Pith. sign in

REVIEW 1 major objections 6 minor 259 references

Machine learning already covers the full SKA pipeline for Cosmic Dawn and reionization 21-cm science.

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

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

T0 review · grok-4.5

2026-07-12 01:13 UTC pith:PARNDIPI

load-bearing objection Solid AASKAII community survey of ML for SKA CD/EoR; useful synthesis, not a new result, and the authors already flag the transfer risks. the 1 major comments →

arxiv 2607.03606 v1 pith:PARNDIPI submitted 2026-07-03 astro-ph.IM astro-ph.CO

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

Anshuman Acharya , Michele Bianco , Daniela Breitman , Huaxi Chen , Abhirup Datta , Kangning Diao , Sambit K. Giri , Caroline S. Heneka
show 13 more authors
Nicholas Kern Adrian Liu Yashrajsinh Mahida Suman Majumdar Samit Kumar Pal Shulei Ni Yannic Pietschke Davide Piras Abinash Kumar Shaw Hayato Shimabukuro Ce Sui Anshuman Tripathi Xiaosheng Zhao
This is my paper
classification astro-ph.IM astro-ph.CO
keywords 21-cm cosmologyCosmic DawnEpoch of ReionizationSKAmachine learningforeground mitigationsimulation-based inferenceemulators
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The Square Kilometre Array will deliver the sensitivity needed for 21-cm tomography of Cosmic Dawn and the Epoch of Reionization, but also data volumes and imaging products that break conventional analysis. This chapter surveys machine-learning methods already proposed for every stage of that pipeline: radio-frequency-interference flagging, ionospheric and gain calibration, beam deconvolution, foreground mitigation (including learned-covariance Gaussian-process regression), field-level and summary-statistic emulators that accelerate simulations, and simulation-based inference that recovers astrophysical parameters without an explicit likelihood. The authors argue that these techniques are no longer peripheral prototypes; they form a coherent, end-to-end toolkit that can turn SKA data into robust constraints on the first billion years. A sympathetic reader cares because the same sensitivity that makes detection possible also makes classical pipelines too slow or too lossy, so the scientific return of the telescope may hinge on whether data-driven methods can be made trustworthy.

Core claim

Machine-learning algorithms already present in the literature span the entire Cosmic Dawn / Epoch of Reionization analysis chain—from low-level instrument modelling through theoretical emulation to field-level inference—and therefore constitute a serious contender for realising the scientific potential of SKA 21-cm tomography.

What carries the argument

The end-to-end pipeline view itself: treating RFI mitigation, calibration, imaging, ML-enhanced Gaussian-process foreground removal, neural emulators, and simulation-based inference as interchangeable modules that can be composed, and ultimately made differentiable, so that gradients flow from raw visibilities to astrophysical parameters.

Load-bearing premise

That methods shown to work on precursor data or on relatively cheap simulations will still perform correctly, without inventing false cosmological signals, once they face the real volume, noise and domain shifts of SKA-Low.

What would settle it

Deploy the surveyed ML-GPR or U-Net foreground pipelines on early SKA-Low commissioning data and check whether recovered 21-cm power spectra remain unbiased relative to classical avoidance methods and to independent multi-probe constraints; any systematic excess or deficit that cannot be absorbed by the training distribution would falsify the transfer claim.

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

If this is right

  • Real-time RFI flagging and calibration can move from heuristic thresholds to learned models that adapt as the RFI environment evolves.
  • Foregrounds can be modelled and subtracted rather than simply avoided, recovering the wedge region and boosting sensitivity by orders of magnitude.
  • Expensive radiative-transfer simulations can be replaced or accelerated by neural emulators, making high-dimensional Bayesian inference practical.
  • Field-level, likelihood-free inference becomes routine, extracting non-Gaussian information that power spectra discard.
  • Fully differentiable end-to-end pipelines become feasible, jointly optimising instrument, foreground and cosmology parameters.

Where Pith is reading between the lines

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

  • Once a single differentiable pipeline exists, multi-probe joint inference with JWST luminosity functions or CMB optical depth becomes a natural gradient step rather than a separate analysis.
  • Foundation-model fine-tuning on 21-cm light-cones may allow transfer from cheap semi-numerical boxes to high-fidelity hydrodynamical runs with only hundreds of expensive samples.
  • The same robustness tests demanded for SKA will force the community to develop standardised domain-shift benchmarks that later instruments can inherit.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 6 minor

Summary. This AASKAII chapter surveys machine-learning methods proposed for Cosmic Dawn and Epoch of Reionization science with the SKA. It organises the literature across the full pipeline: observational analysis (RFI flagging and inpainting, ionospheric TEC/phase-screen modelling, residual gain correction, physics-informed deconvolution such as PI-AstroDeconv, and foreground mitigation including ML-GPR with VAE-learned kernels), theory (field-level radiative-transfer emulators, CosmoUiT, GAN/diffusion/LLM map generators, super-resolution, and summary-statistic emulators for power spectra and bispectra, including Bayesian neural nets), and inference (image segmentation for HI/HII morphology, simulation-based inference with neural density estimators, network-learned summaries, multi-probe combinations, and diagnostics such as SBC/TARP). Sections 5–6 discuss automatic differentiation / differentiable programming and foundation-model trends, and close by arguing that ML is a serious contender for realising SKA 21-cm science while flagging domain shift, model misspecification and robustness.

Significance. As a community review chapter the manuscript’s value is organisational rather than a new derivation: it collates a rapidly growing literature into a single pipeline-oriented narrative and correctly emphasises that most methods remain at the validation or complementary stage. Strengths include the explicit treatment of ML-GPR (including the ELBO and learned-kernel construction), the physics-informed PI-AstroDeconv architecture, the discussion of SDC3a/b outcomes, and the repeated caveats on domain shift, residual systematics and model misspecification. These make the chapter a useful reference for the SKA EoR community provided the presentation issues below are cleaned up.

major comments (1)
  1. The central claim is only that ML methods already proposed span the pipeline and are therefore a serious contender; that claim is supported by the breadth of citations and by the authors’ own caveats (§2.1 final paragraphs; §4.2 SDC3b/model-misspecification discussion; §6 robustness paragraph). No load-bearing technical error or unsupported leap is required for the claim to hold. I therefore raise no major scientific objections.
minor comments (6)
  1. Placeholder citations “SKA foregrounds chapter (2025)” and “SKA imaging chapter (2025)” (and the arXiv:NNNN.MMMMM entries in the reference list) must be replaced by the actual AASKAII report numbers or arXiv identifiers before publication.
  2. §2.2 contains a clear typographical/grammatical break: “a non-negligible component in residual excesses. the upper limits of the CD/EoR power spectrum measurements (Pal et al., 2025).” Please repair the sentence.
  3. Eq. (1) and the surrounding PI-AstroDeconv description are useful, but the figure caption for Figure 1 is extremely long and partially duplicates the main text; a shorter caption with a pointer to the text would improve readability.
  4. Eq. (3) for the ML-GPR kernel is written with an expectation over z ~ q_ϕ(z|x,x′); a one-sentence clarification of how the joint encoding of a pair (x,x′) is obtained in practice would help non-specialist readers.
  5. Several consecutive sentences in §3.1 begin with “Moreover,” and a few acronyms (e.g., CosmoUiT, SDC3a/b) appear before or without expansion; a light copy-edit pass would remove these presentation nits.
  6. Figure 3 (Kern 2025 differentiable pipeline) is informative but dense; ensuring that the five numbered steps remain legible at journal column width would help.

Circularity Check

0 steps flagged

No circularity: literature survey with no derivation chain that reduces predictions or first-principles claims to their own inputs.

full rationale

The manuscript is an AASKAII review chapter whose stated purpose is to overview machine-learning methods already proposed for CD/EoR with the SKA (Abstract; §1; §6). It does not advance a new derivation, uniqueness theorem, fitted-parameter prediction, or first-principles result whose output is forced by construction from its inputs. Descriptions of ML-GPR (VAE-learned kernels, ELBO, k_ML), PI-AstroDeconv, field-level/summary emulators, and SBI are expositions of prior published techniques, with equations that define those methods rather than claim independent predictions. Self-citations by the large author list are numerous and expected in a community working-group survey; they document the literature being reviewed and do not serve as load-bearing external uniqueness or ansatz justifications that close a circular loop. No fitted quantities are re-labeled as predictions, no ansatz is smuggled via self-citation as if external, and no known empirical pattern is merely renamed. The central claim—that proposed ML methods span the pipeline and are a serious contender—rests on the breadth of cited work and the authors’ own caveats on domain shift and robustness, not on a self-referential reduction. Score 0 is therefore the correct honest finding.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

As a literature review the paper inherits the standard assumptions of 21-cm cosmology and of the cited ML papers; it introduces no free parameters of its own and no new physical entities. The load-bearing background assumptions are those of the domain (existence of a detectable 21-cm signal, utility of power spectra / maps, validity of semi-numerical simulators for training) rather than ad-hoc postulates invented here.

axioms (3)
  • domain assumption The cosmological 21-cm signal from CD/EoR is in principle recoverable from SKA-Low data once foregrounds, RFI, ionosphere and calibration systematics are controlled.
    Stated throughout §1–2 and underpins every surveyed application; standard in the field but not proven by this paper.
  • domain assumption Semi-numerical or radiative-transfer simulations (21cmFAST, GRIZZLY, etc.) provide sufficiently realistic training distributions for the ML models discussed.
    Invoked for ML-GPR, emulators, SBI and generative models (§2.5, §3, §4); acknowledged as a source of model misspecification risk.
  • standard math Standard deep-learning architectures (U-Net, VAE, GAN, diffusion, transformers, NDEs) and GPR are valid function approximators for the radio-astronomy tasks surveyed.
    Background mathematical toolkit assumed throughout; no new proofs offered.

pith-pipeline@v1.1.0-grok45 · 37200 in / 2252 out tokens · 23986 ms · 2026-07-12T01:13:43.873148+00:00 · methodology

0 comments
read the original abstract

When operational, the SKA will generate unprecedented amounts of data and provide exquisite sensitivity for 21 cm tomography of Cosmic Dawn (CD) and the Epoch of Reionization (EoR). With this comes opportunities for new data-driven algorithms that unlock new methods for instrument modelling, data analysis, theoretical simulation, and inference for understanding the high-redshift universe. In this chapter, we provide an overview of some machine learning algorithms that have been proposed for CD and EoR science with the SKA

Figures

Figures reproduced from arXiv: 2607.03606 by Abhirup Datta, Abinash Kumar Shaw, Adrian Liu, Anshuman Acharya, Anshuman Tripathi, Caroline S. Heneka, Ce Sui, Daniela Breitman, Davide Piras, Hayato Shimabukuro, Huaxi Chen, Kangning Diao, Michele Bianco, Nicholas Kern, Sambit K. Giri, Samit Kumar Pal, Shulei Ni, Suman Majumdar, Xiaosheng Zhao, Yannic Pietschke, Yashrajsinh Mahida.

Figure 1
Figure 1. Figure 1: The PI-AstroDeconv architecture. The yellow blocks represent convolutional layers, red blocks represent pooling layers, and gray-green blocks represent upsampling operations. The left half of the network depicts the downsampling path, while the right half represents the upsampling path. The number of channels is indicated at the bottom of each block. The arrows above the image symbolize the connections bet… view at source ↗
Figure 2
Figure 2. Figure 2: Possible workflows for likelihood-free inference (Sec. 4): Once a database of 21 cm simulations is created, a suitable summary statistic (e.g. power spectra, higher-order statistics, or network-learned features) is extracted and used as input for simulation-based, likelihood-free inference based on neural density estimation. The resulting posteriors can be quantitatively evaluated using diagnostics such as… view at source ↗
Figure 3
Figure 3. Figure 3: A differentiable, Bayesian forward model for 21 cm cosmology from Kern (2025), highlighting the forward operation of visibility simulation, and the backward operation of reverse-mode backpropagation. Optimization and sampling of the (un-normalized) posterior distribution proceeds iteratively, using the computed gradients to move within the joint parameter space of the model. be mistaken for cosmological si… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

259 extracted references · 79 canonical work pages · 65 internal anchors

  1. [1]

    2019 , eprint=

    Anticipated Performance of the Square Kilometre Array -- Phase 1 (SKA1) , author=. 2019 , eprint=

    2019

  2. [2]

    Characterization Of Inpaint Residuals In Interferometric Measurements of the Epoch Of Reionization

    Characterization of inpaint residuals in interferometric measurements of the epoch of reionization. , keywords =. doi:10.1093/mnras/stad441 , archivePrefix =. 2210.14927 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stad441

  3. [3]

    2026 ,publisher =

    Anshuman Acharya and author2 and author3 and author4 and author5 ,title =. 2026 ,publisher =

    2026

  4. [4]

    2026 ,publisher =

    Anirban Chakraborty and author2 and author3 and author4 and author5 ,title =. 2026 ,publisher =

    2026

  5. [5]

    An Alcock-Paczynski Test on Reionization Bubbles for Cosmology

    Alcock-Paczy \'n ski test on reionization bubbles for cosmology. , keywords =. doi:10.1103/jfkk-7q7l , archivePrefix =. 2502.02638 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1103/jfkk-7q7l

  6. [6]

    2015 , publisher=

    Advancing Astrophysics with the Square Kilometre Array (AASKA14) , author=. 2015 , publisher=

    2015

  7. [7]

    PINION: physics-informed neural network for accelerating radiative transfer simulations for cosmic reionization , volume=

    Korber, Damien and Bianco, Michele and Tolley, Emma and Kneib, Jean-Paul , year=. PINION: physics-informed neural network for accelerating radiative transfer simulations for cosmic reionization , volume=. MNRAS , publisher=. doi:10.1093/mnras/stad615 , number=

    doi:10.1093/mnras/stad615

  8. [8]

    , keywords =

    Direct reconstruction of the Reionization history from 21cm 2D Power Spectra. , keywords =. doi:10.1088/1475-7516/2025/10/039 , adsurl =

    doi:10.1088/1475-7516/2025/10/039 2025

  9. [9]

    Field-level simulation-based inference of galaxy clustering with convolutional neural networks

    Lemos, Pablo and others. Field-level simulation-based inference of galaxy clustering with convolutional neural networks. Phys. Rev. D. 2024. doi:10.1103/PhysRevD.109.083536. arXiv:2310.15256

    Pith/arXiv arXiv doi:10.1103/physrevd.109.083536 2024

  10. [10]

    , keywords =

    Comparison of Bayesian inference methods using the LORELI II database of hydro-radiative simulations of the 21-cm signal. , keywords =. doi:10.1051/0004-6361/202452901 , archivePrefix =. 2411.03093 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/202452901

  11. [11]

    Data Compression and Inference in Cosmology with Self-Supervised Machine Learning

    Akhmetzhanova, Aizhan and Mishra-Sharma, Siddharth and Dvorkin, Cora. Data Compression and Inference in Cosmology with Self-Supervised Machine Learning. Mon. Not. Roy. Astron. Soc. 2024. doi:10.1093/mnras/stad3646. arXiv:2308.09751

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stad3646 2024

  12. [12]

    , keywords =

    Inferring the astrophysics of reionization and cosmic dawn from galaxy luminosity functions and the 21-cm signal. , keywords =. doi:10.1093/mnras/stz032 , archivePrefix =. 1809.08995 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stz032

  13. [13]

    Epoch of reionization parameter estimation with the 21-cm bispectrum

    Epoch of reionization parameter estimation with the 21-cm bispectrum. , keywords =. doi:10.1093/mnras/stab3706 , archivePrefix =. 2102.02310 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stab3706

  14. [14]

    The Journal of Open Source Software , keywords =

    21cmFAST v3: A Python-integrated C code for generating 3D realizations of the cosmic 21cm signal. The Journal of Open Source Software , keywords =. doi:10.21105/joss.02582 , archivePrefix =. 2010.15121 , primaryClass =

    Pith/arXiv arXiv doi:10.21105/joss.02582 2010

  15. [15]

    and others

    Bacon, David J. and others. Cosmology with Phase 1 of the Square Kilometre Array: Red Book 2018: Technical specifications and performance forecasts. Publ. Astron. Soc. Austral. 2020. doi:10.1017/pasa.2019.51. arXiv:1811.02743

    Pith/arXiv arXiv doi:10.1017/pasa.2019.51 2018

  16. [16]

    Cosmology with one galaxy?

    Cosmology with One Galaxy?. , keywords =. doi:10.3847/1538-4357/ac5d3f , archivePrefix =. 2201.02202 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-4357/ac5d3f

  17. [17]

    ML4Astro International Conference , pages=

    Deep Learning 21 cm Lightcones in 3D , author=. ML4Astro International Conference , pages=. 2022 , organization=. doi:10.48550/arXiv.2311.17553 , archivePrefix =. 2311.17553 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2311.17553 2022

  18. [18]

    Galaxy Spectra neural Network (GaSNet). II. Using Deep Learning for Spectral Classification and Redshift Predictions. arXiv e-prints , keywords =. doi:10.48550/arXiv.2311.04146 , archivePrefix =. 2311.04146 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2311.04146

  19. [19]

    The DAWES review 10: The impact of deep learning for the analysis of galaxy surveys

    Huertas-Company, Marc and Lanusse, Fran c ois. The DAWES review 10: The impact of deep learning for the analysis of galaxy surveys. Publ. Astron. Soc. Austral. 2023. doi:10.1017/pasa.2022.55. arXiv:2210.01813

    Pith/arXiv arXiv doi:10.1017/pasa.2022.55 2023

  20. [20]

    arXiv e-prints , keywords =

    The LoReLi database: 21 cm signal inference with 3D radiative hydrodynamics simulations. arXiv e-prints , keywords =. doi:10.48550/arXiv.2310.02684 , archivePrefix =. 2310.02684 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2310.02684

  21. [21]

    Constraining coupled quintessence with the 21cm signal

    Constraining coupled quintessence with the 21 cm signal. , keywords =. doi:10.1088/1475-7516/2020/05/038 , archivePrefix =. 1910.02763 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/1475-7516/2020/05/038 2020

  22. [22]

    , keywords =

    General modified gravity with 21cm intensity mapping: simulations and forecast. , keywords =. doi:10.1088/1475-7516/2018/10/004 , archivePrefix =. 1805.03629 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/1475-7516/2018/10/004 2018

  23. [23]

    Can Diffusion Model Conditionally Generate Astrophysical Images?

    Can diffusion model conditionally generate astrophysical images?. , keywords =. doi:10.1093/mnras/stad2778 , archivePrefix =. 2307.09568 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stad2778

  24. [24]

    , keywords =

    SKA Science Data Challenge 2: analysis and results. , keywords =. doi:10.1093/mnras/stad1375 , archivePrefix =. 2303.07943 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stad1375

  25. [25]

    Photometry of high-redshift blended galaxies using deep learning

    Photometry of high-redshift blended galaxies using deep learning. , keywords =. doi:10.1093/mnras/stz3056 , archivePrefix =. 1905.01324 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stz3056 1905

  26. [26]

    , keywords =

    Gaia DR2 unravels incompleteness of nearby cluster population: new open clusters in the direction of Perseus. , keywords =. doi:10.1051/0004-6361/201834453 , archivePrefix =. 1810.05494 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/201834453

  27. [27]

    , keywords =

    Probing the Intergalactic Medium with Ly and 21 cm Fluctuations. , keywords =. doi:10.3847/1538-4357/aa8eed , archivePrefix =. 1611.09682 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4357/aa8eed

  28. [28]

    The spin-temperature dependence of the 21cm -- LAE cross-correlation

    The spin-temperature dependence of the 21-cm-LAE cross-correlation. , keywords =. doi:10.1093/mnras/staa1517 , archivePrefix =. 2004.10097 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/staa1517 2004

  29. [29]

    Large, fast and accurate HI intensity maps with latent overlap diffusion

    Large, fast and accurate HI intensity maps with latent overlap diffusion. arXiv e-prints , keywords =. doi:10.48550/arXiv.2506.08086 , archivePrefix =. 2506.08086 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2506.08086

  30. [30]

    arXiv e-prints , keywords =

    CosmoGLINT: Cosmological Generative Model for Line Intensity Mapping with Transformer. arXiv e-prints , keywords =. doi:10.48550/arXiv.2506.16843 , archivePrefix =. 2506.16843 , primaryClass =

    work page doi:10.48550/arxiv.2506.16843

  31. [31]

    arXiv e-prints , keywords =

    Large Language Models -- the Future of Fundamental Physics?. arXiv e-prints , keywords =. doi:10.48550/arXiv.2506.14757 , archivePrefix =. 2506.14757 , primaryClass =

    doi:10.48550/arxiv.2506.14757

  32. [32]

    , keywords =

    On the general nature of 21-cm-Lyman emitter cross-correlations during reionization. , keywords =. doi:10.1093/mnras/stad2376 , archivePrefix =. 2306.03156 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stad2376

  33. [33]

    Searching for bias and correlations in a Bayesian way

    Heneka, Caroline and Posada, Alexandre and Marra, Valerio and Amendola, Luca. Searching for bias and correlations in a Bayesian way - Example: SN Ia data. IAU Symp. 2014. doi:10.1017/S1743921315000010. arXiv:1407.2531

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1017/s1743921315000010 2014

  34. [34]

    Stress testing the dark energy equation of state imprint on supernova data

    Moews, Ben and de Souza, Rafael S. and Ishida, Emille E. O. and Malz, Alex I. and Heneka, Caroline and Vilalta, Ricardo and Zuntz, Joe. Stress testing the dark energy equation of state imprint on supernova data. Phys. Rev. D. 2019. doi:10.1103/PhysRevD.99.123529. arXiv:1812.09786

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1103/physrevd.99.123529 2019

  35. [37]

    arXiv e-prints , keywords =

    Masked Autoregressive Flow for Density Estimation. arXiv e-prints , keywords =. doi:10.48550/arXiv.1705.07057 , archivePrefix =. 1705.07057 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1705.07057

  36. [38]

    Neural Spline Flows , url =

    Durkan, Conor and Bekasov, Artur and Murray, Iain and Papamakarios, George , booktitle =. Neural Spline Flows , url =

  37. [39]

    , keywords =

    Constraining the X-ray heating and reionization using 21-cm power spectra with Marginal Neural Ratio Estimation. , keywords =. doi:10.1093/mnras/stad2659 , archivePrefix =. 2303.07339 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stad2659

  38. [40]

    Simulation-Based Inference of the sky-averaged 21-cm signal from CD-EoR with REACH

    Simulation-based inference of the sky-averaged 21-cm signal from CD-EoR with REACH. RAS Techniques and Instruments , keywords =. doi:10.1093/rasti/rzae047 , archivePrefix =. 2403.14618 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/rasti/rzae047

  39. [41]

    Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics , pages =

    Sequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flows , author =. Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics , pages =. 2019 , editor =

    2019

  40. [42]

    arXiv e-prints , keywords =

    Likelihood-free inference with emulator networks. arXiv e-prints , keywords =. doi:10.48550/arXiv.1805.09294 , archivePrefix =. 1805.09294 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1805.09294

  41. [43]

    arXiv e-prints , keywords =

    Flexible statistical inference for mechanistic models of neural dynamics. arXiv e-prints , keywords =. doi:10.48550/arXiv.1711.01861 , archivePrefix =. 1711.01861 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1711.01861

  42. [44]

    Proceedings of the 36th International Conference on Machine Learning , pages =

    Automatic Posterior Transformation for Likelihood-Free Inference , author =. Proceedings of the 36th International Conference on Machine Learning , pages =. 2019 , editor =

    2019

  43. [45]

    Proceedings of the 37th International Conference on Machine Learning , articleno =

    Hermans, Joeri and Begy, Volodimir and Louppe, Gilles , title =. Proceedings of the 37th International Conference on Machine Learning , articleno =. 2020 , publisher =

    2020

  44. [46]

    Proceedings of the 37th International Conference on Machine Learning , pages =

    On Contrastive Learning for Likelihood-free Inference , author =. Proceedings of the 37th International Conference on Machine Learning , pages =. 2020 , editor =

    2020

  45. [47]

    2017 , eprint=

    Density estimation using Real NVP , author=. 2017 , eprint=

    2017

  46. [48]

    2016 , eprint=

    Neural Autoregressive Distribution Estimation , author=. 2016 , eprint=

    2016

  47. [49]

    arXiv e-prints , keywords =

    Validating Bayesian Inference Algorithms with Simulation-Based Calibration. arXiv e-prints , keywords =. doi:10.48550/arXiv.1804.06788 , archivePrefix =. 1804.06788 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1804.06788

  48. [50]

    arXiv e-prints , keywords =

    Radiometer Calibration using Machine Learning. arXiv e-prints , keywords =. doi:10.48550/arXiv.2504.16791 , archivePrefix =. 2504.16791 , primaryClass =

    work page doi:10.48550/arxiv.2504.16791

  49. [51]

    Joint analysis constraints on the physics of the first galaxies with low frequency radio astronomy data

    Joint analysis constraints on the physics of the first galaxies with low-frequency radio astronomy data. , keywords =. doi:10.1093/mnras/stad3194 , archivePrefix =. 2301.03298 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stad3194

  50. [52]

    , keywords =

    Marginal post-processing of Bayesian inference products with normalizing flows and kernel density estimators. , keywords =. doi:10.1093/mnras/stad2997 , archivePrefix =. 2205.12841 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stad2997

  51. [53]

    2, 2022-06-27 , author=

    A path towards autonomous machine intelligence version 0.9. 2, 2022-06-27 , author=. Open Review , volume=

    2022

  52. [54]

    arXiv e-prints , keywords =

    Rapid and Late Cosmic Reionization Driven by Massive Galaxies: a Joint Analysis of Constraints from 21-cm, Lyman Line & CMB Data Sets. arXiv e-prints , keywords =. doi:10.48550/arXiv.2504.09725 , archivePrefix =. 2504.09725 , primaryClass =

    doi:10.48550/arxiv.2504.09725

  53. [55]

    Journal of Computational and Graphical Statistics , volume=

    Validation of software for Bayesian models using posterior quantiles , author=. Journal of Computational and Graphical Statistics , volume=. 2006 , publisher=

    2006

  54. [56]

    40th International Conference on Machine Learning , keywords =

    Sampling-Based Accuracy Testing of Posterior Estimators for General Inference. 40th International Conference on Machine Learning , keywords =. doi:10.48550/arXiv.2302.03026 , archivePrefix =. 2302.03026 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2302.03026

  55. [57]

    Inferring astrophysical parameters using the 2D cylindrical power spectrum from reionisation

    Inferring astrophysical parameters using the 2D cylindrical power spectrum from reionization. , keywords =. doi:10.1093/mnras/stae1984 , archivePrefix =. 2403.14060 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stae1984

  56. [59]

    SciPost Physics Core , keywords =

    Optimal, fast, and robust inference of reionization-era cosmology with the 21cmPIE-INN. SciPost Physics Core , keywords =. doi:10.21468/SciPostPhysCore.8.2.037 , archivePrefix =. 2401.04174 , primaryClass =

    Pith/arXiv arXiv doi:10.21468/scipostphyscore.8.2.037

  57. [60]

    Inferring Astrophysics and Dark Matter Properties from 21cm Tomography using Deep Learning

    Inferring astrophysics and dark matter properties from 21 cm tomography using deep learning. , keywords =. doi:10.1093/mnras/stac218 , archivePrefix =. 2201.07587 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stac218

  58. [61]

    arXiv e-prints , keywords =

    Hybrid Summary Statistics. arXiv e-prints , keywords =. doi:10.48550/arXiv.2410.07548 , archivePrefix =. 2410.07548 , primaryClass =

    doi:10.48550/arxiv.2410.07548

  59. [62]

    Monthly Notices of the Royal Astronomical Society , volume =

    Chardin, Jonathan and Uhlrich, Grégoire and Aubert, Dominique and Deparis, Nicolas and Gillet, Nicolas and Ocvirk, Pierre and Lewis, Joseph , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2019 , month =. doi:10.1093/mnras/stz2605 , url =

    doi:10.1093/mnras/stz2605 2019

  60. [63]

    Bayesian data analysis for sky-averaged 21-cm experiments in the presence of ionospheric effects , volume=

    Shen, Emma and Anstey, Dominic and de Lera Acedo, Eloy and Fialkov, Anastasia , year=. Bayesian data analysis for sky-averaged 21-cm experiments in the presence of ionospheric effects , volume=. Monthly Notices of the Royal Astronomical Society , publisher=. doi:10.1093/mnras/stac1900 , number=

    work page doi:10.1093/mnras/stac1900

  61. [64]

    2025 , eprint=

    Cosmological super-resolution of the 21-cm signal , author=. 2025 , eprint=

    2025

  62. [66]

    Eliminating Primary Beam Effect in Foreground Subtraction of Neutral Hydrogen Intensity Mapping Survey with Deep Learning

    Ni, Shulei and Li, Yichao and Gao, Li-Yang and Zhang, Xin. Eliminating Primary Beam Effect in Foreground Subtraction of Neutral Hydrogen Intensity Mapping Survey with Deep Learning. Astrophys. J. 2022. doi:10.3847/1538-4357/ac7a34. arXiv:2204.02780

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-4357/ac7a34 2022

  63. [67]

    Spectral Redundancy for Calibrating Interferometers and Suppressing the Foreground Wedge in 21cm Cosmology

    Cox, Tyler A. and Parsons, Aaron R. and Dillon, Joshua S. and Ewall-Wice, Aaron and Pascua, Robert. Spectral redundancy for calibrating interferometers and suppressing the foreground wedge in 21 \, cm cosmology. Mon. Not. Roy. Astron. Soc. 2024. doi:10.1093/mnras/stae1612. arXiv:2311.01422

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stae1612 2024

  64. [68]

    Demonstration of hybrid foreground removal on CHIME data

    Wang, Haochen and others. Demonstration of hybrid foreground removal on CHIME data. Phys. Rev. D. 2025. doi:10.1103/PhysRevD.111.103531. arXiv:2408.08949

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1103/physrevd.111.103531 2025

  65. [69]

    Astronomy and Astrophysics Supplement, Vol

    Aperture synthesis with a non-regular distribution of interferometer baselines , author=. Astronomy and Astrophysics Supplement, Vol. 15, p. 417 , volume=

  66. [70]

    Astronomy & Astrophysics , volume=

    A multi-scale multi-frequency deconvolution algorithm for synthesis imaging in radio interferometry , author=. Astronomy & Astrophysics , volume=. 2011 , publisher=

    2011

  67. [71]

    Monthly Notices of the Royal Astronomical Society , volume=

    WSCLEAN: an implementation of a fast, generic wide-field imager for radio astronomy , author=. Monthly Notices of the Royal Astronomical Society , volume=. 2014 , publisher=

    2014

  68. [72]

    2024 , eprint=

    PI-AstroDeconv: A Physics-Informed Unsupervised Learning Method for Astronomical Image Deconvolution , author=. 2024 , eprint=

    2024

  69. [73]

    The Astrophysical Journal , volume=

    Application of Physics-informed Neural Networks in Removing Telescope Beam Effects , author=. The Astrophysical Journal , volume=. 2025 , publisher=

    2025

  70. [74]

    Deep learning-based radiointerferometric imaging with GAN-aided training

    Deep-learning-based radiointerferometric imaging with GAN-aided training. , keywords =. doi:10.1051/0004-6361/202347073 , archivePrefix =. 2307.14100 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1051/0004-6361/202347073

  71. [75]

    and Geyer, F

    Schmidt, K. and Geyer, F. and Fröse, S. and Blomenkamp, P.-S. and Brüggen, M. and de Gasperin, F. and Elsässer, D. and Rhode, W. , year=. Deep learning-based imaging in radio interferometry , volume=. doi:10.1051/0004-6361/202142113 , journal=

    doi:10.1051/0004-6361/202142113

  72. [76]

    , keywords =

    Deep learning from 21-cm tomography of the cosmic dawn and reionization. , keywords =. doi:10.1093/mnras/stz010 , archivePrefix =. 1805.02699 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stz010

  73. [77]

    Reionisation time fields reconstruction from 21 cm signal maps

    Reionisation time field reconstruction from 21 cm signal maps. , keywords =. doi:10.1051/0004-6361/202346495 , archivePrefix =. 2307.00609 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1051/0004-6361/202346495

  74. [78]

    arXiv e-prints , keywords =

    Transfer learning for multifidelity simulation-based inference in cosmology. arXiv e-prints , keywords =. doi:10.48550/arXiv.2505.21215 , archivePrefix =. 2505.21215 , primaryClass =

    work page doi:10.48550/arxiv.2505.21215

  75. [79]

    , keywords =

    Deeper multi-redshift upper limits on the epoch of reionisation 21 cm signal power spectrum from LOFAR between z = 8.3 and z = 10.1. , keywords =. doi:10.1051/0004-6361/202554158 , archivePrefix =. 2503.05576 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/202554158

  76. [80]

    arXiv e-prints , keywords =

    Square Kilometre Array Science Data Challenge 3a: foreground removal for an EoR experiment. arXiv e-prints , keywords =. doi:10.48550/arXiv.2503.11740 , archivePrefix =. 2503.11740 , primaryClass =

    doi:10.48550/arxiv.2503.11740

  77. [81]

    arXiv e-prints , keywords =

    First upper limits on the 21-cm signal power spectrum of neutral hydrogen at z=9.16 from the LOFAR 3C196 field. arXiv e-prints , keywords =. doi:10.48550/arXiv.2504.18534 , archivePrefix =. 2504.18534 , primaryClass =

    doi:10.48550/arxiv.2504.18534

  78. [82]

    , keywords =

    Revised LOFAR upper limits on the 21-cm signal power spectrum at z 9.1 using machine learning and gaussian process regression. , keywords =. doi:10.1093/mnrasl/slae078 , archivePrefix =. 2408.10051 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnrasl/slae078

  79. [83]

    , keywords =

    First upper limits on the 21 cm signal power spectrum from cosmic dawn from one night of observations with NenuFAR. , keywords =. doi:10.1051/0004-6361/202348329 , archivePrefix =. 2311.05364 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/202348329

  80. [84]

    21-cm Signal from the Epoch of Reionization: A Machine Learning upgrade to Foreground Removal with Gaussian Process Regression

    21-cm signal from the Epoch of Reionization: a machine learning upgrade to foreground removal with Gaussian process regression. , keywords =. doi:10.1093/mnras/stad3701 , archivePrefix =. 2311.16633 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stad3701

Showing first 80 references.