pith. machine review for the scientific record. sign in

arxiv: 2605.10105 · v2 · submitted 2026-05-11 · 🌌 astro-ph.CO

Recognition: no theorem link

Application of Machine Learning to 21 cm Cosmology

Hayato Shimabukuro
Authors on Pith no claims yet

Pith reviewed 2026-05-13 03:15 UTC · model grok-4.3

classification 🌌 astro-ph.CO
keywords 21 cm cosmologymachine learningEpoch of Reionizationcosmic dawnSKA-Lowforeground contaminationuncertainty propagation
0
0 comments X

The pith

Machine learning extracts reliable information from 21 cm data when it preserves physical structure and tracks uncertainties explicitly.

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

This review examines applications of machine learning to 21 cm observations of the cosmic dawn and Epoch of Reionization. It summarizes the non-Gaussian character of the signal and its sensitivity to early galaxy properties, then details major analysis bottlenecks from foreground contamination, radio-frequency interference, ionospheric distortions, calibration errors, and the cost of repeated forward modeling. The ML literature is organized into observation-domain methods on contaminated data, theory-domain methods that accelerate modeling, and inference-domain methods that map observables to parameters. The central argument is that ML delivers the greatest value by retaining physically relevant structures and propagating uncertainties explicitly rather than serving as an opaque substitute for physics-based forward models.

Core claim

Machine learning can address bottlenecks in 21 cm cosmology observations, but its greatest utility occurs when methods preserve physically relevant structure in the data and explicitly propagate uncertainties, instead of acting as opaque replacements for the underlying forward models.

What carries the argument

The classification of ML approaches into observation-domain methods operating on contaminated measurements, theory-domain methods accelerating forward modeling, and inference-domain methods connecting observables to constraints, with emphasis on retaining physical structure and uncertainty information.

If this is right

  • Observation-domain ML can reduce the impact of bright foreground contamination and radio-frequency interference on SKA-Low measurements.
  • Theory-domain ML can lower the computational cost of exploring high-dimensional parameter spaces describing early galaxies and radiation backgrounds.
  • Inference-domain ML can map non-Gaussian observables to astrophysical and cosmological constraints while retaining physical fidelity.
  • Hybrid approaches that combine ML with explicit physics models produce more robust extraction of information from the 21 cm signal.

Where Pith is reading between the lines

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

  • These principles could extend to joint analyses of 21 cm data with other probes such as the cosmic microwave background by providing consistent uncertainty estimates.
  • Validation on end-to-end simulations that include all listed systematics would directly test whether structure preservation reduces biases relative to standard pipelines.
  • The same structure-preserving requirements may apply to related radio intensity mapping experiments beyond the Epoch of Reionization.

Load-bearing premise

Machine learning methods can be engineered to retain physically relevant structure and propagate uncertainty without introducing new biases when handling observational systematics such as foregrounds, interference, ionosphere, and calibration errors.

What would settle it

A side-by-side test on identical simulated 21 cm datasets with known inputs, comparing parameter constraints from structure-preserving ML pipelines against those from fully opaque ML models and against the true simulation values.

Figures

Figures reproduced from arXiv: 2605.10105 by Hayato Shimabukuro.

Figure 1.1
Figure 1.1. Figure 1.1: Schematic timeline of cosmic history from the Big Bang to the present [PITH_FULL_IMAGE:figures/full_fig_p002_1_1.png] view at source ↗
Figure 1.2
Figure 1.2. Figure 1.2: Schematic redshift evolution of the sky-averaged 21 cm brightness tem [PITH_FULL_IMAGE:figures/full_fig_p004_1_2.png] view at source ↗
Figure 1.3
Figure 1.3. Figure 1.3: Comparison between theoretical predictions of the dimensionless 21 cm [PITH_FULL_IMAGE:figures/full_fig_p005_1_3.png] view at source ↗
Figure 1.4
Figure 1.4. Figure 1.4: Schematic overview of ML workflows in 21 cm cosmology. The arrows [PITH_FULL_IMAGE:figures/full_fig_p009_1_4.png] view at source ↗
Figure 1.5
Figure 1.5. Figure 1.5: Illustrative schematic of ML-based likelihood-free parameter inference for [PITH_FULL_IMAGE:figures/full_fig_p014_1_5.png] view at source ↗
read the original abstract

This chapter reviews how machine learning (ML) can be used to extract astrophysical and cosmological information from redshifted 21 cm observations of the cosmic dawn and the Epoch of Reionization, with an emphasis on SKA-Low science. We first summarize the basic physics of the global signal and spatial fluctuations, highlighting why the signal is intrinsically non-Gaussian and highly sensitive to poorly constrained properties of early galaxies and radiation backgrounds. We then discuss the main analysis bottlenecks that dominate current and future observations: bright foreground contamination, radio-frequency interference, ionospheric distortions, calibration errors, and the computational burden of repeated forward modeling in high-dimensional parameter spaces. Building on this context, we organize the ML literature by its role in the pipeline: observation-domain methods that operate on contaminated measurements and image products, theory-domain methods that accelerate or compress forward modeling, and inference-domain methods that map complex observables to astrophysical and cosmological constraints. The central message is that ML is most useful in 21 cm cosmology when it preserves physically relevant structure and propagates uncertainty explicitly, rather than acting as an opaque replacement for the underlying forward model.

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. This manuscript is a review chapter summarizing the physics of the redshifted 21 cm global signal and spatial fluctuations during cosmic dawn and the Epoch of Reionization, emphasizing their non-Gaussian character and sensitivity to early galaxy and radiation properties. It identifies dominant observational bottlenecks (foreground contamination, RFI, ionospheric distortions, calibration errors, and computational costs of high-dimensional forward modeling) and organizes the ML literature into observation-domain, theory-domain, and inference-domain applications, with a focus on SKA-Low science. The central prescriptive claim is that ML is most useful when it preserves physically relevant structure and propagates uncertainty explicitly rather than serving as an opaque replacement for the forward model.

Significance. If the synthesis and recommendations hold, the review offers a timely, field-specific guide for integrating ML into 21 cm analyses ahead of SKA-Low data. By framing ML as a complement to physical modeling rather than a substitute, it could reduce the risk of unquantified biases in a systematics-dominated domain and help prioritize methods that maintain interpretability and uncertainty calibration.

minor comments (3)
  1. [Abstract] Abstract: the opening sentence refers to 'This chapter'; for a journal submission, rephrase to 'This review' or explicitly note the book context to avoid reader confusion.
  2. [Bottlenecks discussion] Bottlenecks section: the high-level list of challenges is clear, but adding one or two quantitative references (e.g., dynamic range requirements or computational scaling) per bottleneck would better motivate why ML is needed.
  3. [ML literature organization] Literature organization: while the three-domain taxonomy is useful, a summary table listing representative papers, techniques, and their domain assignments would improve scannability and allow readers to locate specific methods quickly.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive and constructive assessment of our review manuscript. We appreciate the recognition that the work provides a timely guide for integrating ML into 21 cm analyses ahead of SKA-Low, and we are pleased that the central prescriptive claim—that ML is most useful when it preserves physical structure and propagates uncertainty—is viewed as a strength. The recommendation for minor revision is noted, and we will incorporate any editorial or minor clarifications in the revised version.

Circularity Check

0 steps flagged

No significant circularity: literature review with no derivations or self-referential claims

full rationale

This manuscript is explicitly a review chapter that organizes and summarizes existing literature on ML methods for 21 cm cosmology across observation, theory, and inference domains. It presents no original derivations, equations, fitted parameters, or novel predictions that could reduce to inputs by construction. The central prescriptive message—that ML is most useful when it preserves physical structure and propagates uncertainty—is a qualitative synthesis of prior work rather than a load-bearing claim justified by self-citation chains or ansatzes. No self-definitional steps, fitted-input predictions, or uniqueness theorems appear. The paper is therefore self-contained against external benchmarks with zero circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

As this is a review paper, no new free parameters, axioms, or invented entities are introduced by the authors.

pith-pipeline@v0.9.0 · 5484 in / 1079 out tokens · 119791 ms · 2026-05-13T03:15:19.511613+00:00 · methodology

Review history (2 revisions) →

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

101 extracted references · 101 canonical work pages

  1. [1]

    2024a, MNRAS, 527, 3, 7835

    Acharya, A., Mertens, F., Ciardi, B., et al. 2024a, MNRAS, 527, 3, 7835. doi:10.1093/mnras/stad3701

    work page doi:10.1093/mnras/stad3701

  2. [2]

    2024b, MNRAS, 534, 1, L30

    Acharya, A., Mertens, F., Ciardi, B., et al. 2024b, MNRAS, 534, 1, L30. doi:10.1093/mnrasl/slae078

    work page doi:10.1093/mnrasl/slae078

  3. [3]

    2017, Astronomy and Com- puting, 18, 35

    Akeret, J., Chang, C., Lucchi, A., & Refregier, A. 2017, Astronomy and Com- puting, 18, 35. doi:10.1016/j.ascom.2016.12.002

    work page doi:10.1016/j.ascom.2016.12.002 2017

  4. [4]

    2019, MNRAS, 488, 3,

    Alsing, J., Charnock, T., Feeney, S., & Wandelt, B. 2019, MNRAS, 488, 3,

    work page 2019

  5. [5]

    doi:10.1093/mnras/stz1960

    work page doi:10.1093/mnras/stz1960

  6. [6]

    K., Iliev, I

    Bianco, M., Giri, S. K., Iliev, I. T., et al. 2021, MNRAS, 505, 3, 3982. doi:10.1093/mnras/stab1518

    work page doi:10.1093/mnras/stab1518 2021

  7. [7]

    K., Mellema, G., et al

    Bianco, M., Giri, S. K., Mellema, G., et al. 2024, MNRAS, 528, 4, 5212. doi:10.1093/mnras/stae257

    work page doi:10.1093/mnras/stae257 2024

  8. [8]

    K., Iliev, I

    Bianco, M., Giri, S. K., Iliev, I. T., et al. 2025, MNRAS, 541, 1, 234. doi:10.1093/mnras/staf973

    work page doi:10.1093/mnras/staf973 2025

  9. [9]

    A., et al

    Brackenhoff, S. A., et al. 2024, MNRAS, 533, 1, 632. doi:10.1093/mnras/stae1856

    work page doi:10.1093/mnras/stae1856 2024

  10. [10]

    L., Gnedin, N

    Carilli, C. L., Gnedin, N. Y ., & Owen, F. 2002, ApJ, 577, 1, 22. doi:10.1086/342179

    work page doi:10.1086/342179 2002

  11. [11]

    & Jeli ´c, V

    Chapman, E. & Jeli ´c, V . 2019, arXiv e-prints, arXiv:1909.12369. doi:10.48550/arXiv.1909.12369

    work page doi:10.48550/arxiv.1909.12369 2019

  12. [12]

    2022, MNRAS, 512, 4, 5010

    Choudhury, M., Datta, A., & Majumdar, S. 2022, MNRAS, 512, 4, 5010. doi:10.1093/mnras/stac736

    work page doi:10.1093/mnras/stac736 2022

  13. [13]

    2020, Proceedings of the National Academy of Science, 117, 30055, doi: 10.1073/pnas.1912789117

    Cranmer, K., Brehmer, J., & Louppe, G. 2020, Proceedings of the National Academy of Sciences, 117, 48, 30055. doi:10.1073/pnas.1912789117

    work page doi:10.1073/pnas.1912789117 2020

  14. [14]

    D., & Carilli, C

    Datta, A., Bowman, J. D., & Carilli, C. L. 2010, ApJ, 724, 1, 526. doi:10.1088/0004-637X/724/1/526

    work page doi:10.1088/0004-637x/724/1/526 2010

  15. [15]

    W., Navarro J

    Diao K., Chen Z., Chen X., Mao Y ., 2024, ApJ, 974, 141. doi:10.3847/1538- 4357/ad6c40

    work page doi:10.3847/1538- 2024

  16. [16]

    A., et al

    de Gasperin, F., Mevius, M., Rafferty, D. A., et al. 2018, A&A, 615, A179. doi:10.1051/0004-6361/201833012

    work page doi:10.1051/0004-6361/201833012 2018

  17. [17]

    S., Mesinger, A., & Hewitt, J

    Ewall-Wice, A., Dillon, J. S., Mesinger, A., & Hewitt, J. 2014, MNRAS, 441, 3, 2476. doi:10.1093/mnras/stu666

    work page doi:10.1093/mnras/stu666 2014

  18. [18]

    Furlanetto, S. R. 2006, MNRAS, 370, 4, 1867. doi:10.1111/j.1365- 2966.2006.10603.x 1 Application of Machine Learning to 21 cm Cosmology 17

    work page doi:10.1111/j.1365- 2006

  19. [19]

    Furlanetto, S. R. & Loeb, A. 2002, ApJ, 579, 1, 1. doi:10.1086/342757

    work page doi:10.1086/342757 2002

  20. [20]

    R., Oh, S

    Furlanetto, S. R., Oh, S. P., & Briggs, F. H. 2006, Phys. Rep., 433, 4-6, 181. doi:10.1016/j.physrep.2006.08.002

    work page doi:10.1016/j.physrep.2006.08.002 2006

  21. [21]

    2021, MNRAS, 504, 4, 4716

    Gagnon-Hartman, S., Cui, Y ., Liu, A., & Ravanbakhsh, S. 2021, MNRAS, 504, 4, 4716. doi:10.1093/mnras/stab1158

    work page doi:10.1093/mnras/stab1158 2021

  22. [22]

    2019, MNRAS, 484, 1, 282

    Gillet, N., Mesinger, A., Greig, B., et al. 2019, MNRAS, 484, 1, 282. doi:10.1093/mnras/stz010

    work page doi:10.1093/mnras/stz010 2019

  23. [23]

    & Mesinger, A

    Greig, B. & Mesinger, A. 2015, MNRAS, 449, 4, 4246. doi:10.1093/mnras/stv571

    work page doi:10.1093/mnras/stv571 2015

  24. [24]

    & Mesinger, A

    Greig, B. & Mesinger, A. 2017, MNRAS, 472, 3, 2651. doi:10.1093/mnras/stx2118

    work page doi:10.1093/mnras/stx2118 2017

  25. [25]

    & Mesinger, A

    Greig, B. & Mesinger, A. 2018, MNRAS, 477, 3, 3217. doi:10.1093/mnras/sty796

    work page doi:10.1093/mnras/sty796 2018

  26. [26]

    2019, MNRAS, 483, 2, 2524

    Hassan, S., Liu, A., Kohn, S., et al. 2019, MNRAS, 483, 2, 2524. doi:10.1093/mnras/sty3282

    work page doi:10.1093/mnras/sty3282 2019

  27. [27]

    J., Trott, C

    Jennings, R. J., Trott, C. M., Gaensler, B. M., et al. 2019, MNRAS, 483, 2,

    work page 2019

  28. [28]

    doi:10.1093/mnras/sty3168

    work page doi:10.1093/mnras/sty3168

  29. [29]

    2024, MNRAS, 529, 4, 3684

    Kennedy, J., Gagnon-Hartman, S., Liu, A., & Ravanbakhsh, S. 2024, MNRAS, 529, 4, 3684. doi:10.1093/mnras/stae760

    work page doi:10.1093/mnras/stae760 2024

  30. [30]

    Kern, N. S. & Liu, A. 2021, MNRAS, 501, 1, 1463. doi:10.1093/mnras/staa3736

    work page doi:10.1093/mnras/staa3736 2021

  31. [31]

    S., Liu, A., Parsons, A

    Kern, N. S., Liu, A., Parsons, A. R., Mesinger, A., & Greig, B. 2017, ApJ, 848, 1, 23. doi:10.3847/1538-4357/aa8bb4

    work page doi:10.3847/1538-4357/aa8bb4 2017

  32. [32]

    R., et al

    Kerrigan, J. R., et al. 2019, MNRAS, 488, 2, 2605. doi:10.1093/mnras/stz1865

    work page doi:10.1093/mnras/stz1865 2019

  33. [33]

    D., Jacobs, D

    Kittiwisit, P., Bowman, J. D., Jacobs, D. C., et al. 2018, MNRAS, 474, 4, 4487. doi:10.1093/mnras/stx3099

    work page doi:10.1093/mnras/stx3099 2018

  34. [34]

    2016, PASJ, 68, 4, 61

    Kubota, K., Yoshiura, S., Shimabukuro, H., et al. 2016, PASJ, 68, 4, 61. doi:10.1093/pasj/psw059

    work page doi:10.1093/pasj/psw059 2016

  35. [35]

    & Shaw, J

    Liu, A. & Shaw, J. R. 2020, PASP, 132, 1012, 062001. doi:10.1088/1538- 3873/ab5bfd

    work page doi:10.1088/1538- 2020

  36. [36]

    R., & Trott, C

    Liu, A., Parsons, A. R., & Trott, C. M. 2014, Phys. Rev. D, 90, 2, 023018. doi:10.1103/PhysRevD.90.023018

    work page doi:10.1103/physrevd.90.023018 2014

  37. [37]

    R., & Trott, C

    Liu, A., Parsons, A. R., & Trott, C. M. 2014, Phys. Rev. D, 90, 2, 023019. doi:10.1103/PhysRevD.90.023019

    work page doi:10.1103/physrevd.90.023019 2014

  38. [38]

    Mack, K. J. & Wyithe, J. S. B. 2012, MNRAS, 425, 4, 2988. doi:10.1111/j.1365-2966.2012.21561.x

    work page doi:10.1111/j.1365-2966.2012.21561.x 2012

  39. [39]

    R., Mondal, R., et al

    Majumdar, S., Pritchard, J. R., Mondal, R., et al. 2018, MNRAS, 476, 3, 4007. doi:10.1093/mnras/sty535

    work page doi:10.1093/mnras/sty535 2018

  40. [40]

    R., et al

    Majumdar, S., Kamran, M., Pritchard, J. R., et al. 2020, MNRAS, 499, 4, 5090. doi:10.1093/mnras/staa3168

    work page doi:10.1093/mnras/staa3168 2020

  41. [41]

    2025, A&A, 698, A80

    M ´eriot, R., Semelin, B., & Cornu, D. 2025, A&A, 698, A80. doi:10.1051/0004-6361/202452901

    work page doi:10.1051/0004-6361/202452901 2025

  42. [42]

    G., Mevius, M., Koopmans, L

    Mertens, F. G., Mevius, M., Koopmans, L. V . E., et al. 2020, MNRAS, 493, 2,

    work page 2020

  43. [43]

    doi:10.1093/mnras/staa327 18 Hayato Shimabukuro

    work page doi:10.1093/mnras/staa327

  44. [44]

    G., Bobin, J., & Carucci, I

    Mertens, F. G., Bobin, J., & Carucci, I. P. 2024, MNRAS, 527, 2, 3517. doi:10.1093/mnras/stad3430

    work page doi:10.1093/mnras/stad3430 2024

  45. [45]

    , keywords =

    Mesinger, A., Furlanetto, S., & Cen, R. 2011, MNRAS, 411, 2, 955. doi:10.1111/j.1365-2966.2010.17731.x

    work page doi:10.1111/j.1365-2966.2010.17731.x 2011

  46. [46]

    2014, MNRAS, 439, 4, 3262

    Mesinger, A., Ewall-Wice, A., & Hewitt, J. 2014, MNRAS, 439, 4, 3262. doi:10.1093/mnras/stu125

    work page doi:10.1093/mnras/stu125 2014

  47. [47]

    N., et al

    Mevius, M., van der Tol, S., Pandey, V . N., et al. 2016, Radio Science, 51, 7,

    work page 2016

  48. [48]

    doi:10.1002/2016RS006028

    work page doi:10.1002/2016rs006028

  49. [49]

    2022 Cross-loop Propagation of a Quasiperiodic Extreme-ultraviolet Wave Train Triggered by Successive Stretching of Magnetic Field Structures during a Solar Eruption.Astrophys

    Moriwaki, K. & Yoshida, N. 2021b, ApJ, 923, 1, L7. doi:10.3847/2041- 8213/ac3cc0

    work page doi:10.3847/2041- 2041

  50. [50]

    2020, MNRAS, 496, 1, L54

    Moriwaki, K., Filippova, N., Shirasaki, M., & Yoshida, N. 2020, MNRAS, 496, 1, L54. doi:10.1093/mnrasl/slaa088

    work page doi:10.1093/mnrasl/slaa088 2020

  51. [51]

    B., & Yoshida, N

    Moriwaki, K., Eide, M. B., & Yoshida, N. 2021a, ApJ, 906, 1, L1. doi:10.3847/2041-8213/abd17f

    work page doi:10.3847/2041-8213/abd17f 2041

  52. [52]

    Reports on Progress in Physics , keywords =

    Moriwaki, K., Nishimichi, T., & Yoshida, N. 2023, Reports on Progress in Physics, 86, 7, 076901. doi:10.1088/1361-6633/acd2ea

    work page doi:10.1088/1361-6633/acd2ea 2023

  53. [53]

    2022, MNRAS, 511, 3, 3446

    Neutsch, S., Heneka, C., & Br ¨uggen, M. 2022, MNRAS, 511, 3, 3446. doi:10.1093/mnras/stac218

    work page doi:10.1093/mnras/stac218 2022

  54. [54]

    , keywords =

    Offringa, A. R., de Bruyn, A. G., Biehl, M., et al. 2010, MNRAS, 405, 1, 155. doi:10.1111/j.1365-2966.2010.16471.x

    work page doi:10.1111/j.1365-2966.2010.16471.x 2010

  55. [55]

    R., Wayth, R

    Offringa, A. R., Wayth, R. B., Hurley-Walker, N., et al. 2015, PASA, 32, e008. doi:10.1017/pasa.2015.7

    work page doi:10.1017/pasa.2015.7 2015

  56. [56]

    2023, MNRAS, 520, 4, 5552

    Pagano, M., et al. 2023, MNRAS, 520, 4, 5552. doi:10.1093/mnras/stad441

    work page doi:10.1093/mnras/stad441 2023

  57. [57]

    2017, arXiv e-prints, arXiv:1705.07057

    Papamakarios, G., Pavlakou, T., & Murray, I. 2017, arXiv e-prints, arXiv:1705.07057. doi:10.48550/arXiv.1705.07057

    work page doi:10.48550/arxiv.1705.07057 2017

  58. [58]

    2019, MNRAS, 484, 1, 933

    Park, J., Mesinger, A., Greig, B., et al. 2019, MNRAS, 484, 1, 933. doi:10.1093/mnras/stz032

    work page doi:10.1093/mnras/stz032 2019

  59. [59]

    R., Pober, J

    Parsons, A. R., Pober, J. C., Aguirre, J. E., et al. 2012, ApJ, 756, 2, 165. doi:10.1088/0004-637X/756/2/165

    work page doi:10.1088/0004-637x/756/2/165 2012

  60. [60]

    K., ˇSoltinsk´y, T., Maitra, S., & Kulkarni, G

    Patil, S. K., ˇSoltinsk´y, T., Maitra, S., & Kulkarni, G. 2026, MNRAS, 546, 4, stag236. doi:10.1093/mnras/stag236

    work page doi:10.1093/mnras/stag236 2026

  61. [61]

    Peißker, M

    Prelogovi ´c, D. & Mesinger, A. 2024, A&A, 688, A199. doi:10.1051/0004- 6361/202449309

    work page doi:10.1051/0004- 2024

  62. [62]

    2022, MNRAS, 509, 3, 3852

    Prelogovi ´c, D., Mesinger, A., Murray, S., et al. 2022, MNRAS, 509, 3, 3852. doi:10.1093/mnras/stab3215

    work page doi:10.1093/mnras/stab3215 2022

  63. [63]

    Pritchard, J. R. & Loeb, A. 2012, Reports on Progress in Physics, 75, 8, 086901. doi:10.1088/0034-4885/75/8/086901

    work page doi:10.1088/0034-4885/75/8/086901 2012

  64. [64]

    Rezende, D. J. & Mohamed, S. 2015, arXiv e-prints, arXiv:1505.05770. doi:10.48550/arXiv.1505.05770

    work page Pith review doi:10.48550/arxiv.1505.05770 2015

  65. [65]

    E., Dixon, K

    Ross, H. E., Dixon, K. L., Iliev, I. T., et al. 2017, MNRAS, 468, 4, 3785. doi:10.1093/mnras/stx649

    work page doi:10.1093/mnras/stx649 2017

  66. [66]

    2023, MNRAS, 525, 4, 6097

    Saxena, A., Cole, A., Gazagnes, S., et al. 2023, MNRAS, 525, 4, 6097. doi:10.1093/mnras/stad2659

    work page doi:10.1093/mnras/stad2659 2023

  67. [67]

    Schmit, C. J. & Pritchard, J. R. 2018, MNRAS, 475, 1, 1213. doi:10.1093/mnras/stx3292 1 Application of Machine Learning to 21 cm Cosmology 19

    work page doi:10.1093/mnras/stx3292 2018

  68. [68]

    2025, A&A, 698, A35

    Semelin, B., M ´eriot, R., Mishra, A., & Cornu, D. 2025, A&A, 698, A35. doi:10.1051/0004-6361/202453115

    work page doi:10.1051/0004-6361/202453115 2025

  69. [69]

    2023, Nature Astronomy, 7, 1116

    Shao, Y ., Xu, Y ., Wang, Y ., et al. 2023, Nature Astronomy, 7, 1116. doi:10.1038/s41550-023-02024-7

    work page doi:10.1038/s41550-023-02024-7 2023

  70. [70]

    2025, Phys

    Shao, Y ., Sun, T.-Y ., Zhao, M.-L., & Zhang, X. 2025, Phys. Rev. D, 112, 6, 063513. doi:10.1103/vpd1-1kyj

    work page doi:10.1103/vpd1-1kyj 2025

  71. [71]

    Shimabukuro H., 2026, Phys. Rev. D, 113, 083525. doi:10.1103/bl2w-crry

    work page doi:10.1103/bl2w-crry 2026

  72. [72]

    2025, Research in Astronomy and Astrophysics, 25, 8, 085017

    Shimabukuro, H. 2025, Research in Astronomy and Astrophysics, 25, 8, 085017. doi:10.1088/1674-4527/ade34f

    work page doi:10.1088/1674-4527/ade34f 2025

  73. [73]

    & Semelin, B

    Shimabukuro, H. & Semelin, B. 2017, MNRAS, 468, 4, 3869. doi:10.1093/mnras/stx734

    work page doi:10.1093/mnras/stx734 2017

  74. [74]

    2014, Phys

    Shimabukuro, H., Ichiki, K., Inoue, S., & Yokoyama, S. 2014, Phys. Rev. D, 90, 8, 083003. doi:10.1103/PhysRevD.90.083003

    work page doi:10.1103/physrevd.90.083003 2014

  75. [75]

    2015, MNRAS, 451, 1,

    Shimabukuro, H., Yoshiura, S., Takahashi, K., et al. 2015, MNRAS, 451, 1,

    work page 2015

  76. [76]

    doi:10.1093/mnras/stv965

    work page doi:10.1093/mnras/stv965

  77. [77]

    2016, MNRAS, 458, 3,

    Shimabukuro, H., Yoshiura, S., Takahashi, K., et al. 2016, MNRAS, 458, 3,

    work page 2016

  78. [78]

    doi:10.1093/mnras/stw482

    work page doi:10.1093/mnras/stw482

  79. [79]

    2017, MNRAS, 468, 2,

    Shimabukuro, H., Yoshiura, S., Takahashi, K., et al. 2017, MNRAS, 468, 2,

    work page 2017

  80. [80]

    doi:10.1093/mnras/stx530

    work page doi:10.1093/mnras/stx530

Showing first 80 references.