Deep kernel learning for integral measurements

Adrian Wills; Carl Jidling; Johannes Hendriks; Thomas B. Sch\"on

REVIEW 4 major objections 4 minor 57 references

Deep kernel learning for integral measurements

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A finite sine-basis expansion makes deep kernel learning tractable for line-integral measurements such as computed tomography.

desk verdict A genuinely useful extension of deep kernel learning to line-integral data, with a real computational trick at its core, but the evidence is visual and the key approximation is not stress-tested. read the letter →

arxiv 1909.01844 v1 pith:O65MXHRH submitted 2019-09-04 stat.ML cs.LG

Carl Jidling , Johannes Hendriks , Thomas B. Schön , Adrian Wills This is my paper · ORCID

classification stat.MLcs.LG

keywords deepkernellearningGaussianprocesslineintegralmeasurementscomputedtomographynon-stationarycovarianceHilbertspaceapproximationneuralnetworkpre-traininglimited-datareconstruction

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper sets out to make deep kernel learning work when the data are line integrals of the target function, as in X-ray computed tomography. The authors argue that a finite Hilbert space basis expansion of the covariance function removes the main obstacle: naive implementations need numerical double integrals whose number grows quadratically with data, while the expansion reduces each entry to a single integral and makes total cost linear in data size. They add a practical pre-training step in which the neural-network input warp is fitted to a standard Gaussian-process reconstruction before joint training, which they find to be crucial for convergence. If the method holds, limited-data CT-style problems gain non-stationary GP reconstructions with explicit uncertainty, and related integral-measurement problems such as strain-field tomography become natural next targets.

What carries the argument

The load-bearing object is the finite Hilbert space basis expansion (9), in which a stationary covariance is written as a weighted sum of products of sine eigenfunctions on a rectangular domain, with weights given by the spectral density. Because this factorizes the kernel in its two inputs, each line-integral covariance entry becomes a product of single line integrals of sines (13), which is what reduces the numerical work from double to single integrals. The second piece of machinery is the pre-training objective (14), which fits the network warp to the mean of a standard GP reconstruction before joint training, providing an informed starting point.

What would settle it

On a trained model, compute the exact double-integral covariance entry (4) by numerical integration and compare it with the basis approximation (12) for many pairs of measurement lines, including lines crossing sharp edges; if the relative error becomes comparable to the noise level after training, the central feasibility claim fails.

Watch

Extended reading notes

Core claim

The paper claims that deep kernel learning—a Gaussian process with a neural-network-warped input space—is feasible for line-integral measurements if the covariance is replaced by a truncated Hilbert space basis expansion. This turns the covariance matrix between measurements into $L\approx \Phi\Lambda\Phi^T$, so the challenging double integrals reduce to single integrals and the computation scales linearly with the number of measurements. The paper further claims that pre-training the network warp against a standard GP reconstruction, before joint training of all parameters, substantially improves convergence and the final reconstruction. Demonstrated on one-dimensional step-function inference and on limited-data computed tomography, the method yields non-stationary GP reconstructions with tighter credibility regions than the standard stationary GP and cleaner region separation than filtered back projection.

Load-bearing premise

The finite sine-basis approximation, with a domain size chosen by a rule of thumb rather than tuned, stays accurate for the input locations the trained neural network actually produces.

Editorial extensions

If this is right

Deep kernel learning becomes computationally feasible for line-integral observations: the double-integral bottleneck is replaced by single integrals, and the per-iteration cost scales linearly instead of quadratically in the number of measurements.
Limited-data CT reconstructions can be produced with credibility regions, so uncertainty information is available where filtered back projection gives only a single image.
Pre-training the warp on a standard GP reconstruction makes joint training robust enough for two-dimensional problems; without it, convergence depends strongly on initialization.
The same line-integral machinery transfers to other settings where measurements are integrals of a field, such as strain-field estimation, which the paper identifies as a natural extension.

Reading between the lines

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

Because the factorization of the basis product is what does the work, the same linear-scaling reduction should extend to other linear functionals—weighted integrals, derivatives, boundary fluxes—provided they act separably on the sine basis; only the single-integral integrand changes.
The domain-size choice used for the basis is a heuristic without an error bound; a post-training check comparing the exact double-integral covariance with the basis approximation at the input locations the warp visits would test whether the approximation survives training.
Pre-training by regressing the warp onto a GP reconstruction is a generic warm-start for deep kernel learning and may help point-measurement problems, where poor local minima are also a known difficulty.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Short version: this paper has a real idea and a real computational contribution, but it undersells its own assumptions and over-relies on pictures. I think it deserves referee time, not because it is complete, but because the core mechanism is clean and the gaps are fixable.

What is actually new: the paper adapts deep kernel learning to line-integral measurements, and the trick of expanding the warped kernel in a Hilbert space basis (Eq. 9) to turn double integrals into single integrals (Eq. 13) is a genuine computational reduction. The derivation is sound, and the pre-training scheme (Eq. 14) is a sensible way to initialize the network when the cost landscape is bad. The method is described in enough detail to reimplement, and the related work section is competent.

Where the soft spots are: the truncated basis approximation is load-bearing but under-validated. The paper chooses the domain size Lk using the alpha=5 heuristic from the initial lengthscale, then fixes it while the network and hyperparameters are trained. The paper says the effect is "negligible" but offers no evidence. Since the neural network can push u(x) into regions where the fixed frequency grid no longer covers the spectral mass, the finite-basis approximation could silently become poor. No post-training check or error bound is reported. This is a real hole, not a fatal one: the approximation is standard practice, but with a warped input it deserves more care.

The empirical support is also weaker than the claims. The CT results are visual only: no quantitative reconstruction error, no comparison to a standard GP on the same CT data, no reporting of the number of basis functions, and no runtime or convergence details. The blurriness is acknowledged and blamed on the network architecture, but without ablation it is hard to know whether the method is doing what the paper says. The toy example does compare against a standard GP and looks better, but still visually.

One mild self-reference: pre-training the network against the standard GP reconstruction of the same data is a heuristic initialization, not a circular prediction. It is fine, but it should be described as such.

Overall, the math is honest and the contribution is real. This paper is for people working on limited-data CT, strain tomography, or any integral-observation GP problem who need non-stationary behavior with uncertainty. A serious referee could push the authors to validate the basis approximation after training, add quantitative metrics, and include a standard GP baseline in the CT experiments. I would recommend accepting it conditionally rather than desk-rejecting.

Referee Report

4 major / 4 minor

Summary. The paper proposes a method for deep kernel learning when observations are line integrals of a latent function, as in computed tomography. It approximates the stationary covariance kernel on the network-warped input with a truncated Hilbert-space basis expansion (Eq. 9), turning the double integrals in the Gram matrix (Eq. 4) into single integrals (Eq. 13) and matrix operations that scale linearly in the number of measurements for fixed basis size. It also introduces a pre-training step (Eq. 14) in which the neural network is fit to the standard GP posterior mean before joint training. Experiments are presented for a one-dimensional step function and for simulated and real CT data (Shepp-Logan phantom and carved cheese), with qualitative comparisons against filtered backprojection.

Significance. If the basis expansion remains accurate after training, the core reduction is a useful and nontrivial contribution: it is the first deep kernel learning formulation for integral observations and offers a route to non-stationary GP reconstruction with uncertainty in limited-data CT. The derivation of Eq. (13) is clean, the matrix-inversion-lemma reformulation is standard, and the authors are candid about limitations such as restricting to a single latent output and the heuristic choice of the domain size. However, the paper does not provide code, quantitative error metrics, uncertainty-calibration checks, or validation of the approximation error, so the significance is conditional on the additional evidence requested below.

major comments (4)

[Supplementary 8.1 and Eqs. (9)-(13)] The central reduction is valid only if the truncated basis expansion accurately represents the squared-exponential covariance on the range of u(x) after training, but this is asserted rather than verified. The domain size L_k is fixed before training using the alpha=5 heuristic with the initial lengthscale, while the network and hyperparameters are optimized; nothing prevents u(x) from approaching the Dirichlet boundary or the spectral mass from moving outside the frequency grid [pi/(2L_k), m-tilde*pi/(2L_k)]. The statement at the end of Section 8.1 that the effect of not optimizing L_k is "negligible" is unsupported. Please provide either an error bound for the approximation in Eq. (9) over the set of u values visited during training, or a post-training validation reporting max|u_k(x)|, the optimized lengthscales, and the resulting spectral coverage for each experiment. Without this, Eqs. (12)-(13) are not shown to approximate Eq. (4) for the model actually trained.
[Section 4] The empirical claims are supported only by visual inspection. For the Shepp-Logan phantom the ground truth is available, so the paper should report quantitative reconstruction errors (e.g., RMSE, SSIM, or normalized MSE) for the proposed method, FBP, and a standard GP baseline, with variability over noise realizations; for the real cheese data it should report a quantitative comparison against the dense reconstruction. In addition, on simulated data the paper should report the empirical coverage of the claimed 95% credibility regions, since the uncertainty estimates are one of the advertised outputs. Without these numbers the claims of improved performance over FBP and the standard GP cannot be assessed.
[Algorithm 1 and Eq. (14)] Pre-training is described as having a "crucial impact" on convergence and final quality, but no ablation is provided. Please report results with (i) random initialization, (ii) the proposed pre-training step, and ideally (iii) a fixed u(x) equal to the standard GP mean ft(x), to establish whether the gains come from the deep kernel learning itself or from initialization near the standard GP solution. This also addresses the mild circularity that the network is trained to reproduce the standard GP mean, which the authors themselves note may contain artifacts for discontinuous functions.
[Section 3.1 and Section 4] The headline scalability claim of linear scaling with the data size N is not demonstrated empirically. The derivation of Eq. (13) gives linear scaling for building Phi for fixed m, but m = m-tilde^{D_u} grows exponentially with the latent dimension, as acknowledged in Section 3.2, and no wall-clock times or scaling experiment with varying N are reported. Please include a short experiment measuring computation time per optimization step or per prediction as a function of N (and ideally D_u) to substantiate the claim that the method is practically feasible beyond one-dimensional problems.

minor comments (4)

[Section 4.2.1] In the paragraph after Figure 5, "FPB" should be "FBP" to match the abbreviation used elsewhere.
[References and Section 4.2.2] Reference [1] spells the society as "Finish Inverse Problems Society"; the correct spelling is "Finnish", and the dataset citation [1, 6] should give the full technical-report information in the text.
[Supplementary 8.1] The equation "alpha l_k^{-1} = max_k c_{kj}" uses k both as a dimension index and as a summation index in Eq. (11); please clarify the notation.
[Section 3.2] The sentence "As stated, this approach does not directly generalise to the case D_u > 1" is unclear, since Eq. (14) could in principle be applied to each output component separately; please clarify what is meant by non-generalization.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the predictive derivation is self-contained and no claim reduces to a fitted input or to a self-citation chain.

full rationale

The paper's central derivation is not circular. The Hilbert-space basis expansion in Eq. (9) uses the known spectral density of the squared-exponential covariance, and the reduction of the double integrals in Eq. (4) to single integrals in Eq. (13) is a direct algebraic consequence of the product structure of the basis functions. The predictive expressions in Eqs. (7) and (12) follow from standard GP conditioning and the matrix inversion lemma, with no fitted parameter renamed as a prediction. The pre-training step in Eq. (14) is a warm-start initialization: the neural network is fitted to the standard GP reconstruction ft of the same data, but the final predictor in Eq. (7) is not defined to equal ft, and the paper reports that the proposed method improves over the standard GP in the toy experiment, so the result is not forced by construction. The heuristic choice of the Dirichlet domain size Lk in Supplementary 8.1 is an approximation-accuracy assumption rather than a circular input: it is not a fitted parameter used to define the prediction. Citations to the authors' own prior work appear only as background or as related applications, and the load-bearing basis expansion cites external work [47]. Overall, the derivation is self-contained, and any concern about the uncontrolled approximation in Eq. (9) is a correctness or validation risk, not circularity.

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

The method rests on the quality of the truncated basis expansion (Eq. 9), on the unvalidated claim that Dirichlet boundaries can be made harmless by choosing Lk, and on a pre-training heuristic. The free parameters are mostly architectural choices and the domain-size multiplier; the GP hyperparameters are fit by marginal likelihood or LOO-CV in a standard way and are not listed here.

free parameters (4)

alpha (domain-size multiplier) = 5
Chosen by hand in supplementary 8.1 so that the basis domain covers 99.9% of the spectral mass of the squared exponential kernel. This value directly sets the domain size Lk and thus the approximation quality of the Hilbert space expansion; it is not optimized and its effect on the final result is not studied.
Number of basis functions per dimension (m_tilde)
The total number of basis functions m = m_tilde^Du is not stated in the paper. It controls both approximation accuracy and computational cost, and the paper gives no guidance or sensitivity analysis for this critical choice.
Neural network architecture = (2, 30, 20, 6, 1) for CT experiments
The network depth and width are chosen by hand. The paper explicitly says this structure is 'far from optimal', yet the reported results depend on it.
Number of pre-training points N_t = 10^4 for CT, 100 for 1D
Chosen by hand for the pre-training regression in Eq. (14). The sensitivity of the final reconstruction to N_t is not analyzed.

assumptions (4)

standard math A Gaussian process is closed under linear functionals, so line integrals of a GP are jointly Gaussian (Eq. 2).
This is a standard result from GP theory (Rasmussen and Williams, Sect. 2), used throughout Section 2.1.
domain assumption A stationary covariance function can be approximated by a finite sum of Laplace eigenfunctions (Eq. 9) with Dirichlet boundary conditions on a bounded domain.
This is the Solin-Sarkka reduced-rank approximation. The approximation error is ignored in the derivation; no bound is provided for the warped kernel after neural network training.
domain assumption Choosing Lk large enough so that the Dirichlet boundary conditions 'do not affect the GP solution' is sufficient for accurate predictions.
Stated in supplementary 8.1. This is a practical heuristic, not a theorem, and it is load-bearing because the entire method relies on the accuracy of the approximate covariance L approximately Phi Lambda Phi^T.
ad hoc to paper Pre-training the neural network to match the standard GP mean reconstruction (Eq. 14) provides an initialization from which joint training converges to a better solution.
This is a heuristic claim. The paper observes it works in examples, but gives no convergence guarantees or analysis of when it might fail. It is a central part of Algorithm 1.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Deep kernel learning for integral measurements." pith.science (2026). https://pith.science/paper/O65MXHRH

@misc{pith2026190901844,
  author       = {Pith},
  title        = {Pith review of: Deep kernel learning for integral measurements},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O65MXHRH}},
  note         = {Machine review of arXiv:1909.01844}
}

read the original abstract

Deep kernel learning refers to a Gaussian process that incorporates neural networks to improve the modelling of complex functions. We present a method that makes this approach feasible for problems where the data consists of line integral measurements of the target function. The performance is illustrated on computed tomography reconstruction examples.

Figures

Figures reproduced from arXiv: 1909.01844 by the authors.

**Figure 2.** Illustration of deep kernel learning and two different constructions for the latent mapping [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** One-dimensional toy example of inferring a step function from 50 integral measurements over randomly chosen intervals in [0, 1]. True function in solid grey, standard GP in blue (dashed) and the proposed method in red (dash-dotted). The shades indicate the 95% credibility regions, which is much tighter for the proposed method [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 5.** Figure 5: Left: Shepp-Logan phantom. Middle: FBP reconstruction. Right: proposed method. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 53 canonical work pages

[1]

Accessed: 2019-04-20

Finish inverse problems society.https://www.fips.fi/dataset.php. Accessed: 2019-04-20

work page 2019
[2]

Solving ill-posed inverse problems using iterative deep neural networks

Jonas Adler and Ozan Öktem. Solving ill-posed inverse problems using iterative deep neural networks. Inverse Problems, 33(12):124007, 2017

2017
[3]

Learned primal-dual reconstruction.IEEE Transactions on Medical Imaging, 37(6):1322–1332, 2018

Jonas Adler and Ozan Öktem. Learned primal-dual reconstruction.IEEE Transactions on Medical Imaging, 37(6):1322–1332, 2018

2018
[4]

Wilson, Yunus Saatchi, Zhiting Hu, and Eric P

Maruan Al-Shedivat, Andrew G. Wilson, Yunus Saatchi, Zhiting Hu, and Eric P. Xing. Learning scalable deep kernels with recurrent structure.Journal of Machine Learning Research, 18(82):1–37, 2017

work page 2017
[5]

Bouman and Ken Sauer

Charles A. Bouman and Ken Sauer. A uniﬁed approach to statistical tomography using coordinate descent optimization. IEEE Transactions on Image Processing, 5(3):480–492, 1996

work page 1996
[6]

Bubba, Markus Juvonen, Jonatan Lehtonen, Maximilian März, Alexander Meaney, Zenith Pur- isha, and Samuli Siltanen

Tatiana A. Bubba, Markus Juvonen, Jonatan Lehtonen, Maximilian März, Alexander Meaney, Zenith Pur- isha, and Samuli Siltanen. Tomographic X-ray data of carved cheese. Technical report, arXiv:1705.05732v2, June 2017

arXiv 2017
[7]

Rasmussen, and Marc P

Roberto Calandra, Jan Peters, Carl E. Rasmussen, and Marc P. Deisenroth. Manifold Gaussian processes for regression. InIn Proceedings of the International Joint Conference on Neural Networks (IJCNN), pages 3338–3345, 2016

work page 2016
[8]

Chapra.Applied Numerical Methods with MATLAB for engineers and scientists, 4th edition

Steven C. Chapra.Applied Numerical Methods with MATLAB for engineers and scientists, 4th edition. McGraw Hill, 2017

work page 2017

Show all 57 references

[9]

Allan M. Cormack. Representation of a function by its line integrals, with some radiological applications. Journal of Applied physics, 34(9):2722–2727, 1963

1963
[10]

Lawrence

Zhenwen Dai, Andreas Damianou, Javier Gonzalez, and Neil D. Lawrence. Variationally auto-encoded deep Gaussian processes. InProceedings of the International Conference on Learning Representations (ICLR), volume 3, 2016

2016
[11]

Damianou.Deep Gaussian processes and variational propagation of uncertainty

Andreas C. Damianou.Deep Gaussian processes and variational propagation of uncertainty. PhD thesis, Department of neuroscience, University of Sheﬃeld, 2015

2015
[12]

Damianou and Neil D

Andreas C. Damianou and Neil D. Lawrence. Deep Gaussian processes. InInternational Conference on Artiﬁcial Intelligence and Statistics (AISTATS), 2013. 9

2013
[13]

Alexander G. de G. Matthews, Jiri Hron, Mark Rowland, Richard E. Turner, and Zoubin Ghahramani. Gaussian process behaviour in wide deep neural networks. InInternational Conference on Learning Representations (ICLR), 2018

2018
[14]

Lecture 11: Bayesian quadrature

Roman Garnett. Lecture 11: Bayesian quadrature. University Lecture, 2018. Accessed: 2019-05-01. Avail- able from: http://www.cse.wustl.edu/~garnett/cse515t/spring_2018/files/lecture_notes/11. pdf

2018
[15]

Rasmussen, and Laurence Aitchison

Adri`a Garriga-Alonso, Carl E. Rasmussen, and Laurence Aitchison. Deep convolutional networks as shallow Gaussian processes. InInternational Conference on Learning Representations (ICLR), 2019

2019
[16]

Ghassemi and Marc P

Nooshin H. Ghassemi and Marc P. Deisenroth. Analytic long-term forecasting with periodic Gaussian pro- cesses. InProceedings of the International Conference on Artiﬁcial Intelligence and Statistics (AISTATS), 2014

2014
[17]

Golub and Charles F

Gene H. Golub and Charles F. Van Loan.Matrix Computations. Johns Hopkins University Press; fourth edition edition, 2013

2013
[18]

Shape recovery for sparse-data tomography.Mathematical Methods in the Applied Sciences, 40(18):6649–6669, 2017

Heikki Haario, Aki Kallonen, Marko Laine, Esa Niemi, Zenith Purisha, and Samuli Siltanen. Shape recovery for sparse-data tomography.Mathematical Methods in the Applied Sciences, 40(18):6649–6669, 2017

2017
[19]

A deep learning architecture for limited-angle computed tomography reconstruction

Kerstin Hammernik, Tobias Würﬂ, Thomas Pock, and Andreas Maier. A deep learning architecture for limited-angle computed tomography reconstruction. InBildverarbeitung für die Medizin 2017, pages 92–97. Springer Berlin Heidelberg, 2017

2017
[20]

Hendriks, Alexander W

Johannes N. Hendriks, Alexander W. T. Gregg, Christopher M. Wensrich, Anton S. Tremsin, Takenao Shinohara, Michael Meylan, Erich H. Kisi, Vladimir Luzin, and Oliver Kirsten. Bragg-edge elastic strain tomography for in situ systems from energy-resolved neutron transmission imag...

2017
[21]

Quasi-Newton methods: A new direction.The Journal of Machine Learning Research, 14(1):843–865, 2013

Philipp Hennig and Martin Kiefel. Quasi-Newton methods: A new direction.The Journal of Machine Learning Research, 14(1):843–865, 2013

2013
[22]

Gabor T. Herman. Image reconstruction from projections.Topics in Applied Physics, 32, 1979

1979
[23]

David J. C. MacKay. Introduction to Gaussian processes.NATO ASI Series F Computer and Systems Sciences, 168:133–166, 1998

1998
[24]

Schön, Christopher Wensrich, and Adrian Wills

Carl Jidling, Johannes Hendriks, Niklas Wahlström, Alexander Gregg, Thomas B. Schön, Christopher Wensrich, and Adrian Wills. Probabilistic modelling and reconstruction of strain.Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and...

2018
[25]

Carl Jidling, Niklas Wahlström, Adrian Wills, and Thomas B. Schön. Linearly constrained Gaussian processes. In Advances in Neural Information Processing Systems (NIPS), pages 1215–1224. 2017

2017
[26]

Springer Science & Business Media, 2006

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

2006
[27]

Deep neural networks as Gaussian processes

Jaehoon Lee, Jascha Sohl-dickstein, Jeﬀrey Pennington, Roman Novak, Sam Schoenholz, and Yasaman Bahri. Deep neural networks as Gaussian processes. InInternational Conference on Learning Represen- tations (ICLR), 2018

2018
[28]

William R. B. Lionheart and Philip J. Withers. Diﬀraction tomography of strain.Inverse Problems, 31(4):045005, 2015

2015
[29]

Neal.Bayesian Learning for Neural Networks, volume 118 ofLecture Notes in Statistics

Radford M. Neal.Bayesian Learning for Neural Networks, volume 118 ofLecture Notes in Statistics. Springer New York, 1996

1996
[30]

Unnikrishna Pillai.Probability, random variables, and stochastic processes

Athanasios Papoulis and S. Unnikrishna Pillai.Probability, random variables, and stochastic processes. McGraw-Hill Education, New York, 1991. 10

1991
[31]

Automatic diﬀerentiation in PyTorch

Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic diﬀerentiation in PyTorch. InThe future of gradient-based machine learning software and techniques – workshop at the Neura...

2017
[32]

Pelt, Kees Joost Batenburg, and James A

Daniël M. Pelt, Kees Joost Batenburg, and James A. Sethian. Improving tomographic reconstruction from limited data using mixed-scale dense convolutional neural networks.Journal of Imaging, 4(11), 2018

2018
[33]

Zenith Purisha, Carl Jidling, Niklas Wahlström, Simo Särkkä, and Thomas B. Schön. Probabilistic approach to limited-data computed tomography reconstruction. Technical report, arXiv:1809.03779v2, 2018

2018 arXiv
[34]

Carl. E. Rasmussen and Christopher K. I. Williams.Gaussian processes for machine learning. MIT press, Cambridge, MA, 2006

2006
[35]

Doubly stochastic variational inference for deep Gaussian processes

Hugh Salimbeni and Marc Deisenroth. Doubly stochastic variational inference for deep Gaussian processes. In Advances in Neural Information Processing Systems (NIPS), pages 4588–4599. 2017

2017
[36]

Sampson and Peter Guttorp

Paul D. Sampson and Peter Guttorp. Nonparametric estimation of nonstationary spatial covariance structure. Journal of the American Statistical Association, 87(417):108–119, 1992

1992
[37]

Santisteban, Lyndon Edwards, Mike E

Javier R. Santisteban, Lyndon Edwards, Mike E. Fitzpatrick, Axel Steuwer, Philip J. Withers, M.R Daymond, Michael W. Johnson, Nigel Rhodes, and Erik M. Schooneveld. Strain imaging by Bragg edge neutron transmission. Nuclear Instruments and Methods in Physics Research Section A...

2002
[38]

Santisteban, Lyndon Edwards, Mike E

Javier R. Santisteban, Lyndon Edwards, Mike E. Fizpatrick, Axel Steuwer, and Philip J. Withers. Engineering applications of Bragg-edge neutron transmission.Applied Physics, 74(1):1433–1436, 2002

2002
[39]

Bayesian estimation of 3-D objects from few radiographs.IEEE Transactions on Nuclear Science, 41(5):1780–1790, 1994

Ken Sauer, J Sachs, and C Klifa. Bayesian estimation of 3-D objects from few radiographs.IEEE Transactions on Nuclear Science, 41(5):1780–1790, 1994

1994
[40]

Adams, and Nando De Freitas

Bobak Shahriari, Kevin Swersky, Ziyu Wang, Ryan P. Adams, and Nando De Freitas. Taking the human out of the loop: A review of bayesian optimization.Proceedings of the IEEE, 104(1):148–175, 2016

2016
[41]

Shepp and Joseph B

Lawrence A. Shepp and Joseph B. Kruskal. Computerized tomography: the new medical X-ray technology. American Mathematical Monthly, pages 420–439, 1978

1978
[42]

Shepp and Benjamin F

Lawrence A. Shepp and Benjamin F. Logan. The fourier reconstruction of a head section. IEEE Transactions on Nuclear Science, 21:21–43, 1974

1974
[43]

PyTorch-LBFGS: A PyTorch implementation of L-BFGS

Hao-Jun Michael Shi and Dheevatsa Mudiger. PyTorch-LBFGS: A PyTorch implementation of L-BFGS. https://github.com/hjmshi/PyTorch-LBFGS, 2017

2017
[44]

Variable noise and dimensionality reduction for sparse Gaussian processes

Edward Snelson and Zoubin Ghahramani. Variable noise and dimensionality reduction for sparse Gaussian processes. In Proceedings of Uncertainty in Artiﬁcial Intelligence (UAI), 2006

2006
[45]

Rasmussen

Edward Snelson, Zoubin Ghahramani, and Carl E. Rasmussen. Warped Gaussian processes. InAdvances in Neural Information Processing Systems (NIPS), pages 337–344. 2004

2004
[46]

Know your boundaries: Constraining Gaussian processes by variational harmonic features

Arno Solin and Manon Kok. Know your boundaries: Constraining Gaussian processes by variational harmonic features. InProceedings of Machine Learning Research, volume 89, pages 2193–2202, 2019

2019
[47]

Hilbert space methods for reduced-rank Gaussian process regression

Arno Solin and Simo Särkkä. Hilbert space methods for reduced-rank Gaussian process regression. Technical report, arXiv:1401.5508, January 2014

2014 arXiv
[48]

Stein.Interpolation of Spatial Data

Michael L. Stein.Interpolation of Spatial Data. Springer-Verlag New York, 02 1999

1999
[49]

Schönberger, Juan Nunez-Iglesias, François Boulogne, Joshua D

Stéfan van der Walt, Johannes L. Schönberger, Juan Nunez-Iglesias, François Boulogne, Joshua D. Warner, Neil Yager, Emmanuelle Gouillart, Tony Yu, and the scikit-image contributors. scikit-image: image processing in Python.PeerJ, 2:e453, 2014. 11

2014
[50]

PhD thesis, Humboldt-Universität zu Berlin, Mathematisch- Naturwissenschaftliche Fakultät II, 2012

Sebastian Walter.Structured higher-order algorithmic diﬀerentiation in the forward and reverse mode with application in optimum experimental design. PhD thesis, Humboldt-Universität zu Berlin, Mathematisch- Naturwissenschaftliche Fakultät II, 2012

2012
[51]

Wensrich, Johannes N

Christopher M. Wensrich, Johannes N. Hendriks, Alexander Gregg, Michael H. Meylan, Vladimir Luzin, and Anton S. Tremsin. Bragg-edge neutron transmission strain tomography for in situ loadings.Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with...

2016
[52]

Wills and Thomas B

Adrian G. Wills and Thomas B. Schön. On the construction of probabilistic Newton-type algorithms. In IEEE 56th Annual Conference on Decision and Control (CDC), pages 6499–6504, Melbourne, Australia, Dec 2017

2017
[53]

Wilson and Ryan Adams

Andrew G. Wilson and Ryan Adams. Gaussian process kernels for pattern discovery and extrapolation. In International Conference on Machine Learning (ICML), pages 1067–1075, 2013

2013
[54]

Wilson, Zhiting Hu, Ruslan Salakhutdinov, and Eric P

Andrew G. Wilson, Zhiting Hu, Ruslan Salakhutdinov, and Eric P. Xing. Deep kernel learning. In Proceedings of the 19th International Conference on Artiﬁcial Intelligence and Statistics (AISTATS), pages 370–378, Cadiz, Spain, May 2016

2016
[55]

Wilson, Zhiting Hu, Ruslan R

Andrew G. Wilson, Zhiting Hu, Ruslan R. Salakhutdinov, and Eric P. Xing. Stochastic variational deep kernel learning. InAdvances in Neural Information Processing Systems (NIPS), pages 2586–2594, 2016

2016
[56]

Wilson, and Peter Frazier

Jian Wu, Matthias Poloczek, Andrew G. Wilson, and Peter Frazier. Bayesian optimization with gradients. In Advances in Neural Information Processing Systems (NIPS), pages 5267–5278. 2017

2017
[57]

Ghesu, Vincent Christlein, and Andreas Maier

Tobias Würﬂ, Florin C. Ghesu, Vincent Christlein, and Andreas Maier. Deep learning computed tomography. InMedical Image Computing and Computer-Assisted Intervention (MICCAI), pages 432– 440, 2016. 12

2016

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Accessed: 2019-04-20

Finish inverse problems society.https://www.fips.fi/dataset.php. Accessed: 2019-04-20

work page 2019

[2] [2]

Solving ill-posed inverse problems using iterative deep neural networks

Jonas Adler and Ozan Öktem. Solving ill-posed inverse problems using iterative deep neural networks. Inverse Problems, 33(12):124007, 2017

2017

[3] [3]

Learned primal-dual reconstruction.IEEE Transactions on Medical Imaging, 37(6):1322–1332, 2018

Jonas Adler and Ozan Öktem. Learned primal-dual reconstruction.IEEE Transactions on Medical Imaging, 37(6):1322–1332, 2018

2018

[4] [4]

Wilson, Yunus Saatchi, Zhiting Hu, and Eric P

Maruan Al-Shedivat, Andrew G. Wilson, Yunus Saatchi, Zhiting Hu, and Eric P. Xing. Learning scalable deep kernels with recurrent structure.Journal of Machine Learning Research, 18(82):1–37, 2017

work page 2017

[5] [5]

Bouman and Ken Sauer

Charles A. Bouman and Ken Sauer. A uniﬁed approach to statistical tomography using coordinate descent optimization. IEEE Transactions on Image Processing, 5(3):480–492, 1996

work page 1996

[6] [6]

Bubba, Markus Juvonen, Jonatan Lehtonen, Maximilian März, Alexander Meaney, Zenith Pur- isha, and Samuli Siltanen

Tatiana A. Bubba, Markus Juvonen, Jonatan Lehtonen, Maximilian März, Alexander Meaney, Zenith Pur- isha, and Samuli Siltanen. Tomographic X-ray data of carved cheese. Technical report, arXiv:1705.05732v2, June 2017

arXiv 2017

[7] [7]

Rasmussen, and Marc P

Roberto Calandra, Jan Peters, Carl E. Rasmussen, and Marc P. Deisenroth. Manifold Gaussian processes for regression. InIn Proceedings of the International Joint Conference on Neural Networks (IJCNN), pages 3338–3345, 2016

work page 2016

[8] [8]

Chapra.Applied Numerical Methods with MATLAB for engineers and scientists, 4th edition

Steven C. Chapra.Applied Numerical Methods with MATLAB for engineers and scientists, 4th edition. McGraw Hill, 2017

work page 2017