pith. sign in

arxiv: 2604.26254 · v1 · submitted 2026-04-29 · 🧮 math.NA · cs.NA

Spotlight, priorsketching and Bayesian approximation error paradigms

Daniela Calvetti , Erkki Somersalo This is my paper

Pith reviewed 2026-05-07 13:12 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords Bayesian approximation errorspotlight inversioninverse problemsmodeling errorssketching schemesrandomized linear algebratomography
0
0 comments X

The pith

The Bayesian approximation error and spotlight inversion approaches are closely related but not equivalent methods for addressing modeling errors in inverse problems.

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

This paper analyzes two strategies for managing the effects of using approximate models in large-scale inverse problems, where discrepancies often cause artifacts or blurring in solutions. The Bayesian approximation error (BAE) method and linear algebraic spotlight inversion both aim to suppress these clutter effects. By comparing them, the authors demonstrate that the methods are closely related through their handling of model discrepancies but are not the same, and they share connections with sketching schemes from randomized linear algebra. This relation is demonstrated through examples in X-ray tomography and electrical impedance tomography, where both methods successfully reduce most of the unwanted effects.

Core claim

The paper shows that the Bayesian approximation error (BAE) method and spotlight inversion, while both effective at suppressing clutter from approximate models in inverse problems, are closely related but not equivalent, with a connection to sketching schemes in randomized linear algebra.

What carries the argument

Orthogonal projections in spotlight inversion to suppress clutter, paralleled by prior sketching in the BAE approach for handling model discrepancy.

If this is right

  • The similarities suggest that techniques from randomized linear algebra could enhance both methods for computational efficiency.
  • Successful application in tomography indicates potential for broader use in other ill-posed inverse problems.
  • Non-equivalence implies that the choice between methods depends on the specific type of modeling error and computational constraints.
  • Both methods lower computational cost by allowing simpler models without major loss in solution quality.

Where Pith is reading between the lines

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

  • Hybrid approaches combining projection-based suppression with Bayesian priors might yield even better artifact reduction.
  • The sketching connection could inspire new parameter-free methods for model error correction in high-dimensional problems.
  • Similar paradigms might apply to machine learning models where approximate architectures are used.

Load-bearing premise

The model discrepancy from approximate models manifests primarily as suppressible artifacts or blurring that the BAE and spotlight methods can address for the specific types of modeling errors considered.

What would settle it

An inverse problem example where using the spotlight projection or BAE adjustment leaves significant artifacts or blurring that cannot be explained by other factors.

Figures

Figures reproduced from arXiv: 2604.26254 by Daniela Calvetti, Erkki Somersalo.

Figure 1
Figure 1. Figure 1: A schematic picture of the coarsening. Here, the pink square is the region of interest view at source ↗
Figure 2
Figure 2. Figure 2: On the left, we show the least squares solution of the problem view at source ↗
Figure 3
Figure 3. Figure 3: On the left we show the reconstruction x n naive obtained by ignoring the approximation error and solving the equation A nx n = b by the LSQR algorithm, stopping the iterations when the discrepancy reaches the noise level. On the right, we show the difference xnaive − xref. absorption coefficient is non-negative, we define a non-Gaussian prior model, writing X = γ Σξ0,α Ξ  , (29) where Ξ is a Gaussian ran… view at source ↗
Figure 4
Figure 4. Figure 4: Three draws from the prior model for the fine grain variable (top row) and the corresponding coarse view at source ↗
Figure 5
Figure 5. Figure 5: On the left, we show the reconstruction x n BME obtained by using the BAE reduced model. The solution is calculated by using the Conjugate Gradient solver regularized by terminating the iterations as soon as the discrepancy norm reached the noise level. On the right, we show the difference xBAE − xref. To test this approach, we generate a sample of 250 draws from the prior described in the previous section… view at source ↗
Figure 6
Figure 6. Figure 6: The spotlight solution x n spot, the range of the approximation error being estimated by using the prior sketching, and the difference x n spot − x n ref. 4.2 Electrical impedance tomography In the second example, we consider the inverse problem of electrical impedance tomography (EIT). Given a bounded domain Ω with connected complement, the goal is to estimate the electric conductivity σ(x) > 0 inside Ω b… view at source ↗
Figure 7
Figure 7. Figure 7: The left panel shows the generative model. The conductivity is set to the background value view at source ↗
Figure 8
Figure 8. Figure 8: Three random conductivity structures inside the corresponding randomly varying domains. view at source ↗
Figure 9
Figure 9. Figure 9: On the left, reconstruction obtained by ignoring the approximation error. The geometry artifacts are view at source ↗
read the original abstract

A way to lower computational cost in large scale inverse problems and problems depending on poorly known model parameters is to replace the detailed model by an approximate one. Inverse problems are typically ill-posed, and the model discrepancy introduced by using approximate models often shows up in the computed solutions as disturbing artifacts or blurring. In this article, we consider two methods of addressing certain types of modeling errors, the Bayesian approximation error (BAE) method and linear algebraic spotlight inversion to suppress clutter in the computational model by orthogonal projections. Through the process of analyzing the two approaches, we show that they turn out to be closely related but not equivalent, and we highlight a connection to sketching schemes in randomized linear algebra. The similarities between the methods and their successful suppression of most of the clutter effects is elucidated with two computed examples, one addressing of X-ray tomography and the other electrical impedance tomography.

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 analyzes the Bayesian approximation error (BAE) method and linear algebraic spotlight inversion (also linked to priorsketching) as approaches to mitigate modeling discrepancies in large-scale inverse problems. It establishes that the two methods are closely related but not equivalent, draws an explicit connection to sketching schemes from randomized linear algebra, and illustrates successful suppression of clutter artifacts via two numerical examples in X-ray computed tomography and electrical impedance tomography.

Significance. If the derivations and non-equivalence argument hold, the work offers a useful unification of statistical (BAE) and algebraic (spotlight/projection) techniques for handling approximation errors that commonly produce artifacts or blurring in ill-posed inversions. The sketching link may enable transfer of randomized linear algebra tools for reduced computational cost. The computed examples provide concrete evidence of practical utility for the targeted error regimes.

minor comments (3)
  1. The abstract states that the methods 'turn out to be closely related but not equivalent' after analysis; the main text should include an explicit side-by-side comparison (e.g., a table or equations) of the two operators or posterior covariances to make the distinction immediate for readers.
  2. In the X-ray CT and EIT examples, quantitative metrics (relative L2 errors, PSNR, or residual norms against a reference full-model solution) would strengthen the claim of 'successful suppression of most of the clutter effects' beyond visual inspection.
  3. The title references 'priorsketching' while the abstract and body emphasize 'spotlight inversion'; a brief clarifying sentence on the relationship between these terms would improve readability.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, recognition of the unification between Bayesian approximation error and spotlight/projection methods, and the link to randomized linear algebra sketching. The recommendation for minor revision is appreciated, and we will prepare the revised manuscript accordingly. Since no specific major comments were listed in the report, we provide no point-by-point responses below.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper performs an explicit comparative analysis of the Bayesian approximation error (BAE) method and spotlight inversion for suppressing modeling artifacts in inverse problems, concluding they are closely related but not equivalent while noting a connection to sketching schemes. This relation is derived from direct examination of the two approaches and demonstrated via independent computed examples (X-ray CT and EIT), without any reduction of the central claim to a fitted parameter, self-definition, or load-bearing self-citation chain. The derivation chain remains self-contained and externally verifiable through the provided examples and algebraic comparisons.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The abstract relies on standard frameworks from Bayesian inverse problems and linear algebra without introducing new free parameters, axioms beyond domain norms, or invented entities.

axioms (2)
  • domain assumption Inverse problems are typically ill-posed and model discrepancies appear as artifacts or blurring.
    Stated directly in the abstract as the motivation for the methods.
  • domain assumption Orthogonal projections and Bayesian error modeling can suppress certain modeling errors.
    Core premise underlying both spotlight and BAE approaches.

pith-pipeline@v0.9.0 · 5444 in / 1290 out tokens · 50604 ms · 2026-05-07T13:12:38.767099+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

22 extracted references · 22 canonical work pages

  1. [1]

    S. R. ARRIDGE, J. P. KAIPIO, V. KOLEHMAINEN, M. SCHWEIGER, E. SOMERSALO, T. TARVAINEN, ANDM. VAUHKONEN,Approximation errors and model reduction with an application in optical diffusion tomography, Inverse problems, 22 (2006), pp. 175–195. 21

    work page 2006

  2. [2]

    T. A. BUBBA, A. HAUPTMANN, S. HUOTARI, J. RIMPEL ¨AINEN,ANDS. SILTANEN,Tomographic X-ray data of a lotus root filled with attenuating objects, Sept. 2016

    work page 2016

  3. [3]

    CALVETTI, M

    D. CALVETTI, M. DUNLOP, E. SOMERSALO,ANDA. STUART,Iterative updating of model error for bayesian inversion, Inverse Problems, 34 (2018), p. 025008

    work page 2018

  4. [4]

    CALVETTI, N

    D. CALVETTI, N. HYV ¨ONEN, V. KOLEHMAINEN,ANDE. SOMERSALO,Spotlight inversion by orthogonal projections, arXiv preprint arXiv:2509.15512, (2025)

    work page arXiv 2025

  5. [5]

    CALVETTI, M

    D. CALVETTI, M. PRAGLIOLA,ANDE. SOMERSALO,Sparsity-promoting hierarchical bayesian model for eit with a blocky target, Journal of Computational Physics, (2025), p. 114255

    work page 2025

  6. [6]

    CALVETTI ANDE

    D. CALVETTI ANDE. SOMERSALO,Hypermodels in the bayesian imaging framework, Inverse Problems, 24 (2008), p. 034013

    work page 2008

  7. [7]

    ,Bayesian scientific computing, Springer, 2023

    work page 2023

  8. [8]

    CANDIANI, A

    V. CANDIANI, A. HANNUKAINEN,ANDN. HYVONEN,Computational framework for applying electrical impedance tomography to head imaging, SIAM Journal on Scientific Computing, 41 (2019), pp. B1034– B1060

    work page 2019

  9. [9]

    J. M. HUTTUNEN, J. P. KAIPIO,ANDE. SOMERSALO,Approximation errors in nonstationary inverse prob- lems, Inverse problems and imaging, 1 (2007), p. 77

    work page 2007

  10. [10]

    HYV ¨ONEN, H

    N. HYV ¨ONEN, H. MAJANDER,ANDS. STABOULIS,Compensation for geometric modeling errors by posi- tioning of electrodes in electrical impedance tomography, Inverse Problems, 33 (2017), p. 035006

    work page 2017

  11. [11]

    J ¨A ¨ASKEL ¨AINEN, J

    A. J ¨A ¨ASKEL ¨AINEN, J. TOIVANEN, A. H ¨ANNINEN, V. KOLEHMAINEN,ANDN. HYV ¨ONEN,Projection- based preprocessing for electrical impedance tomography to reduce the effect of electrode contacts, SIAM Journal on Imaging Sciences, 18 (2025), pp. 1681–1706

    work page 2025

  12. [12]

    KAIPIO ANDV

    J. KAIPIO ANDV. KOLEHMAINEN,Approximate marginalization over modeling errors and uncertainties in inverse problems, Bayesian theory and applications, (2013), pp. 644–672

    work page 2013

  13. [13]

    KAIPIO ANDE

    J. KAIPIO ANDE. SOMERSALO,Statistical inverse problems: discretization, model reduction and inverse crimes, Journal of computational and applied mathematics, 198 (2007), pp. 493–504

    work page 2007

  14. [14]

    J. P. KAIPIO, T. HUTTUNEN, T. LUOSTARI, T. L ¨AHIVAARA,ANDP. B. MONK,A bayesian approach to improving the born approximation for inverse scattering with high-contrast materials, Inverse Problems, 35 (2019), p. 084001

    work page 2019

  15. [15]

    J. P. KAIPIO ANDE. SOMERSALO,Statistical and computational inverse problems, Springer, 2005

    work page 2005

  16. [16]

    M. E. KILMER ANDD. P. O’LEARY,Choosing regularization parameters in iterative methods for ill-posed problems, SIAM Journal on matrix analysis and applications, 22 (2001), pp. 1204–1221

    work page 2001

  17. [17]

    KOLEHMAINEN, M

    V. KOLEHMAINEN, M. LASSAS,ANDP. OLA,Electrical impedance tomography problem with inaccurately known boundary and contact impedances, IEEE transactions on medical imaging, 27 (2008), pp. 1404–1414

    work page 2008

  18. [18]

    KOLEHMAINEN, M

    V. KOLEHMAINEN, M. LASSAS, P. OLA,ANDS. SILTANEN,Recovering boundary shape and conductivity in electrical impedance tomography, Inverse Problems and Imaging, 7 (2013), pp. 217–242

    work page 2013

  19. [19]

    KOULOURI, V

    A. KOULOURI, V. RIMPIL ¨AINEN, M. BROOKES,ANDJ. P. KAIPIO,Compensation of domain modelling errors in the inverse source problem of the poisson equation: Application in electroencephalographic imaging, Applied Numerical Mathematics, 106 (2016), pp. 24–36. 22

    work page 2016

  20. [20]

    NISSINEN, V

    A. NISSINEN, V. KOLEHMAINEN,ANDJ. P. KAIPIO,Reconstruction of domain boundary and conductivity in electrical impedance tomography using the approximation error approach, Visualization of Mechanical Processes: An International Online Journal, 1 (2011)

    work page 2011

  21. [21]

    NISSINEN, V

    A. NISSINEN, V. P. KOLEHMAINEN,ANDJ. P. KAIPIO,Compensation of modelling errors due to unknown domain boundary in electrical impedance tomography, IEEE transactions on medical imaging, 30 (2010), pp. 231–242

    work page 2010

  22. [22]

    SOMERSALO, M

    E. SOMERSALO, M. CHENEY,ANDD. ISAACSON,Existence and uniqueness for electrode models for electric current computed tomography, SIAM Journal on Applied Mathematics, 52 (1992), pp. 1023–1040. 23

    work page 1992