Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Reproducibility of machine learning analyses of 21 cm reionization maps

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A CNN trained on 21 cm reionization maps can score near-perfectly on slices from boxes it already saw while failing on fresh simulation boxes, meaning it has memorized the boxes rather than learned the physics.

desk verdict Valuable cautionary study showing CNN box-memorization in 21 cm inference, but the diagnosis is stronger than the evidence without an identifiability baseline. read the letter →

arxiv 2412.15893 v1 pith:NBEHTYSC submitted 2024-12-20 astro-ph.CO

Kimeel Sooknunan , Emma Chapman , Luke Conaboy , Daniel Mortlock , Jonathan Pritchard This is my paper · ORCID
classification astro-ph.CO
keywords 21cmcosmologyreionizationconvolutionalneuralnetworksparameterinferencegeneralizationout-of-distributionnetworkmemorizationbrightnesstemperaturemaps
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 reports a reproduction study of convolutional neural networks built to infer cosmological and astrophysical parameters from simulated 21 cm reionization maps. Its central claim is that these networks often achieve high scores not by learning the physics of reionization but by learning to recognise individual simulation boxes: slices drawn from the same box share an underlying density field, and the network can exploit that fingerprint. When tested on slices from boxes seen in training, a six-parameter network reaches an average R2 of about 0.994; when tested on slices from freshly simulated boxes, the R2 values collapse to near zero or negative, for example -1.276 for the reionization index Dion. The paper concludes that such a network is not useful as it stands for application to real observations, where the true density field cannot be matched by the training set. A reader should care because this exposes a failure mode that can make published ML-based inference look successful while being fragile.

What carries the argument

The central object is the CNN architecture taken from Hassan et al. (2020): four convolutional blocks (two convolutional layers, batch normalisation, ReLU) joined by max pooling, followed by three fully connected blocks, trained with mean-squared error and the Adam optimiser. The argument is carried by the R2 score measured under two testing protocols: slices from boxes represented in the training set versus slices from newly simulated boxes, and by a control scan in which the fraction P/120 of slices per box shared between training and testing is varied. That shared-slice fraction is the mechanism of memorization: when the network sees many slices from one box, it can identify the box through its shared density field and simply return the box's parameter labels, which produces inflated same-box R2 but fails at P=0, the regime relevant to real observations.

What would settle it

Take a fixed set of new simulation boxes and compare the CNN against a non-neural baseline that uses explicit summaries of the same slices, such as a power-spectrum and bispectrum measurement fed into a Gaussian process emulator. If the baseline also fails to recover the six parameters from single slices, the poor new-box R2 is a property of the inference problem rather than evidence of memorization; if the baseline succeeds, the CNN's failure is confirmed as a learning failure. A second decisive test is to train the CNN with exactly one slice per box and check whether new-box R2 stays high.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that CNNs trained on 21 cm brightness-temperature maps frequently learn box-level memorization rather than a transferable mapping from image features to parameters. The evidence chain is: reproducing the network architecture of Hassan et al. (2020) on 30,000 Simfast21 boxes; showing same-box test R2 of about 0.994 for six parameters; showing new-box R2 values of -1.276, -1.157, -0.236, -0.094, 0.433 and 0.511 across the six parameters; and demonstrating that R2 rises as the fraction of slices per box shared with the training set increases. The paper also finds that some targets generalise: the neutral fraction and the CMB optical depth remain recoverable on new boxes, while the reionization duration fails because the training set samples it poorly. The interpretation is that a network which appears to perform excellently can be silently exploiting artificial properties of the training set, placing responsibility on users to test generalization on completely new simulations before trusting CNN-based parameter inference.

Load-bearing premise

The load-bearing premise is that a single 2D slice of a 150 Mpc simulation box contains enough information to determine all six astrophysical and cosmological parameters; if that inference problem is ill-posed, a network could fail on new boxes even while learning real physics.

Editorial extensions

If this is right

  • A reported R2 above 0.99 on a held-out test set is not evidence of generalisation when training and test slices share simulation boxes; new-box performance must be reported separately.
  • Targets like the neutral fraction and the CMB optical depth appear learnable across new boxes, while parameters such as h, sigma8 and fesc do not, so claims about those parameters should be treated cautiously.
  • Providing the network with a cube of slices across redshift improves new-box recovery of some parameters, suggesting that temporal evolution carries much of the usable information.
  • A network trained and tested under the common shared-box protocol is not ready for real 21 cm observations, because real data correspond to P=0.
  • Users of CNN-based inference in 21 cm cosmology need to check the training-set simulation statistics and add out-of-distribution tests before drawing physical conclusions.

Reading between the lines

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

  • If the single-slice inference problem is fundamentally ill-posed, the same-box R2 could be inflated by correlations between the density field and the parameters even without memorization; a summary-statistic baseline on the same new boxes would separate those cases.
  • The redshift-cube improvement in Section 9 may be driven as much by adding extra constraints as by the network learning physics; ablating single redshift slices would show which slices carry the recoverable signal.
  • The same shared-box diagnostic could be applied to other image-based cosmological inference pipelines, since any training set that draws multiple views from a single realisation carries the same memorization risk.
  • A practical discipline suggested by the paper is to train with at most one slice per simulation realisation, or to generate enough independent boxes that shared density fields cannot be a shortcut.
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.

Referee Report

3 major / 6 minor

Summary. This paper presents an empirical benchmark of a convolutional neural network used for 21 cm reionization parameter inference. The authors reproduce the architecture of Hassan et al. (2020), train it on 30,000 Simfast21 simulations, and compare performance when test slices come from boxes already represented in training versus from freshly simulated boxes. They find that a six-parameter regression (Ωm, h, σ8, fesc, Cion, Dion) achieves an average R2 ≈ 0.994 on shared-box test slices but near-zero or negative R2 for four of the six parameters on new boxes. A controlled experiment (Sec. 6) varies the fraction P/120 of slices per box shared between training and test at fixed total training size, and shows a monotonic rise in average R2 with P/120. The paper also studies recovery of the neutral fraction, reionization midpoint, duration, and CMB optical depth, with and without redshift-stacked inputs. It concludes that the CNN often memorizes box-specific features rather than learning physical mappings, so that current architectures are not suitable for real 21 cm data.

Significance. The paper is a timely and valuable cautionary contribution to a field where ML-based inference from 21 cm maps is being adopted rapidly. Its central empirical findings are stark and well controlled: the near-perfect shared-box R2 and the collapse to near-zero/negative R2 on new simulation boxes, complemented by the P/120 trend that isolates the effect of shared-box information at fixed training-set size. The authors ship the code, which supports reproducibility. If the conclusions hold, the paper should motivate the community to adopt stricter train/test separation protocols (new simulation boxes, not just new slices) and to test the robustness of published CNN-based parameter inference. The main qualifications are that the evidence comes from a single architecture and from single training runs, and that the memorization diagnosis would be strengthened by an explicit identifiability argument or baseline estimator.

major comments (3)
  1. [Sec. 5.2 / Sec. 6, Figs. 7 and 8] The R2 collapse in Fig. 7 is interpreted in Sec. 5.2 as direct evidence that the network 'has instead learnt to associate slices with individual boxes.' This interpretation presumes that a well-trained network could in principle recover all six parameters from a single 2D slice of a 150 Mpc Simfast21 box. If the single-slice problem is not identifiable, then a perfect network would also fail on new boxes, and the high same-box R2 would reflect box-specific correlations rather than a failure to learn physics. The P/120 experiment in Fig. 8 provides controlled evidence for memorization, because same-box test performance rises with the fraction of training slices per box at fixed total training size. However, the paper does not explicitly explain why this trend rules out the identifiability alternative; a reader must supply that argument. Please add a brief identifiability discussion and, ideally, a simple baseline estimator (e.g., linear regression on power-spectrum or topology summary statistics, or a nearest-neighbour predictor) to quantify the new-box performance achievable without memorization.
  2. [Secs. 7–9, Table 1] The 3D case studies (Secs. 7–9) feed datacubes of shape (200,200,20) into the network, but Table 1 specifies an input shape of (200,200,1). The paper does not state whether the first convolutional layer was changed to accept 20 input channels, whether 3D convolutions were used, or how the redshift/line-of-sight dimension was handled. Because the paper's stated purpose is reproducibility, this architecture detail is essential and should be specified explicitly (or the table amended for the 3D runs).
  3. [General / Fig. 8] Every configuration in the paper is trained once, and Fig. 8 is reported without error bars. The main qualitative conclusions are probably robust given the size of the R2 differences, but the paper should at least state that single runs are used and provide some estimate of the expected variability (e.g., from a few random seeds for a subset of configurations, or from multiple data splits). Without this, a reader cannot assess whether the monotonic trend in Fig. 8 and the exact positions of the orange/red points are stable.
minor comments (6)
  1. [Abstract] The abstract states the paper applies networks to '21cmFast and Simfast21 simulations,' but the full text only uses Simfast21; please correct the abstract or clarify that 21cmFast refers to the original papers being reproduced.
  2. [Sec. 2.3 / Sec. 7.1] Sec. 2.3 says each simulation had 21 boxes, while Sec. 7.1 refers to 20 redshift boxes; the counting should be made consistent.
  3. [Sec. 2.3] The text mentions 'minor adjustments to the code' to vary Cion and Dion, but the exact modifications are not described. Since the code is available, a brief description or pointer would improve reproducibility.
  4. [Sec. 9 / Fig. 12] The claim that the network is 'able to infer four out of six parameters' is somewhat strong given the R2 values in Fig. 12 (e.g., σ8 R2 = 0.283, Dion R2 = 0.494); please temper the wording or define a threshold for 'able to infer.'
  5. [Abstract / Conclusions] The paper generalizes from a single CNN architecture (Network I of Hassan et al. 2020) to 'CNNs often learn...' in the abstract; consider qualifying the claim as 'the tested CNN' or 'at least one widely used architecture.'
  6. [Throughout] Minor typos include 'criterea' (Sec. 7.1), 'refered' (Sec. 3.1), and 'telecopes' (Sec. 2).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is established by external benchmarks on freshly simulated boxes, not by fitted parameters, self-citations, or definitional reductions.

full rationale

This is an empirical benchmarking study, and its central claim—that CNNs trained on 21 cm maps often memorize simulation boxes rather than the underlying reionization physics (Abstract; Sec. 10)—is established by comparing network outputs to ground truth on simulations that were held out from training. The same-box/new-box comparison in Sec. 5.2 (Figs. 6 and 7) is an external test relative to the trained model: the new boxes are freshly simulated with different initial seeds, and no network parameter is fitted to those new-box labels. The P/120 experiment in Sec. 6 (Fig. 8) is also an empirical measurement, not a fitted prediction: R2 is computed on held-out slices and is not determined by construction from P/120, even though the shared-box design deliberately makes memorization visible. The architecture and preprocessing follow Hassan et al. (2020) and Mangena et al. (2020), but those are external references and are not used to justify the new conclusion; the new result is the degradation on unseen boxes. The acknowledged limitations—for example, the tau label not being the true CMB optical depth (Sec. 8.1) and the poor sampling of Delta z (Sec. 7.2)—are internal caveats that reduce the strength of some sub-claims but are not circular. A legitimate scientific concern is that no identifiability analysis or baseline estimator is given to show that all six parameters can be recovered from a single 2D slice (Sec. 5.2), but this is a correctness or interpretability concern, not a circularity: the paper does not define the target in terms of the network output, nor does it rename a fit as a prediction. No self-citation chain is load-bearing.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper is an empirical robustness study; it introduces no new fitted physical constants or entities. The central claim rests on simulation realism, the learnability of the inference task, and the representativeness of one CNN architecture.

assumptions (5)
  • domain assumption Simfast21 semi-numerical simulations provide a valid and sufficiently diverse proxy for the 21 cm signal during reionization.
    All training, validation and test data are drawn from this code; the paper does not test on radiative-transfer simulations or real data. Invoked throughout Sec 2.2-2.3.
  • domain assumption A single 2D slice from a 150 Mpc box contains enough information to infer the six astrophysical and cosmological parameters in principle.
    The interpretation of the R2 drop in Sec 5.2 as memorization presumes the inference task is learnable; the paper gives no identifiability analysis.
  • domain assumption The chosen CNN architecture (Hassan et al. 2020 Network I) is representative of CNNs used for 21 cm parameter inference.
    The conclusion that CNNs "often" learn box features generalizes from a single architecture; the architecture is introduced in Sec 3.2.
  • domain assumption Differences in reionization morphology are independent of the random seed used to generate the initial density field, so "new boxes" are a fair generalization test.
    The paper assumes seeds only change the density field realization, not the physical parameter mapping; this is used in Sec 2.3 and Sec 4.3.
  • domain assumption The spin-temperature saturated regime approximation (Ts >> Tcmb) does not affect the robustness conclusions.
    Stated in Sec 2.1; the paper also sets peculiar velocities to zero, which could alter which features the network can use.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reproducibility of machine learning analyses of 21 cm reionization maps." pith.science (2026). https://pith.science/paper/NBEHTYSC

@misc{pith2026241215893,
  author       = {Pith},
  title        = {Pith review of: Reproducibility of machine learning analyses of 21 cm reionization maps},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NBEHTYSC}},
  note         = {Machine review of arXiv:2412.15893}
}
read the original abstract

Machine learning (ML) methods have become popular for parameter inference in cosmology, although their reliance on specific training data can cause difficulties when applied across different data sets. By reproducing and testing networks previously used in the field, and applied to 21cmFast and Simfast21 simulations, we show that convolutional neural networks (CNNs) often learn to identify features of individual simulation boxes rather than the underlying physics, limiting their applicability to real observations. We examine the prediction of the neutral fraction and astrophysical parameters from 21 cm maps and find that networks typically fail to generalise to unseen simulations. We explore a number of case studies to highlight factors that improve or degrade network performance. These results emphasise the responsibility on users to ensure ML models are applied correctly in 21 cm cosmology.

Figures

Figures reproduced from arXiv: 2412.15893 by the authors.

Figure 1
Figure 1. Brightness temperature fields for two example SIMFAST21 simulation boxes of size of 150 Mpc, going from redshift 𝑧 = 7.7 (left) to 𝑧 = 5.8 (right). The top row shows the evolution for a late model; the bottom row shows the evolution for an early model. is specified in [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Our network architecture, for more details on each layer see [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Histograms of the ionization fraction (𝑥Hii) distribution of our dataset at redshifts of 𝑧 = 9 and 𝑧 = 10. The dataset is very skewed towards higher ionization fractions (top panel); this is fixed by balancing the dataset (bottom panel) such that equal numbers of slices are drawn from each ioniza￾tion fraction bin. data from the same box in the different data sets not as a design choice, but in order to follow the l… view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Predicted ionized fraction against true ionization fraction for the case where testing is on slices from boxes represented in the training set (top panel) and on slices from newly simulated boxes (bottom panel). While 𝑥Hii is still recovered to the degree that 𝑅 2 is o…
Figure 6
Figure 6. Figure 6: Plot showing the results of training the network on the dataset described in section Sec. 5.1. Slices from each simulation box were sampled and put into the training set, then different, uncorrelated slices, were sampled and put into the testing set. These plots show h…
Figure 7
Figure 7. Figure 7: Plot showing results of training the network on the dataset described in section Sec. 5.1. Unlike in [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: The average 𝑅 2 score as a function of the fraction of slices in the training set. The orange point corresponds to the case of [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Plot showing results of training network on 3D SimFast21 data with 𝑧mid (left) or Δ𝑧 (right) labels. It can be seen that the network performs well for 𝑧mid with a 𝑅 2 ≃ 0.988 and significantly less well for Δ𝑧. of the 20 redshift boxes and combine them to provide a sin…
Figure 11
Figure 11. Figure 11: Left panel: The results of training network on 3D SimFast21 data with 𝜏 labels. Middle and right panels: Results of training network on 3D SimFast21 data with 𝜏 label. Each point is coloured according to the value of its 𝑧mid (middle) or Δ𝑧 (right) label. 0.20 0.25 0.…
Figure 12
Figure 12. Figure 12: The results of training the network on the new 3D dataset. ACKNOWLEDGEMENTS KS acknowledges funding by an Imperial College President’s Scholarship. EC acknowledges the support of a Royal Society Dorothy Hodgkin Fellowship and a Royal Society Enhancement Award. LC ackn…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Machine Learning and the SKA for Cosmic Dawn and the Epoch of Reionization

    astro-ph.IM 2026-07 accept novelty 2.5 of 10

    A multi-author overview of machine-learning algorithms proposed for instrument modelling, data analysis, simulation and inference in SKA Cosmic Dawn and Epoch of Reionization science.

Reference graph

Works this paper leans on

91 extracted references · 6 canonical work pages · cited by 1 Pith paper

  1. [1]

    Towards cosmological inference on unlabeled out-of-distribution HI observational data

    Andrianomena S., Hassan S., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2411.10515 , https://ui.adsabs.harvard.edu/abs/2024arXiv241110515A p. arXiv:2411.10515

    work page Pith review arXiv doi:10.48550/arxiv.2411.10515 2024

  2. [2]

    Atek H., et al., 2024, @doi [ ] 10.1038/s41586-024-07043-6 , https://ui.adsabs.harvard.edu/abs/2024Natur.626..975A 626, 975

    doi:10.1038/s41586-024-07043-6 2024

  3. [3]

    Di Matteo, P

    Baek, S. Di Matteo, P. Semelin, B. Combes, F. Revaz, Y. 2009, @doi [A&A] 10.1051/0004-6361:200810757 , 495, 389

    doi:10.1051/0004-6361:200810757 2009

  4. [4]

    D., Davies F

    Becker G. D., Davies F. B., Furlanetto S. R., Malkan M. A., Boera E., Douglass C., 2018, , 863, 92

    2018

  5. [5]

    Bevins H. T. J., Handley W. J., Fialkov A., de Lera Acedo E., Javid K., 2021, @doi [ ] 10.1093/mnras/stab2737 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.508.2923B 508, 2923

    doi:10.1093/mnras/stab2737 2021

  6. [6]

    K., Iliev I

    Bianco M., Giri S. K., Iliev I. T., Mellema G., 2021, @doi [ ] 10.1093/mnras/stab1518 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.505.3982B 505, 3982

    doi:10.1093/mnras/stab1518 2021

  7. [7]

    K., Prelogovi \'c D., Chen T., Mertens F

    Bianco M., Giri S. K., Prelogovi \'c D., Chen T., Mertens F. G., Tolley E., Mesinger A., Kneib J.-P., 2024, @doi [ ] 10.1093/mnras/stae257 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.528.5212B 528, 5212

    doi:10.1093/mnras/stae257 2024

  8. [8]

    S., La Plante P., Aguirre J

    Billings T. S., La Plante P., Aguirre J. E., 2021, PASP, https://ui.adsabs.harvard.edu/abs/2021arXiv210314563B 133, 044001

    2021

Show all 91 references
  1. [9]

    Bosman S. E. I., et al., 2022, @doi [ ] 10.1093/mnras/stac1046 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.514...55B 514, 55

    2022 doi

  2. [10]

    L., Green J

    Braun R., Bourke T. L., Green J. A., Keane E., Wagg J., 2015, in , Vol. 215, PoS(AASKA14). Giardini Naxos, Italy

    2015

  3. [11]

    G., Prelogovi \'c D., Qin Y., Trotta R., 2024, @doi [ ] 10.1093/mnras/stad3849 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.527.9833B 527, 9833

    Breitman D., Mesinger A., Murray S. G., Prelogovi \'c D., Qin Y., Trotta R., 2024, @doi [ ] 10.1093/mnras/stad3849 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.527.9833B 527, 9833

    2024 doi

  4. [12]

    H., Portillo S

    Bye C. H., Portillo S. K. N., Fialkov A., 2022, @doi [ ] 10.3847/1538-4357/ac6424 , https://ui.adsabs.harvard.edu/abs/2022ApJ...930...79B 930, 79

    2022 doi

  5. [13]

    arXiv:2405.18485

    Carniani S., et al., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2405.18485 , https://ui.adsabs.harvard.edu/abs/2024arXiv240518485C p. arXiv:2405.18485

    2024 doi arXiv

  6. [15]

    Choudhury M., Datta A., Chakraborty A., 2020, @doi [ ] 10.1093/mnras/stz3107 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.491.4031C 491, 4031

    2020 doi

  7. [16]

    R., 2021, @doi [ ] 10.1093/mnras/stab180 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.502.2815C 502, 2815

    Choudhury M., Chatterjee A., Datta A., Choudhury T. R., 2021, @doi [ ] 10.1093/mnras/stab180 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.502.2815C 502, 2815

    2021 doi

  8. [17]

    Choudhury M., Datta A., Majumdar S., 2022, @doi [ ] 10.1093/mnras/stac736 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.512.5010C 512, 5010

    2022 doi

  9. [18]

    L., Boser B., Denker J

    Cun Y. L., Boser B., Denker J. S., Howard R. E., Habbard W., Jackel L. D., Henderson D., 1990, Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, Chapt. Handwritten Digit Recognition with a Back-propagation Network, pp 396--404, http://dl.acm.org/citation.cfm?id=109230.109279

    1990

  10. [19]

    Curtis-Lake E., et al., 2023, @doi [Nature Astronomy] 10.1038/s41550-023-01918-w , https://ui.adsabs.harvard.edu/abs/2023NatAs...7..622C 7, 622

    2023 doi

  11. [20]

    R., et al., 2017, @doi [ ] 10.1088/1538-3873/129/974/045001 , https://ui.adsabs.harvard.edu/abs/2017PASP..129d5001D 129, 045001

    DeBoer D. R., et al., 2017, @doi [ ] 10.1088/1538-3873/129/974/045001 , https://ui.adsabs.harvard.edu/abs/2017PASP..129d5001D 129, 045001

    2017 doi

  12. [21]

    Doussot A., Eames E., Semelin B., 2019, @doi [ ] 10.1093/mnras/stz2429 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.490..371D 490, 371

    2019 doi

  13. [22]

    P., Whitler L., Topping M

    Endsley R., Stark D. P., Whitler L., Topping M. W., Chen Z., Plat A., Chisholm J., Charlot S., 2023, @doi [ ] 10.1093/mnras/stad1919 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.524.2312E 524, 2312

    2023 doi

  14. [23]

    Endsley R., et al., 2024, @doi [ ] 10.1093/mnras/stae1857 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.533.1111E 533, 1111

    2024 doi

  15. [24]

    Fan X., et al., 2006, , 131, 1203

    2006

  16. [25]

    B., 1958, @doi [Proc

    Field G. B., 1958, @doi [Proc. of the IRE] 10.1109/JRPROC.1958.286741 , https://ui.adsabs.harvard.edu/abs/1958PIRE...46..240F 46, 240

    1958

  17. [26]

    B., 1959, @doi [ ] 10.1086/146653 , https://ui.adsabs.harvard.edu/abs/1959ApJ...129..536F 129, 536

    Field G. B., 1959, @doi [ ] 10.1086/146653 , https://ui.adsabs.harvard.edu/abs/1959ApJ...129..536F 129, 536

    1959 doi

  18. [27]

    L., et al., 2023, @doi [ ] 10.3847/2041-8213/acade4 , https://ui.adsabs.harvard.edu/abs/2023ApJ...946L..13F 946, L13

    Finkelstein S. L., et al., 2023, @doi [ ] 10.3847/2041-8213/acade4 , https://ui.adsabs.harvard.edu/abs/2023ApJ...946L..13F 946, L13

    2023 doi

  19. [28]

    R., Zaldarriaga M., Hernquist L., 2004, @doi [ ] 10.1086/423025 , https://ui.adsabs.harvard.edu/abs/2004ApJ...613....1F 613, 1

    Furlanetto S. R., Zaldarriaga M., Hernquist L., 2004, @doi [ ] 10.1086/423025 , https://ui.adsabs.harvard.edu/abs/2004ApJ...613....1F 613, 1

    2004 doi

  20. [29]

    R., Oh S

    Furlanetto S. R., Oh S. P., Briggs F. H., 2006, @doi [ ] 10.1016/j.physrep.2006.08.002 , https://ui.adsabs.harvard.edu/abs/2006PhR...433..181F 433, 181

    2006 doi

  21. [30]

    R., Datta K

    Ghara R., Choudhury T. R., Datta K. K., 2015, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stu2512 , 447, 1806

    2015 doi

  22. [31]

    Gillet N., Mesinger A., Greig B., Liu A., Ucci G., 2019, @doi [ ] 10.1093/mnras/stz010 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.484..282G 484, 282

    2019 doi

  23. [32]

    W., Titterington M., eds, Proceedings of Machine Learning Research Vol

    Glorot X., Bengio Y., 2010, in Teh Y. W., Titterington M., eds, Proceedings of Machine Learning Research Vol. 9, Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics. PMLR, Chia Laguna Resort, Sardinia, Italy, pp 249--256, https://pr...

    2010

  24. [33]

    arXiv:2311.18007

    Gondhalekar Y., Hassan S., Saphra N., Andrianomena S., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2311.18007 , https://ui.adsabs.harvard.edu/abs/2023arXiv231118007G p. arXiv:2311.18007

    2023 doi arXiv

  25. [34]

    The MIT Press

    Goodfellow I., Bengio Y., Courville A., 2016, Deep Learning. The MIT Press

    2016

  26. [35]

    Greig B., Mesinger A., 2015, @doi [ ] 10.1093/mnras/stv571 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.449.4246G 449, 4246

    2015 doi

  27. [36]

    Greig B., Mesinger A., 2017a, @doi [ ] 10.1093/mnras/stw3026 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.465.4838G 465, 4838

    doi

  28. [37]

    Greig B., Mesinger A., 2017b, @doi [ ] 10.1093/mnras/stx2118 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.472.2651G 472, 2651

    doi

  29. [38]

    Harikane Y., et al., 2023, @doi [ ] 10.3847/1538-4365/acaaa9 , https://ui.adsabs.harvard.edu/abs/2023ApJS..265....5H 265, 5

    2023 doi

  30. [39]

    G., 2016, @doi [ ] 10.1093/mnras/stv3001 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.457.1550H 457, 1550

    Hassan S., Dav \'e R., Finlator K., Santos M. G., 2016, @doi [ ] 10.1093/mnras/stv3001 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.457.1550H 457, 1550

    2016 doi

  31. [40]

    Hassan S., Andrianomena S., Doughty C., 2020, @doi [ ] 10.1093/mnras/staa1151 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.494.5761H 494, 5761

    2020 doi

  32. [41]

    Y.-Y., et al., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2305.03042 , https://ui.adsabs.harvard.edu/abs/2023arXiv230503042H p

    Hsiao T. Y.-Y., et al., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2305.03042 , https://ui.adsabs.harvard.edu/abs/2023arXiv230503042H p. arXiv:2305.03042

    2023 doi arXiv

  33. [42]

    Hutter A., 2018, @doi [ ] 10.1093/mnras/sty683 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.477.1549H 477, 1549

    2018 doi

  34. [44]

    T., Mellema G., Ahn K., Shapiro P

    Iliev I. T., Mellema G., Ahn K., Shapiro P. R., Mao Y., Pen U.-L., 2014, @doi [ ] 10.1093/mnras/stt2497 , https://ui.adsabs.harvard.edu/abs/2014MNRAS.439..725I 439, 725

    2014 doi

  35. [45]

    Iliev I., Santos M., Mesinger A., Majumdar S., Mellema G., 2015, in Advancing Astrophysics with the Square Kilometre Array (AASKA14). p. 7 ( @eprint arXiv 1501.04213 )

    2015 arXiv

  36. [46]

    C., Weinberger L

    Keating L. C., Weinberger L. H., Kulkarni G., Haehnelt M. G., Chardin J., Aubert D., 2020, @doi [ ] 10.1093/mnras/stz3083 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.491.1736K 491, 1736

    2020 doi

  37. [47]

    R., Gaikwad P., 2016, @doi [ ] 10.1093/mnras/stw192 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.457.4051K 457, 4051

    Khaire V., Srianand R., Choudhury T. R., Gaikwad P., 2016, @doi [ ] 10.1093/mnras/stw192 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.457.4051K 457, 4051

    2016 doi

  38. [48]

    P., Ba J., 2014, @doi [arXiv e-prints] 10.48550/arXiv.1412.6980 , https://ui.adsabs.harvard.edu/abs/2014arXiv1412.6980K p

    Kingma D. P., Ba J., 2014, @doi [arXiv e-prints] 10.48550/arXiv.1412.6980 , https://ui.adsabs.harvard.edu/abs/2014arXiv1412.6980K p. arXiv:1412.6980

    2014 doi arXiv

  39. [49]

    C., Haehnelt M

    Kulkarni G., Keating L. C., Haehnelt M. G., Bosman S. E. I., Puchwein E., Chardin J., Aubert D., 2019, , 485, L24

    2019

  40. [50]

    E., Park I., 2020, @doi [Journal of Korean Physical Society] 10.3938/jkps.77.49 , https://ui.adsabs.harvard.edu/abs/2020JKPS...77...49K 77, 49

    Kwon Y., Hong S. E., Park I., 2020, @doi [Journal of Korean Physical Society] 10.3938/jkps.77.49 , https://ui.adsabs.harvard.edu/abs/2020JKPS...77...49K 77, 49

    2020 doi

  41. [51]

    La Plante P., Ntampaka M., 2019, @doi [ ] 10.3847/1538-4357/ab2983 , https://ui.adsabs.harvard.edu/abs/2019ApJ...880..110L 880, 110

    2019 doi

  42. [52]

    Convolutional Networks for Images, Speech, and Time Series, pp 255--258, http://dl.acm.org/citation.cfm?id=303568.303704

    LeCun Y., Bengio Y., 1998, MIT Press, Cambridge, MA, USA, Chapt. Convolutional Networks for Images, Speech, and Time Series, pp 255--258, http://dl.acm.org/citation.cfm?id=303568.303704

    1998

  43. [53]

    R., 2020, @doi [ ] 10.1088/1538-3873/ab5bfd , https://ui.adsabs.harvard.edu/abs/2020PASP..132f2001L 132, 062001

    Liu A., Shaw J. R., 2020, @doi [ ] 10.1088/1538-3873/ab5bfd , https://ui.adsabs.harvard.edu/abs/2020PASP..132f2001L 132, 062001

    2020 doi

  44. [54]

    J., 1997, @doi [ ] 10.1086/303549 , https://ui.adsabs.harvard.edu/abs/1997ApJ...475..429M 475, 429

    Madau P., Meiksin A., Rees M. J., 1997, @doi [ ] 10.1086/303549 , https://ui.adsabs.harvard.edu/abs/1997ApJ...475..429M 475, 429

    1997 doi

  45. [55]

    R., Mondal R., Watkinson C

    Majumdar S., Pritchard J. R., Mondal R., Watkinson C. A., Bharadwaj S., Mellema G., 2018, @doi [ ] 10.1093/mnras/sty535 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.476.4007M 476, 4007

    2018 doi

  46. [56]

    L., Lancaster L., Villaescusa-Navarro F., Melchior P., Ho S., Perreault-Levasseur L., Spergel D

    Makinen T. L., Lancaster L., Villaescusa-Navarro F., Melchior P., Ho S., Perreault-Levasseur L., Spergel D. N., 2021, @doi [ ] 10.1088/1475-7516/2021/04/081 , https://ui.adsabs.harvard.edu/abs/2021JCAP...04..081M 2021, 081

    2021 doi

  47. [57]

    G., 2020, @doi [ ] 10.1093/mnras/staa750 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.494..600M 494, 600

    Mangena T., Hassan S., Santos M. G., 2020, @doi [ ] 10.1093/mnras/staa750 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.494..600M 494, 600

    2020 doi

  48. [58]

    McQuinn M., Lidz A., Zahn O., Dutta S., Hernquist L., Zaldarriaga M., 2007, @doi [Monthly Notices of the Royal Astronomical Society] 10.1111/j.1365-2966.2007.11489.x , 377, 1043

    2007

  49. [59]

    T., Pen U.-L., Shapiro P

    Mellema G., Iliev I. T., Pen U.-L., Shapiro P. R., 2006, @doi [Monthly Notices of the Royal Astronomical Society] 10.1111/j.1365-2966.2006.10919.x , 372, 679

    2006

  50. [60]

    G., Bobin J., Carucci I

    Mertens F. G., Bobin J., Carucci I. P., 2024, @doi [ ] 10.1093/mnras/stad3430 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.527.3517M 527, 3517

    2024 doi

  51. [61]

    Mesinger A., Furlanetto S., 2007, @doi [ ] 10.1086/521806 , 669, 663

    2007 doi

  52. [62]

    Mesinger A., Greig B., Sobacchi E., 2016, @doi [ ] 10.1093/mnras/stw831 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.459.2342M 459, 2342

    2016 doi

  53. [63]

    G., Finlator K., 2019, @doi [ ] 10.1093/mnras/stz2171 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.489.5594M 489, 5594

    Molaro M., Dav \'e R., Hassan S., Santos M. G., Finlator K., 2019, @doi [ ] 10.1093/mnras/stz2171 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.489.5594M 489, 5594

    2019 doi

  54. [64]

    P., et al., 2022, @doi [ ] 10.3847/2041-8213/ac9b22 , https://ui.adsabs.harvard.edu/abs/2022ApJ...940L..14N 940, L14

    Naidu R. P., et al., 2022, @doi [ ] 10.3847/2041-8213/ac9b22 , https://ui.adsabs.harvard.edu/abs/2022ApJ...940L..14N 940, L14

    2022 doi

  55. [65]

    Neutsch S., Heneka C., Br \"u ggen M., 2022, @doi [ ] 10.1093/mnras/stac218 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.511.3446N 511, 3446

    2022 doi

  56. [66]

    Ni S., Li Y., Gao L.-Y., Zhang X., 2022, @doi [ ] 10.3847/1538-4357/ac7a34 , https://ui.adsabs.harvard.edu/abs/2022ApJ...934...83N 934, 83

    2022 doi

  57. [67]

    Planck Collaboration 2020, , 641, A6

    2020

  58. [68]

    C., et al., 2014, @doi [ ] 10.1088/0004-637X/782/2/66 , https://ui.adsabs.harvard.edu/abs/2014ApJ...782...66P 782, 66

    Pober J. C., et al., 2014, @doi [ ] 10.1088/0004-637X/782/2/66 , https://ui.adsabs.harvard.edu/abs/2014ApJ...782...66P 782, 66

    2014 doi

  59. [69]

    Prelogovi \'c D., Mesinger A., Murray S., Fiameni G., Gillet N., 2022, @doi [ ] 10.1093/mnras/stab3215 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.509.3852P 509, 3852

    2022 doi

  60. [70]

    R., Loeb A., 2012, @doi [Reports on Progress in Physics] 10.1088/0034-4885/75/8/086901 , https://ui.adsabs.harvard.edu/abs/2012RPPh...75h6901P 75, 086901

    Pritchard J. R., Loeb A., 2012, @doi [Reports on Progress in Physics] 10.1088/0034-4885/75/8/086901 , https://ui.adsabs.harvard.edu/abs/2012RPPh...75h6901P 75, 086901

    2012 doi

  61. [71]

    G., Kadota K., Asorey J., Park I., 2022, @doi [ ] 10.1088/1475-7516/2022/01/020 , https://ui.adsabs.harvard.edu/abs/2022JCAP...01..020S 2022, 020

    Sabiu C. G., Kadota K., Asorey J., Park I., 2022, @doi [ ] 10.1088/1475-7516/2022/01/020 , https://ui.adsabs.harvard.edu/abs/2022JCAP...01..020S 2022, 020

    2022 doi

  62. [72]

    B., Mishra-Sharma S., Youn T., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2407.21097 , https://ui.adsabs.harvard.edu/abs/2024arXiv240721097S p

    Sabti N., Reddy R., Mu \ n oz J. B., Mishra-Sharma S., Youn T., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2407.21097 , https://ui.adsabs.harvard.edu/abs/2024arXiv240721097S p. arXiv:2407.21097

    2024 doi arXiv

  63. [73]

    G., Amblard A., Pritchard J., Trac H., Cen R., Cooray A., 2008, @doi [ ] 10.1086/592487 , https://ui.adsabs.harvard.edu/abs/2008ApJ...689....1S 689, 1

    Santos M. G., Amblard A., Pritchard J., Trac H., Cen R., Cooray A., 2008, @doi [ ] 10.1086/592487 , https://ui.adsabs.harvard.edu/abs/2008ApJ...689....1S 689, 1

    2008 doi

  64. [74]

    Saxena A., et al., 2024, @doi [ ] 10.1051/0004-6361/202347132 , https://ui.adsabs.harvard.edu/abs/2024A&A...684A..84S 684, A84

    2024 doi

  65. [75]

    Schmidhuber J., 2014, preprint, https://ui.adsabs.harvard.edu/\#abs/2014arXiv1404.7828S (arXiv:1404.7828)

    2014 arXiv

  66. [76]

    J., Pritchard J

    Schmit C. J., Pritchard J. R., 2018, @doi [ ] 10.1093/mnras/stx3292 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.475.1213S 475, 1213

    2018 doi

  67. [77]

    Shimabukuro H., Semelin B., 2017, @doi [ ] 10.1093/mnras/stx734 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.468.3869S 468, 3869

    2017 doi

  68. [78]

    Shimabukuro H., Yoshiura S., Takahashi K., Yokoyama S., Ichiki K., 2016, @doi [ ] 10.1093/mnras/stw482 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.458.3003S 458, 3003

    2016 doi

  69. [79]

    Shimabukuro H., Mao Y., Tan J., 2022, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/ac4ca3 , https://ui.adsabs.harvard.edu/abs/2022RAA....22c5027S 22, 035027

    2022 doi

  70. [80]

    Shin M.-S., Trac H., Cen R., 2008, @doi [The Astrophysical Journal] 10.1086/588247 , 681, 756

    2008 doi

  71. [81]

    Sikder S., Barkana R., Reis I., Fialkov A., 2024, @doi [ ] 10.1093/mnras/stad3699 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.527.9977S 527, 9977

    2024 doi

  72. [82]

    Simonyan K., Zisserman A., 2014, preprint, https://ui.adsabs.harvard.edu/abs/2014arXiv1409.1556S (arXiv:1409.1556)

    2014 arXiv

  73. [83]

    M., et al., 2009, @doi [Monthly Notices of the Royal Astronomical Society] 10.1111/j.1365-2966.2008.14206.x , 393, 32

    Thomas R. M., et al., 2009, @doi [Monthly Notices of the Royal Astronomical Society] 10.1111/j.1365-2966.2008.14206.x , 393, 32

    2009

  74. [84]

    K., Majumdar S., Kamran M., Choudhury M., 2022, @doi [ ] 10.1088/1475-7516/2022/04/045 , https://ui.adsabs.harvard.edu/abs/2022JCAP...04..045T 2022, 045

    Tiwari H., Shaw A. K., Majumdar S., Kamran M., Choudhury M., 2022, @doi [ ] 10.1088/1475-7516/2022/04/045 , https://ui.adsabs.harvard.edu/abs/2022JCAP...04..045T 2022, 045

    2022 doi

  75. [85]

    M., et al., 2016, @doi [ ] 10.3847/0004-637X/818/2/139 , https://ui.adsabs.harvard.edu/abs/2016ApJ...818..139T 818, 139

    Trott C. M., et al., 2016, @doi [ ] 10.3847/0004-637X/818/2/139 , https://ui.adsabs.harvard.edu/abs/2016ApJ...818..139T 818, 139

    2016 doi

  76. [86]

    Umeda H., Ouchi M., Nakajima K., Harikane Y., Ono Y., Xu Y., Isobe Y., Zhang Y., 2024, @doi [ ] 10.3847/1538-4357/ad554e , https://ui.adsabs.harvard.edu/abs/2024ApJ...971..124U 971, 124

    2024 doi

  77. [87]

    Villanueva-Domingo P., Villaescusa-Navarro F., 2021, @doi [ ] 10.3847/1538-4357/abd245 , https://ui.adsabs.harvard.edu/abs/2021ApJ...907...44V 907, 44

    2021 doi

  78. [88]

    R., 2007, @doi [The Astrophysical Journal] 10.1086/509597 , 654, 12

    Zahn O., Lidz A., McQuinn M., Dutta S., Hernquist L., Zaldarriaga M., Furlanetto S. R., 2007, @doi [The Astrophysical Journal] 10.1086/509597 , 654, 12

    2007 doi

  79. [89]

    E., 2011, @doi [ ] 10.1111/j.1365-2966.2011.18439.x , https://ui.adsabs.harvard.edu/abs/2011MNRAS.414..727Z 414, 727

    Zahn O., Mesinger A., McQuinn M., Trac H., Cen R., Hernquist L. E., 2011, @doi [ ] 10.1111/j.1365-2966.2011.18439.x , https://ui.adsabs.harvard.edu/abs/2011MNRAS.414..727Z 414, 727

    2011

  80. [90]

    B., 1970, , https://ui.adsabs.harvard.edu/abs/1970A&A.....5...84Z 5, 84

    Zel'dovich Y. B., 1970, , https://ui.adsabs.harvard.edu/abs/1970A&A.....5...84Z 5, 84

    1970

  81. [91]

    D., 2022, @doi [ ] 10.3847/1538-4357/ac457d , https://ui.adsabs.harvard.edu/abs/2022ApJ...926..151Z 926, 151

    Zhao X., Mao Y., Cheng C., Wandelt B. D., 2022, @doi [ ] 10.3847/1538-4357/ac457d , https://ui.adsabs.harvard.edu/abs/2022ApJ...926..151Z 926, 151

    2022 doi

  82. [93]

    L., 2022, @doi [Publications of the Astronomical Society of the Pacific] 10.1088/1538-3873/ac5f5d , 134, 044001

    Zhou Y., Plante P. L., 2022, @doi [Publications of the Astronomical Society of the Pacific] 10.1088/1538-3873/ac5f5d , 134, 044001

    2022 doi

  83. [94]

    P., et al., 2013, @doi [ ] 10.1051/0004-6361/201220873 , https://ui.adsabs.harvard.edu/abs/2013A&A...556A...2V 556, A2

    van Haarlem M. P., et al., 2013, @doi [ ] 10.1051/0004-6361/201220873 , https://ui.adsabs.harvard.edu/abs/2013A&A...556A...2V 556, A2

    2013 doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.