3XII5M submitted 2026-07-12 math.ST stat.TH

An Efficient Bayesian Framework for Uncertainty Quantification in Nonlinear Imaging Inverse Problems

Anuj Abhishek , Sakshi Arya , Madhu Gupta This is my paper
classification math.ST stat.TH MSC 62G2062F1535R3065N21
keywords Bayesian inverse problemspushforward posteriorposterior contractionquantitative photoacoustic tomographyelectrical impedance tomographyuncertainty quantificationDirichlet-to-Neumann mapGaussian process priors
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Bayesian recovery of material parameters from PDE-based imaging is usually too slow because every MCMC step solves a PDE. This paper shows that for quantitative photoacoustic tomography and electrical impedance tomography one can instead put a Bayesian update on an auxiliary linear object (absorbed energy density, or a shifted Dirichlet-to-Neumann operator) whose posterior is available in closed form, then push that measure forward through a classical deterministic reconstruction map. The resulting pushforward is rigorously the Bayesian posterior under the induced prior, and it contracts around the true parameter at rates controlled by the auxiliary rate and the map's continuity. Numerically the method yields point estimates, pixelwise credible bands, and slice intervals at a fraction of the usual cost, because sampling stays exact and the expensive reconstructions can be run in parallel.

Core claim

The pushforward of a closed-form auxiliary posterior through a deterministic reconstruction map is itself the Bayesian posterior associated with the induced prior and the composed forward map; for QPAT and EIT this posterior contracts around the true absorption or conductivity, and exact samples of it can be drawn without MCMC.

What carries the argument

Two-stage pushforward (Theorem 3.1): first obtain the Gaussian (or low-rank Gaussian) posterior on the auxiliary variable, then push it through the recovery map e (pointwise division for QPAT) or R (D-bar/OOEIT for EIT) so that contraction and credibility transfer by the map's Lipschitz or modulus-of-continuity properties.

Load-bearing premise

The deterministic reconstruction map from operators to conductivities must stay continuous on sets that carry almost all posterior mass; if that continuity fails for the numerical solvers or truncated priors used, the transfer of contraction and credible regions to conductivity space no longer holds.

What would settle it

On a fixed EIT phantom with known conductivity, draw exact auxiliary posterior samples, push them through the same D-bar or OOEIT map used in the paper, and check whether the empirical L2 diameter of the conductivity ensemble shrinks at the rate predicted by the auxiliary contraction plus the claimed modulus of continuity as noise level decreases; systematic failure of that scaling falsifies the transfer theorems.

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

If this is right

  • Exact posterior samples for absorption (QPAT) or conductivity (EIT) can be generated by sampling a linear Gaussian model once and applying a classical reconstructor in parallel, eliminating sequential MCMC PDE solves.
  • Posterior contraction rates for the nonlinear parameters follow from known linear rates plus the Lipschitz or logarithmic stability of the reconstruction map.
  • Credible regions constructed on the auxiliary space push forward to credible regions of exact Bayesian credibility and frequentist coverage for the material parameters.
  • The same pipeline applies to any nonlinear inverse problem that admits a closed-form auxiliary posterior and a deterministic recovery map with controlled continuity.

Where Pith is reading between the lines

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

  • If the auxiliary prior can be chosen so that its support already lies inside the set where the reconstructor is uniformly continuous, the logarithmic stability of the Calderón problem need not prevent practical contraction of pixelwise credible intervals.
  • Operator-learning surrogates could replace the classical reconstructor R while still preserving the measure-theoretic interpretation of the pushforward posterior.
  • The framework suggests a natural hybrid workflow: use the cheap pushforward ensemble for real-time UQ, then refine only high-uncertainty regions with a short MCMC run.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper develops a two-stage pushforward Bayesian framework for the nonlinear inverse problems of QPAT (recovering absorption from internal energy density) and EIT (recovering conductivity from boundary DtN data). Bayesian inference is performed in closed form on an auxiliary linear problem (Gaussian regression for absorbed energy H or for a PCA-truncated shifted DtN operator), after which the auxiliary posterior is pushed forward through a deterministic reconstruction map (pointwise division for QPAT; D-bar or OOEIT for EIT) to induce a posterior on the parameter of interest. Theorem 3.1 supplies a measure-theoretic justification that the pushforward is the Bayesian posterior under the induced prior and transformed forward map. Posterior contraction rates are derived (Proposition 3.3 for QPAT via local Lipschitz of e; Theorem 3.7 for EIT under modulus-of-continuity assumptions on R), credible-region transfer is proved (Theorems 3.4 and 3.8), and numerical experiments illustrate reconstructions plus pixelwise/slicewise 95% credible intervals from 100 exact pushforward samples at modest noise levels.

Significance. If the claims hold, the work supplies a practical MCMC-free route to Bayesian UQ for two canonical PDE imaging problems, with exact sampling from the auxiliary posterior and parallelizable pushforward. Strengths include the clean measure-theoretic foundation (Theorem 3.1), the first contraction analysis for pushforward posteriors in this setting, explicit Lipschitz transfer for the QPAT map (Lemma 3.2), and concrete numerical pipelines that reuse existing D-bar and OOEIT solvers. The computational separation of closed-form linear inference from deterministic reconstruction is a genuine practical advantage over pCN-style methods that require millions of PDE solves. The EIT operator-space prior construction via empirical PCA of Matérn log-conductivity ensembles is a useful technical extension of the function-space framework of the cited precursor [41].

major comments (3)
  1. [Theorem 3.7, Assumption (A3); §3.2.1; §4.2] Theorem 3.7 (and the subsequent transfer in Theorem 3.8 / Corollary 3.9) rests on Assumption (A3): the deterministic reconstruction map R (D-bar or OOEIT Gauss-Newton) must satisfy a modulus of continuity ω on the high-posterior sets A_ε of the rank-R PCA Gaussian prior constructed in §3.2.1. The paper only invokes the classical logarithmic stability of the continuum Calderón problem; it never verifies that the concrete numerical solvers inherit any such modulus on the support of the truncated PCA prior actually used in the experiments of §4.2. Without this, operator-space contraction does not imply L^{2} conductivity contraction, so the pushforward posterior need not concentrate and the credible regions of Figures 4–5 lack the theoretical coverage guarantee claimed in Theorem 3.8. This is load-bearing for the EIT half of the central claim.
  2. [Theorem 3.1; Theorem 3.8] Theorem 3.1 requires the reconstruction map (e for QPAT, R for EIT) to be injective with Borel-measurable inverse on the relevant support. For the numerical EIT maps this is not established; D-bar and iterative OOEIT reconstructions are known to be non-injective under noise and truncation. The paper proceeds as if the pushforward relation still yields a well-defined Bayesian posterior, but the Radon–Nikodym representation and the exact credibility/coverage identities of Theorem 3.8 become formal rather than rigorous once injectivity fails.
  3. [Abstract; §1; §4; §5] The abstract, introduction and conclusion repeatedly claim “arguably lower” / “significantly lower” computational cost than standard MCMC Bayesian approaches, yet §4 contains no wall-clock timings, no PDE-solve counts, and no comparison against a baseline (e.g., pCN or delayed-acceptance MCMC with the same forward solvers). The only quantitative remark is that 100 exact samples are drawn, which is not itself evidence of cost superiority. Either supply a controlled cost comparison or temper the claim to “MCMC-free exact sampling of the auxiliary posterior.”
minor comments (5)
  1. [Abstract] Abstract and page 1: “at a arguably lower” → “at an arguably lower”.
  2. [§4] Figure 3–5 captions and §4.1–4.2: the number of pushforward samples (100) and the precise noise model (relative vs absolute) should be stated uniformly in every caption; currently they appear only in the text.
  3. [§3] Notation for the reconstruction map switches between e (QPAT) and R (EIT) without a single unifying symbol; a short remark that both play the role of the Borel map of Theorem 3.1 would improve readability.
  4. [§4.1.1; §4.2.1] The free parameters of the Matérn priors (length-scales ℓ = 0.06 for QPAT, 0.15/0.05 for EIT) and the PCA rank R = 300 are chosen by “empirical Bayes” or fixed ad hoc; a short sensitivity check or cross-validation statement would strengthen reproducibility.
  5. [§1] Reference [41] is the methodological precursor; a clearer one-sentence statement of what is new versus what is directly inherited would help the reader locate the contribution.

Circularity Check

0 steps flagged

No load-bearing circularity: auxiliary closed-form posterior is pushed by an independent deterministic map; self-cites supply background only.

full rationale

The derivation chain begins with a linear Gaussian observation model on an auxiliary variable (absorbed energy density H for QPAT; shifted DtN operator for EIT) whose posterior is available in closed form under a Gaussian/Matérn or PCA-Gaussian prior (Sections 3.1.2 and 3.2.2). Theorem 3.1 then supplies a standard measure-theoretic identity: the pushforward of that posterior through a Borel-measurable injective reconstruction map e (or R) is exactly the Bayesian posterior for the induced prior and the composed forward map. Contraction rates (Prop. 3.3, Thm. 3.7) transfer the known auxiliary rates via the map’s Lipschitz or modulus-of-continuity properties; these properties are either proved (local Lipschitz of the explicit QPAT map e) or assumed (A3 for EIT) rather than derived from the target posterior itself. Numerical length-scale choices are empirical-Bayes and fixed; they are not re-used as “predictions.” Citations to the authors’ earlier QPAT/EIT papers and to the external two-stage framework [41] provide context and classical stability estimates, not the identity of the pushforward posterior. The construction therefore does not reduce the claimed posterior or its contraction rate to its own inputs by definition or by self-citation chain. The only residual softness (unproved modulus for the concrete numerical solvers D-bar/OOEIT under the truncated PCA prior) is an assumption gap, not circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claim rests on standard Bayesian nonparametric and elliptic PDE theory plus domain regularity of reconstruction maps; free parameters appear only in numerics (prior hyperparameters, PCA rank, sample sizes). No new physical entities are postulated. The main non-standard modeling choices are the Gaussian PCA approximation of the pushforward prior on operators and the transfer of contraction through unproved continuity of numerical EIT solvers on the truncated prior support.

free parameters (5)
  • Matérn length-scale ℓ (QPAT H prior) = 0.06
    Chosen by empirical Bayes then fixed (ℓ=0.06); controls posterior smoothness and credible-band width for the auxiliary field.
  • Matérn length-scales for log-conductivity ensemble (EIT) = 0.15 (D-bar), 0.05 (OOEIT)
    Hand-chosen differently for D-bar (0.15) and OOEIT (0.05); shapes the empirical PCA prior on boundary operators.
  • PCA truncation rank R for DtN prior = 300
    Rank-300 low-rank Gaussian approximation of the operator prior; truncation bias enters the EIT contraction rate via b_R.
  • Number of conductivity ensemble samples N for EIT prior = 2000
    N=2000 samples used to form empirical mean and covariance of boundary operators; affects quality of the Gaussian approximation.
  • Posterior sample count for pushforward UQ = 100
    100 exact auxiliary posterior samples pushed through reconstruction maps; determines empirical quantiles and means in figures.
axioms (5)
  • standard math Gaussian process priors under linear white-noise observation models yield closed-form Gaussian posteriors with known contraction rates (Knapik et al.-type theory).
    Used for auxiliary posteriors in QPAT §3.1.2 and EIT sequence model §3.2.2; Assumption (A1) in Prop. 3.3.
  • domain assumption Uniformly elliptic operators with Lipschitz coefficients and positive boundary data give unique positive weak solutions in H1.
    Underpins well-definedness of u=Kv+g̃ and the pointwise map e(v)=v/u for QPAT (Section 2.1, Lemma 3.2).
  • domain assumption Shifted DtN maps for admissible conductivities lie in Hilbert–Schmidt spaces Hr and the Calderón problem has logarithmic stability.
    Cited from Abraham–Nickl [2]; used for observation model (2.7) and Corollary 3.9 diameter bounds.
  • ad hoc to paper The deterministic reconstruction map R (D-bar or iterative OOEIT) is Borel measurable, injective on the relevant support, and continuous in a modulus-of-continuity sense on high-posterior sets.
    Required by Theorem 3.1 and Assumption (A3) of Theorem 3.7; not proved for the numerical solvers on the truncated PCA prior support.
  • ad hoc to paper Empirical PCA of DtN matrices from Matérn log-conductivity samples adequately approximates the pushforward prior for posterior contraction analysis.
    Construction in §3.2.1 following Calvetti et al. [14]; Gaussian approximation replaces the true nonlinear pushforward law without quantitative error bounds in the theory.

pith-pipeline@v1.1.0-grok45 · 28715 in / 3780 out tokens · 53680 ms · 2026-07-14T09:01:46.656236+00:00 · methodology

0 comments
read the original abstract

Bayesian methods provide a natural framework for estimating a parameter in non-linear inverse problems and quantifying uncertainty in the estimation. However, when the forward model for such non-linear inverse problems is given by some Partial Differential Equation (PDE), Bayesian inference is typically carried out by resorting to MCMC methods. Since each MCMC iteration requires solving a PDE, these methods become computationally expensive and are often impractical for large-scale imaging problems. In this work, we develop a computationally efficient Bayesian framework for two such nonlinear imaging inverse problems: Quantitative Photoacoustic Tomography (QPAT) and Electrical Impedance Tomography (EIT). Building on a recently proposed two-stage pushforward methodology, we first formulate a Bayesian regression problem for an auxiliary variable whose posterior is available in closed form. This posterior is then pushed forward through a deterministic reconstruction map to obtain a posterior on the unknown parameter, avoiding MCMC sampling. We give a rigorous measure-theoretic justification to interpret the induced posterior as a Bayesian posterior and derive posterior contraction rates for both QPAT and EIT. Numerical results show that the proposed method provides accurate reconstructions and reliable uncertainty estimates at a arguably lower computational cost than standard Bayesian approaches.

Figures

Figures reproduced from arXiv: 2607.10817 by Anuj Abhishek, Madhu Gupta, Sakshi Arya.

Figure 1
Figure 1. Figure 1: Two-step Bayesian procedure: inference is first performed on the auxiliary variable [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Two-step Bayesian procedure: inference is first performed on the auxiliary variable [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Posterior mean reconstructions for QPAT (4% relative noise): The first column shows the [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Posterior reconstructions for EIT using D-bar method (2% relative noise): The first [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Posterior reconstructions for EIT using OOEIT method (2% relative noise): The first [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗

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

62 extracted references · 1 linked inside Pith

  1. [1]

    However, extending the framework introduced in [41] to specific inverse problems is not straight- forward and requires careful problem-specific analysis

    and projection-posterior methods, [50]. However, extending the framework introduced in [41] to specific inverse problems is not straight- forward and requires careful problem-specific analysis. For QPAT, one needs to identify a suitable auxiliary variable and establish the required properties of the corresponding reconstruction map. For EIT, the auxiliary...

  2. [2]

    For completeness’ sake, we give a self-contained proof of the above statement

    in a slightly different setting. For completeness’ sake, we give a self-contained proof of the above statement. Proof.In order to do the proof, we recall the abstract set-up from [29, section 2.6]. Clearly,H r is a seperable Hilbert space and the mapG: Γ→H r given byG(σ) = ˜Λσ is measurable. Thus it follows from [29, Lemma 2.6.3] that Λ =E( ˜Λσ) := R Hr ˜...

    2000

  3. [3]

    Abhishek, M

    A. Abhishek, M. Bonnet, and S. Moskow. Modified forward and inverse Born series for the Calderon and diffuse-wave problems.Inverse Problems, 36(11):114001, 19, 2020

    2020

  4. [4]

    Abraham and R

    K. Abraham and R. Nickl. On statistical Calder´ on problems.Math. Stat. Learn., 2(2):165–216, 2019

    2019

  5. [5]

    B. M. Afkham, K. Knudsen, A. K. Rasmussen, and T. Tarvainen. A bayesian approach for consistent reconstruction of inclusions.Inverse Problems, 40(4):045004, 2024

    2024

  6. [6]

    J. P. Agnelli, A. C ¸ ¨ ol, M. Lassas, R. Murthy, M. Santacesaria, and S. Siltanen. Classifica- tion of stroke using neural networks in electrical impedance tomography.Inverse Problems, 36(11):115008, oct 2020

    2020

  7. [7]

    Ammari, editor.Mathematical modeling in biomedical imaging

    H. Ammari, editor.Mathematical modeling in biomedical imaging. II, volume 2035 ofLecture Notes in Mathematics. Springer, Heidelberg, 2012. Optical, ultrasound, and opto-acoustic tomographies, Mathematical Biosciences Subseries

    2035

  8. [8]

    Arridge, S

    S. Arridge, S. Moskow, and J. C. Schotland. Inverse Born series for the Calderon problem. Inverse Problems, 28(3):035003, 16, 2012

    2012

  9. [9]

    A. B. Bakushinski˘ i. Iterative methods for solving nonlinear operator equations without regularity. A new approach.Dokl. Akad. Nauk, 330(3):282–284, 1993

    1993

  10. [10]

    G. Bal. Hybrid inverse problems and internal functionals. InInverse problems and applications: inside out. II, volume 60 ofMath. Sci. Res. Inst. Publ., pages 325–368. Cambridge Univ. Press, Cambridge, 2013

    2013

  11. [11]

    Bal and K

    G. Bal and K. Ren. Multi-source quantitative photoacoustic tomography in a diffusive regime. Inverse Problems, 27(7):075003, 20, 2011. 26

    2011

  12. [12]

    Bazahica, V

    L. Bazahica, V. Kaarnioja, and L. Roininen. Uncertainty quantification for electrical impedance tomography using quasi-monte carlo methods.Inverse Problems, 41(6):065002, 2025

    2025

  13. [13]

    L. Borcea. Electrical impedance tomography.Inverse problems, 18(6):R99, 2002

    2002

  14. [14]

    Calder´ on

    A.-P. Calder´ on. On an inverse boundary value problem. InSeminar on Numerical Analysis and its Applications to Continuum Physics (Rio de Janeiro, 1980), pages 65–73. Soc. Brasil. Mat., Rio de Janeiro, 1980

    1980

  15. [15]

    Calvetti, J

    D. Calvetti, J. P. Kaipio, and E. Somersalo. Inverse problems in the Bayesian framework [Preface].Inverse Problems, 30(11):110301, 4, 2014

    2014

  16. [16]

    Calvetti, S

    D. Calvetti, S. Nakkireddy, and E. Somersalo. Approximation of continuous eit data from electrode measurements with bayesian methods.Inverse Problems, 35(4):045012, mar 2019

    2019

  17. [17]

    Calvetti and E

    D. Calvetti and E. Somersalo. Inverse problems: From regularization to Bayesian inference. WIREs Computational Statistics, 10(3):e1427, 2018

    2018

  18. [18]

    Castro, C

    J. Castro, C. Mu˜ noz, and N. Valenzuela. The Calder´ on’s problem via DeepONets.Vietnam J. Math., 52(3):775–806, 2024

    2024

  19. [19]

    Cheney, D

    M. Cheney, D. Isaacson, and J. C. Newell. Electrical impedance tomography.SIAM review, 41(1):85–101, 1999

    1999

  20. [20]

    Cherepenin, A

    V. Cherepenin, A. Karpov, A. Korjenevsky, V. Kornienko, Y. Kultiasov, M. Ochapkin, O. Trochanova, and J. Meister. Three-dimensional eit imaging of breast tissues: system design and clinical testing.IEEE Transactions on Medical Imaging, 21(6):662–667, 2002

    2002

  21. [21]

    S. L. Cotter, G. O. Roberts, A. M. Stuart, and D. White. Mcmc methods for functions: modifying old algorithms to make them faster.Statistical Science, pages 424–446, 2013

    2013

  22. [22]

    T. Cui, Y. Marzouk, and K. Willcox. Scalable posterior approximations for large-scale Bayesian inverse problems via likelihood-informed parameter and state reduction.Journal of Computa- tional Physics, 315:363–387, 2016

    2016

  23. [23]

    T. Cui, Y. M. Marzouk, and K. E. Willcox. Data-driven model reduction for the Bayesian solution of inverse problems.International Journal for Numerical Methods in Engineering, 102(5):966–990, 2015

    2015

  24. [24]

    Dashti and A

    M. Dashti and A. M. Stuart.The Bayesian Approach to Inverse Problems. Lecture Notes,

  25. [25]

    Available as arXiv:1302.6989

    Pith/arXiv arXiv

  26. [26]

    Debnath and P

    L. Debnath and P. Mikusi´ nski.Introduction to Hilbert Spaces with Applications. Academic Press, Amsterdam; Boston, 3rd edition, 2005

    2005

  27. [27]

    M. M. Dunlop and A. M. Stuart. The Bayesian formulation of EIT: analysis and algorithms. Inverse Probl. Imaging, 10(4):1007–1036, 2016

    2016

  28. [28]

    H. W. Engl, M. Hanke, and A. Neubauer.Regularization of inverse problems, volume 375. Springer Science & Business Media, 1996. 27

    1996

  29. [29]

    L. C. Evans.Partial differential equations, volume 19 ofGraduate Studies in Mathematics. American Mathematical Society, Providence, RI, 1998

    1998

  30. [30]

    Frerichs, J

    I. Frerichs, J. Hinz, P. Herrmann, G. Weisser, G. Hahn, M. Quintel, and G. Hellige. Regional lung perfusion as determined by electrical impedance tomography in comparison with electron beam ct imaging.IEEE Transactions on Medical Imaging, 21(6):646–652, 2002

    2002

  31. [31]

    Z. Gao, L. Yan, and T. Zhou. Adaptive operator learning for infinite-dimensional Bayesian inverse problems.SIAM/ASA Journal on Uncertainty Quantification, 12(4):1389–1423, 2024

    2024

  32. [32]

    Gin´ e and R

    E. Gin´ e and R. Nickl.Mathematical Foundations of Infinite-Dimensional Statistical Models. Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, 2021

    2021

  33. [33]

    Gupta, R

    M. Gupta, R. K. Mishra, and S. Roy. Sparse reconstruction of log-conductivity in current density impedance tomography.J. Math. Imaging Vision, 62(2):189–205, 2020

    2020

  34. [34]

    Gupta, R

    M. Gupta, R. K. Mishra, and S. Roy. Sparsity-based nonlinear reconstruction of optical parameters in two-photon photoacoustic computed tomography.Inverse Problems, 37(4):Paper No. 044001, 24, 2021

    2021

  35. [35]

    Hanke, N

    M. Hanke, N. Hyv¨ onen, and S. Reusswig. Convex backscattering support in electric impedance tomography.Numer. Math., 117(2):373–396, 2011

    2011

  36. [36]

    T. Hohage. Logarithmic convergence rates of the iteratively regularized Gauss-Newton method for an inverse potential and an inverse scattering problem.Inverse Problems, 13(5):1279–1299, 1997

    1997

  37. [37]

    M. A. Iglesias, K. J. H. Law, and A. M. Stuart. Ensemble Kalman methods for inverse problems.Inverse Problems, 29(4):045001, 20, 2013

    2013

  38. [38]

    B. Jin, T. Khan, and P. Maass. A reconstruction algorithm for electrical impedance to- mography based on sparsity regularization.International Journal for Numerical Methods in Engineering, 89(3):337–353, 2012

    2012

  39. [39]

    Jin and P

    B. Jin and P. Maass. An analysis of electrical impedance tomography with applications to Tikhonov regularization.ESAIM Control Optim. Calc. Var., 18(4):1027–1048, 2012

    2012

  40. [40]

    Kaipio and E

    J. Kaipio and E. Somersalo.Statistical and computational inverse problems, volume 160. Springer Science & Business Media, 2006

    2006

  41. [41]

    J. P. Kaipio, V. Kolehmainen, E. Somersalo, and M. Vauhkonen. Statistical inversion and Monte Carlo sampling methods in electrical impedance tomography.Inverse Problems, 16(5):1487–1522, 2000

    2000

  42. [42]

    Kaltenbacher, A

    B. Kaltenbacher, A. Neubauer, and O. Scherzer.Iterative regularization methods for nonlinear ill-posed problems, volume 6 ofRadon Series on Computational and Applied Mathematics. Walter de Gruyter GmbH & Co. KG, Berlin, 2008

    2008

  43. [43]

    B. T. Knapik, A. W. van der Vaart, and J. H. van Zanten. Bayesian inverse problems with gaussian priors.The Annals of Statistics, 39(5):2626–2657, 2011. Accessed: 2026-05-17. 28

    2011

  44. [44]

    Koers, B

    G. Koers, B. Szabo, and A. van der Vaart. Linear methods for non-linear inverse problems, 2024

    2024

  45. [45]

    Kuchment.The Radon transform and medical imaging, volume 85 ofCBMS-NSF Regional Conference Series in Applied Mathematics

    P. Kuchment.The Radon transform and medical imaging, volume 85 ofCBMS-NSF Regional Conference Series in Applied Mathematics. Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 2014

    2014

  46. [46]

    Kuusela and M

    P. Kuusela and M. Vauhkonen. Ooeit: An open source object oriented electrical impedance tomography software package.Applied Mathematics for Modern Challenges, 2(2):243–261, 2024

    2024

  47. [47]

    Kuusela and M

    P. Kuusela and M. Vauhkonen. OOEIT: Object Oriented Electrical Impedance Tomography, an open source code package for Matlab.GitHub, 2024

    2024

  48. [48]

    Li and Y

    J. Li and Y. M. Marzouk. Adaptive construction of surrogates for the Bayesian solution of inverse problems.SIAM Journal on Scientific Computing, 36(3):A1163–A1186, 2014

    2014

  49. [49]

    Majee, A

    S. Majee, A. Abhishek, T. Strauss, and T. Khan. MCMC-Net: accelerating Markov Chain Monte Carlo with neural networks for inverse problems.Inverse Problems, 41(9):Paper No. 095013, 38, 2025

    2025

  50. [50]

    Mueller, S

    J. Mueller, S. Siltanen, and J. Tamminen. D-bar method for EIT.GitHub, 2024

    2024

  51. [51]

    J. L. Mueller and S. Siltanen. Direct reconstructions of conductivities from boundary mea- surements.SIAM J. Sci. Comput., 24(4):1232–1266, 2003

    2003

  52. [52]

    J. L. Mueller and S. Siltanen. The D-bar method for electrical impedance tomography— demystified.Inverse Problems, 36(9):093001, 28, 2020

    2020

  53. [53]

    Pal and S

    S. Pal and S. Ghosal. Projection-posterior for variable selection: weak limit and coverage. Electron. J. Stat., 19(2):3743–3770, 2025

    2025

  54. [54]

    Somersalo, M

    E. Somersalo, M. Cheney, and D. Isaacson. Existence and uniqueness for electrode models for electric current computed tomography.SIAM J. Appl. Math., 52(4):1023–1040, 1992

    1992

  55. [55]

    A. M. Stuart. Inverse problems: A bayesian perspective.Acta Numerica, 19:451–559, 2010

    2010

  56. [56]

    Tarantola.Inverse problem theory and methods for model parameter estimation

    A. Tarantola.Inverse problem theory and methods for model parameter estimation. SIAM, 2005

    2005

  57. [57]

    Tarvainen

    T. Tarvainen. Quantitative photoacoustic tomography in Bayesian framework. InThe Radon transform—the first 100 years and beyond, volume 22 ofRadon Ser. Comput. Appl. Math., pages 239–271. Walter de Gruyter, Berlin, [2019]©2019

    2019

  58. [58]

    Tarvainen and B

    T. Tarvainen and B. Cox. Quantitative photoacoustic tomography: Modeling and inverse problems.Journal of Biomedical Optics, 29(S1):S11509, 2024

    2024

  59. [59]

    Tarvainen, B

    T. Tarvainen, B. T. Cox, J. P. Kaipio, and S. R. Arridge. Reconstructing absorption and scat- tering distributions in quantitative photoacoustic tomography.Inverse Problems, 28(8):084009, 17, 2012. 29

    2012

  60. [60]

    G. Uhlmann. Electrical impedance tomography and Calder´ on’s problem.Inverse Problems, 25(12):123011, 39, 2009

    2009

  61. [61]

    Yan and Y.-X

    L. Yan and Y.-X. Zhang. Convergence analysis of surrogate-based methods for Bayesian inverse problems.Inverse Problems, 33(12):125001, 2017

    2017

  62. [62]

    L. Y. Zhou et al. An adaptive surrogate modeling based on deep neural networks for large- scale Bayesian inverse problems.Communications in Computational Physics, 28(5):2180–2205, 2020. 30

    2020