Pith. sign in

REVIEW 1 major objections 5 minor 55 references

Mapping Diffuse Radio Sources Using TUNA: A Transformer-Based Deep Learning Approach

T0 review · 1 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A Transformer-based network trained only on simulated radio skies maps faint diffuse emission, including megahalos and cluster bridges, directly from native-resolution LOFAR survey images, without source subtraction or re-imaging.

desk verdict A credible, data-released application of TransUNet to diffuse radio source segmentation; the real-data validation leans on the pipeline it aims to replace, so the strongest claims need a tighter evaluation. read the letter →

arxiv 2507.11320 v2 pith:JKQVQNEM submitted 2025-07-15 astro-ph.IM

Nicoletta Sanvitale , Claudio Gheller , Franco Vazza , Annalisa Bonafede , Virginia Cuciti , Emanuele De Rubeis , Federica Govoni , Matteo Murgia
show 1 more author
Valentina Vacca
This is my paper · ORCID
classification astro-ph.IM
keywords diffuseradioemissiongalaxyclustersmegahalosLOFARsurveysdeeplearningsegmentationVisionTransformersbridgesimageprocessing
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

The paper introduces TUNA, a deep-learning segmentation network built by fusing a Vision Transformer into a U-Net, and claims that it can map faint, extended radio sources directly in native-resolution LOFAR survey images. Trained exclusively on mock observations generated from cosmological simulations, the network generalizes to real LOFAR data without retraining, detecting diffuse emission that normally becomes visible only after sources are removed and the image is reprocessed at 4-6 times coarser resolution. On the LoTSS-DR2/PSZ2 cluster sample, TUNA attains an IoU of 0.43 and recall of 0.61 against source-subtracted tapered images, more than doubling and tripling the R-UNet CNN baseline. It also recovers the A399-A401 ridge, the Abell 1758 bridge, and all four known megahalos directly from high-resolution or native-resolution input. If the claim holds, automated pipelines could screen the huge next-generation radio survey volumes for rare diffuse sources without expensive reprocessing.

What carries the argument

The load-bearing object is TUNA, a customized TransUNet: a U-Net whose encoder is a hybrid CNN-Transformer, with a ResNet-50 feature extractor feeding a 12-layer Vision Transformer that applies self-attention across image patches, followed by bilinear upsampling blocks in the decoder. This lets the model combine long-range contextual reasoning, which diffuse sources need because they extend over large angular scales and must be distinguished from calibration artifacts, with local boundary fidelity. Equally essential is the training-data machinery: over 500 mock 1.1 by 1.1 degree LOFAR HBA observations are generated from cosmological MHD simulations by projecting synchrotron emission from the shock-acceleration model into light cones, adding Gaussian noise at LoTSS noise levels, and imaging with WSClean at 6 and 20 arcsec resolutions. The network is trained to reproduce binary masks from the noiseless sky images, learning to ignore the artifacts and noise of the clean images.

What would settle it

Apply TUNA to LoTSS pointings with no previously known diffuse emission, then independently re-image the flagged fields with deep source subtraction and heavy uv-tapering; any confident TUNA mask that has no counterpart at 3-sigma or above in the reprocessed image would show that the network is detecting imaging artifacts rather than low-surface-brightness sky emission.

Watch

Extended reading notes

Core claim

The paper's central claim is that a Transformer-enhanced U-Net trained on synthetic LOFAR-like observations can segment real low-surface-brightness radio emission from survey images at their native roughly 6 arcsec resolution, with no manual subtraction of compact sources and no low-resolution re-imaging. The authors argue that the self-attention mechanism gives TUNA the long-range context needed to tell large, faint diffuse structures apart from imaging artifacts, while the U-Net decoder preserves boundary detail. Applied to the 246 usable LoTSS-DR2/PSZ2 clusters, TUNA's masks best match the diffuse emission visible in source-subtracted images at 20-40 arcsec resolution, equivalent to reprocessing the input 4-6 times coarser, and the network recovers confirmed examples of a radio ridge, a bridge, and four megahalos. The authors further claim that TUNA outperforms the earlier R-UNet on the same sample, with IoU 0.43 against 0.19 and recall 0.61 against 0.20, and that it generalizes to source types never seen in training, including AGN jets, which they present as evidence for blind source detection.

Load-bearing premise

The load-bearing premise is that mock images built from shock-accelerated radio emission alone, without turbulent re-acceleration, resemble real halos, bridges, and megahalos closely enough that TUNA learns genuine diffuse-source morphology rather than simulation-specific patterns.

Editorial extensions

If this is right

  • The full LoTSS-DR2 pointing P128+37, at 9528 by 9528 pixels, is segmented in about 193 seconds at 6 arcsec resolution on one Ampere 100 GPU, against up to a day for conventional source subtraction and tapered re-imaging, so whole-survey diffuse-source screening becomes practical.
  • Megahalos, radio bridges, and radio halos can be recovered from public native-resolution survey images without manual source subtraction or low-resolution reprocessing, making archival LoTSS data re-mineable for rare sources.
  • Because the network also detects AGN jets and compact sources it never saw in training, the same approach can be extended toward blind source detection and classification in SKA-era surveys.
  • Feeding the network native-resolution data instead of degraded images lowers confusion noise and reduces the chance that blended point sources are misclassified as diffuse objects, according to the paper's own analysis.

Reading between the lines

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

  • We infer that the network's success on turbulence-dominated megahalos, despite training only on shock-accelerated emission, suggests the observed morphology of these sources at LOFAR sensitivity is set more by projection and magnetic-field structure than by the acceleration mechanism; this would make simulation-based training more robust than the authors' caveat implies.
  • We infer that the evaluation ground truth is itself a product of the source-subtraction and tapering pipeline TUNA is meant to replace, so the reported IoU and recall measure agreement with that pipeline's output, not directly with the sky; independent follow-up of TUNA-only candidates is needed to establish true detection reliability.
  • We infer that the next decisive experiment is to retrain TUNA on mock images that include turbulent reacceleration and compare detections; any change in which faint sources are recovered would reveal how much of the current performance depends on the omitted emission channel.
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

1 major / 5 minor

Summary. The paper presents TUNA, a TransUNet-based deep learning segmentation model for detecting faint, diffuse radio emission in LOFAR images. The model is trained on mock observations derived from cosmological MHD simulations that include only shock-accelerated synchrotron emission, and is then applied, without retraining, to real LoTSS-DR2 data. The authors report that TUNA outperforms a prior R-UNet on both simulated and real data, and that it can recover diffuse emission at scales equivalent to images reprocessed 4–6 times coarser than the native resolution, without manual source subtraction or low-resolution re-imaging. Qualitative demonstrations include the A399–A401 ridge, the A1758 bridge, and four known megahalos.

Significance. If the central claims hold, TUNA would provide a fast, automated alternative to traditional source subtraction and uv-tapered re-imaging for detecting diffuse cluster radio emission, which is valuable for current and future large-area surveys. The paper ships a well-described architecture, public data products, and quantitative performance metrics on both simulated and real data, and it is honest about several limitations. The key risk lies in the domain gap between the shock-only simulated training set and the turbulence-dominated real sources, and in the fact that the real-data ground truth is itself derived from the very re-imaging pipeline TUNA aims to replace.

major comments (1)
  1. [Section 4.1] The performance gains over R-UNet in Table 2 are reported with large cluster-to-cluster standard deviations (e.g., IoU 0.43±0.13 vs 0.19±0.15; recall 0.61±0.18 vs 0.20±0.16). The authors do not provide any statistical significance test for these differences. Given the modest sample size (131 clusters for Fig. 6a, 15 clusters for the 60''–120'' claim) and the large scatter, a paired bootstrap or Wilcoxon signed-rank test over the same clusters would substantially strengthen the conclusion that TUNA's improvement is not due to chance.
minor comments (5)
  1. [Abstract] The phrase "groundbreaking capability" in the abstract is promotional; consider a more neutral formulation such as "a capability that was previously unavailable".
  2. [Eq. (1)] Equation (1) uses H' and W' but does not define the floor division or clarify that these are the output feature-map dimensions after the CNN backbone; please clarify the notation.
  3. [Section 3.3] The caption of Fig. 4 contains a duplicated word: "for for".
  4. [Section 5] The conclusion states that TUNA "generalizes to diverse source types not present in the training set, including AGN and their associated jets," but no quantitative evaluation of AGN detection is provided; if this claim is retained, it should be supported by at least a qualitative figure or a reference to the online material.
  5. [References] Reference "Sanvitale N., Gheller C., Bowman E., 2022, Granular Matter, 24" appears unrelated to the radio-astronomy tiling method cited in Section 3.1; please verify that this is the correct citation.

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity: TUNA is trained on synthetic mocks and evaluated against external reference masks; the self-cited simulation pipeline is minor and non-load-bearing, and the turbulence-omission gap is a correctness risk, not a circular step.

  1. other [Section 3.2 (Training Data); Section 4.1 (Planck Catalog from LoTSS-DR2)]
    "Images have been produced to closely resemble actual LOFAR observations, following the methodology outlined in Gheller & Vazza (2022) and Stuardi et al. (2024). ... we did not account for additional radio emission generated by the reacceleration by turbulence on relativistic electrons, which likely has a key role in the formation of radio halos, bridges, or megahalos. However, the morphology and emissivity of diffuse radio sources can loosely resemble even those of radio halos."

    Candidate circular link: the training set's realism premise (loosely resemble actual LOFAR observations) is inherited from the co-authors' own simulation pipeline (Gheller & Vazza 2022; Stuardi et al. 2024) and asserted only qualitatively, and the real-data ground truth is the source-subtracted uv-tapered pipeline (Botteon et al. 2022) TUNA claims to supersede. On inspection this does not reduce by construction: TUNA is trained only on mock images with a simple brightness-threshold ground truth, never on the Botteon products; the reference masks are an external, independently produced benchmark; the headline metrics (IoU 0.43, recall 0.61) are measured overlaps, not fitted targets; and the 4-6 times coarser claim is the empirically located IoU peak at 20-40 arcsec, not an identity.

full rationale

The claimed derivation chain is: (1) mock LOFAR-like images are synthesized from cosmological MHD simulations using the co-authors' prior pipeline (Gheller & Vazza 2022; Stuardi et al. 2024), with synchrotron emission modeled by diffusive shock acceleration only (Hoeft & Bruggen 2007); (2) TUNA (TransUNet with ResNet-50 and ImageNet21k pretraining) is trained on these mocks with a brightness-threshold ground truth; (3) the trained network is applied without retraining to real LoTSS-DR2/PSZ2 images; (4) predictions are scored against 3-sigma masks of the source-subtracted, uv-tapered images of Botteon et al. (2022), and against previously published detections (A399-A401 ridge, A1758 bridge, the four megahalos of Cuciti et al. 2022). No parameter is fitted to real data, so pattern 2 (fitted input called prediction) does not apply; no equation defines its own output (pattern 1); no uniqueness theorem is imported (pattern 4); no prior ansatz is smuggled via citation (pattern 5), because the shock-only emission assumption is stated openly with the mitigation 'loosely resemble'; and no known result is renamed (pattern 6). The only near-link is the self-citation of the training-simulation methodology (Section 3.2); it is minor and non-load-bearing because generalization is independently falsifiable against the external Botteon benchmark, and the earlier R-UNet work was itself validated on real LoTSS data. The acknowledged turbulence-omission and the 'proxy for ground truth' approximation (Sections 3.2 and 4.1) are domain-gap and correctness risks, not circularities. Verdict: no significant circularity; score 2 reflects the minor self-citation only.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claim depends on two domain assumptions (simulated shock-only emission is a valid training proxy; 3 sigma masks from reprocessed images are a valid ground truth) and on a small set of hand-set thresholds and hyperparameters. No new physical entities are introduced. The architecture (TransUNet) and the simulation pipeline are taken from prior work by the same group and others.

free parameters (7)
  • Ground truth surface brightness threshold = 1e-8 Jy/pixel
    Section 3.2: converts simulated sky images to binary masks. Hand-set to match LOFAR sensitivity; defines what the network learns as 'diffuse emission'.
  • Input normalization range = 1e-8 to 1e-2 Jy (logarithmic)
    Section 4: real and simulated images are log-scaled and clipped to this range before inference/training. Hand-chosen and applied uniformly.
  • Confidence threshold for predictions = 0.5 default; 0.9 for A399-A401 and ZwCl0634.1+4750
    Sections 4.2-4.3: threshold varied per target, higher when input resolution is below the training resolution. Post hoc choice affects reported masks and metrics.
  • Class weight w1 in loss = 1 (values 1, 2, 10 tested)
    Section 3.3: weighting the minority class did not improve results, so the weight was left at 1. A free parameter that proved ineffective.
  • Training hyperparameters (learning rate, batch size, tile size) = 0.005, 24, 512
    Section 3.3 and Appendix A: selected by grid search based on validation IoU/precision/recall. Not derived from theory.
  • Training epochs = 200
    Section 3.3: chosen because Class 1 losses kept decreasing until 200 epochs.
  • Reference mask threshold for real-data evaluation = 3 sigma
    Section 4.1: pixels above 3 sigma in 100 kpc/50 kpc source-subtracted tapered images define the proxy ground truth. Affects all real-data metrics.
assumptions (4)
  • domain assumption The Hoeft & Brüggen (2007) shock acceleration model adequately approximates the radio emission of real diffuse cluster sources, despite the omission of turbulent reacceleration.
    Section 3.2 states the emission is calculated from shocks only; the authors acknowledge turbulence is likely key for halos, bridges, and megahalos, yet assume the simulated morphology is close enough to train a generalizable network.
  • domain assumption Pixels with surface brightness above 3 sigma in the 100 kpc/50 kpc source-subtracted tapered images represent the true spatial extent of diffuse emission.
    Section 4.1 uses this as the evaluation ground truth for real data; if residual artifacts or missing faint emission bias this mask, the reported scores do not measure true performance.
  • domain assumption Imaging artifacts in mock LOFAR observations reproduce the appearance of real calibration and deconvolution artifacts closely enough that a network trained on mock data transfers to real survey images.
    Implicit in Sections 3.2 and 4: the training images are produced with WSClean 'predict' and LoSiTo noise, and the network is applied without retraining to LoTSS data.
  • standard math The TransUNet architecture, ImageNet-pretrained ResNet-50/ViT encoder, and standard training procedures are valid building blocks for this segmentation task.
    Section 3.1; no need to re-derive the Transformer or U-Net.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mapping Diffuse Radio Sources Using TUNA: A Transformer-Based Deep Learning Approach." pith.science (2026). https://pith.science/paper/JKQVQNEM

@misc{pith2026250711320,
  author       = {Pith},
  title        = {Pith review of: Mapping Diffuse Radio Sources Using TUNA: A Transformer-Based Deep Learning Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JKQVQNEM}},
  note         = {Machine review of arXiv:2507.11320}
}
read the original abstract

Vision Transformers are used via a customized TransUNet architecture, which is a hybrid model combining Transformers into a U-Net backbone, to achieve precise, automated, and fast segmentation of radio astronomy data affected by calibration and imaging artifacts, addressing the identification of faint, diffuse radio sources. Trained on mock radio observations from numerical simulations, the network is applied to the LOFAR Two-meter Sky Survey data. It is then evaluated on key use cases, specifically megahalos and bridges between galaxy clusters, to assess its performance in targeting sources at different resolutions and at the sensitivity limits of the telescope. The network is capable of detecting low surface brightness radio emission without manual source subtraction or re-imaging. The results demonstrate its groundbreaking capability to identify sources that typically require reprocessing at resolutions 4-6 times lower than that of the input image, accurately capturing their morphology and ensuring detection completeness. This approach represents a significant advancement in accelerating discovery within the large datasets generated by next-generation radio telescopes.

Figures

Figures reproduced from arXiv: 2507.11320 by the authors.

Figure 1
Figure 1. Schematic overview of the TransUNet architecture where z𝑙 is the encoded image representation. The flattened hidden features produced by the Transformer encoder are reshaped back into a 2D spatial structure, making it compatible with the U-Net decoder. The resulting low-resolution feature map is progressively restored to the full resolution H×W to obtain the final segmentation prediction using a stack of bilinear up… view at source ↗
Figure 2
Figure 2. TUNA network validation losses as a function of the training epoch for the 6 ′′ dataset. The blue curves show the Cross Entropy (CE), the Dice and the Total (average of the previous two) losses for Class 1 pixels. The red curves Show the same losses but considering both Classes. Standard deviations are estimated over five different trainings of the network. equation: L = LClass0 + 𝑤1LClass1 (12) where 𝑤1 is the weig… view at source ↗
Figure 3
Figure 3. The two rows show examples of Clean images at 6 ′′ used as input for the network (left panels), sky maps with surface brightness exceeding 10−8 Jy/pixel (our ground truth, right panels), predictions from TUNA (mid-left panels), and predictions from R-UNet (mid-right panels), all shown at the 0.5 confidence level [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Surface brightness distribution for for the simulated dataset at 6”, considering pixels above 10−8 Jy/pixel, accounting for the pixels identified by TUNA and R-UNet This dataset, hereafter referred to as LoTSS-DR2/PSZ2, represents the most comprehensive study to date o…
Figure 5
Figure 5. Figure 5: Left panel: radio images used as input for the networks. Central panels: prediction from the networks at a confidence level ≥ 0.5 with blue and green contours representing the confidence level = 0.5 for the R-UNet and TUNA networks, respectively. Right panel: 100 kpc t…
Figure 6
Figure 6. Figure 6: Panel (a): performance scores of TUNA and R-UNet as a function of the average resolution of the 100 kpc source-subtracted tapered images. Panel (b): performance metrics of TUNA as a function of the average resolution of the reference images, including 100 kpc and 50 kp…
Figure 7
Figure 7. Figure 7: shows the surface brightness for pixels above 3𝜎, alongside the surface brightness identified by TUNA and R-UNet predictions that overlap this threshold mask. Above 10−2 Jy/beam, both networks detect the majority of source pixels. TUNA continues to identify a significa…
Figure 8
Figure 8. Figure 8: On the left, the A399-A401 ridge; on the right, the Abell 1758 bridge. Top panels: radio images used as input for TUNA. Central panels: TUNA predictions at 0.5 confidence level; cyan contours show the 0.9 confidence level for A399-A401 and 0.5 for A1758. Bottom panels:…
Figure 9
Figure 9. Figure 9: The LoTSS-DR2 pointing P128+37 enclosing the cluster Abell 697 (one of the four megahalos). The top row shows the high resolution (6 ′′) radio observation (left panel) and the inferred TUNA mask (right panel). The bottom row shows the low resolution (20′′) radio observ…
Figure 10
Figure 10. Figure 10: The four megahalos ZwCl 0634.1+4750, Abell 665, Abell 697 and Abell 2218. Left column: LOFAR 144 MHz radio images used as input for TUNA. Right column: low resolution, source subtracted images with 3𝜎 contours (white) and TUNA confidence contours (cyan) superimposed. …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 28 canonical work pages

  1. [1]

    Astropy Collaboration et al., 2013, @doi [ ] 10.1051/0004-6361/201322068 , http://adsabs.harvard.edu/abs/2013A

    doi:10.1051/0004-6361/201322068 2013

  2. [2]

    Botteon A., et al., 2020, Monthly Notices of the Royal Astronomical Society: Letters, 499, L11

    work page 2020

  3. [3]

    Botteon A., et al., 2022, Astronomy & Astrophysics, 660, A78

    work page 2022

  4. [4]

    S., 1995, in American Astronomical Society Meeting Abstracts

    Briggs D. S., 1995, in American Astronomical Society Meeting Abstracts. p. 112.02

    work page 1995

  5. [5]

    D., 2011, @doi [Journal of Astrophysics and Astronomy] 10.1007/s12036-011-9114-4 , http://ads.ari.uni-heidelberg.de/abs/2011JApA...32..577B 32, 577

    Brown S. D., 2011, @doi [Journal of Astrophysics and Astronomy] 10.1007/s12036-011-9114-4 , http://ads.ari.uni-heidelberg.de/abs/2011JApA...32..577B 32, 577

    work page doi:10.1007/s12036-011-9114-4 2011

  6. [6]

    Brunetti G., 2011, @doi [Journal of Astrophysics and Astronomy] 10.1007/s12036-011-9103-7 , https://ui.adsabs.harvard.edu/abs/2011JApA...32..437B 32, 437

    work page doi:10.1007/s12036-011-9103-7 2011

  7. [7]

    W., 2014, International Journal of Modern Physics D, 23, 1430007

    Brunetti G., Jones T. W., 2014, International Journal of Modern Physics D, 23, 1430007

    work page 2014

  8. [8]

    Brunetti G., Vazza F., 2020, @doi [ ] 10.1103/PhysRevLett.124.051101 , https://ui.adsabs.harvard.edu/abs/2020PhRvL.124e1101B 124, 051101

    doi:10.1103/physrevlett.124.051101 2020

Show all 55 references
  1. [9]

    L., et al., 2014, @doi [ ] 10.1088/0067-0049/211/2/19 , http://adsabs.harvard.edu/abs/2014ApJS..211...19B 211, 19

    Bryan G. L., et al., 2014, @doi [ ] 10.1088/0067-0049/211/2/19 , http://adsabs.harvard.edu/abs/2014ApJS..211...19B 211, 19

    2014 doi

  2. [10]

    K., Bekki K., Groves B

    Cavanagh M. K., Bekki K., Groves B. A., 2021, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stab1552 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.506..659C 506, 659

    2021 doi

  3. [11]

    Chen J., et al., 2021, arXiv preprint arXiv:2102.04306

    2021 arXiv

  4. [12]

    Cuciti V., et al., 2022, Nature, 609, 911

    2022

  5. [13]

    arXiv:2412.19437

    DeepSeek-AI et al., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2412.19437 , https://ui.adsabs.harvard.edu/abs/2024arXiv241219437D p. arXiv:2412.19437

    2024 doi arXiv

  6. [14]

    Association for Computational Linguistics, Minneapolis, Minnesota, pp 4171--4186, @doi 10.18653/v1/N19-1423 , https://aclanthology.org/N19-1423/

    Devlin J., Chang M.-W., Lee K., Toutanova K., 2019, in Burstein J., Doran C., Solorio T., eds, Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). Ass...

    2019 doi

  7. [15]

    Dosovitskiy A., 2020, arXiv preprint arXiv:2010.11929

    2020 arXiv

  8. [16]

    https://openreview.net/forum?id=YicbFdNTTy

    Dosovitskiy A., et al., 2021, in International Conference on Learning Representations. https://openreview.net/forum?id=YicbFdNTTy

    2021

  9. [17]

    Feretti L., Giovannini G., Govoni F., Murgia M., 2012, The Astronomy and Astrophysics Review, 20, 1

    2012

  10. [18]

    arXiv:2312.11805

    Gemini Team et al., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2312.11805 , https://ui.adsabs.harvard.edu/abs/2023arXiv231211805G p. arXiv:2312.11805

    2023 doi arXiv

  11. [19]

    Gheller C., Vazza F., 2022, @doi [ ] 10.1093/mnras/stab3044 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.509..990G 509, 990

    2022 doi

  12. [20]

    Gheller C., Vazza F., Bonafede A., 2018, @doi [ ] 10.1093/mnras/sty2102 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.480.3749G 480, 3749

    2018 doi

  13. [21]

    Govoni F., et al., 2019, Science, 364, 981

    2019

  14. [22]

    pp 770--778

    He K., Zhang X., Ren S., Sun J., 2016, in Proceedings of the IEEE conference on computer vision and pattern recognition. pp 770--778

    2016

  15. [23]

    Hodgson T., Vazza F., Johnston-Hollitt M., McKinley B., 2021, @doi [ ] 10.1017/pasa.2021.32 , https://ui.adsabs.harvard.edu/abs/2021PASA...38...47H 38, e047

    2021 doi

  16. [24]

    Hoeft M., Br \"u ggen M., 2007, @doi [ ] 10.1111/j.1365-2966.2006.11111.x , http://adsabs.harvard.edu/abs/2007MNRAS.375...77H 375, 77

    2007

  17. [25]

    Hotan A., et al., 2021, Publications of the Astronomical Society of Australia, 38, e009

    2021

  18. [26]

    Jonas J., Team M., 2016, MeerKAT Science: On the Pathway to the SKA, p. 1

    2016

  19. [27]

    Jones A., et al., 2023, @doi [ ] 10.1051/0004-6361/202245102 , https://ui.adsabs.harvard.edu/abs/2023A&A...680A..31J 680, A31

    2023 doi

  20. [28]

    pp 1137--1143

    Kohavi R., 1995, in Proceedings of the 14th International Joint Conference on Artificial Intelligence (IJCAI). pp 1137--1143

    1995

  21. [29]

    J., Nagai D., Finner K., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2311.06340 , https://ui.adsabs.harvard.edu/abs/2023arXiv231106340L p

    Lee W., Pillepich A., ZuHone J., Nelson D., Jee M. J., Nagai D., Finner K., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2311.06340 , https://ui.adsabs.harvard.edu/abs/2023arXiv231106340L p. arXiv:2311.06340

    2023 doi arXiv

  22. [30]

    Nishiwaki K., Brunetti G., Vazza F., Gheller C., 2024, @doi [ ] 10.3847/1538-4357/ad11ce , https://ui.adsabs.harvard.edu/abs/2024ApJ...961...15N 961, 15

    2024 doi

  23. [31]

    R., Smirnov O., 2017, @doi [MNRAS] 10.1093/mnras/stx1547 , 471, 301

    Offringa A. R., Smirnov O., 2017, @doi [MNRAS] 10.1093/mnras/stx1547 , 471, 301

    2017 doi

  24. [32]

    R., McKinley B., Hurley-Walker et al., 2014, @doi [MNRAS] 10.1093/mnras/stu1368 , 444, 606

    Offringa A. R., McKinley B., Hurley-Walker et al., 2014, @doi [MNRAS] 10.1093/mnras/stu1368 , 444, 606

    2014 doi

  25. [33]

    A., Jubelgas M., 2006, @doi [ ] 10.1111/j.1365-2966.2005.09953.x , http://adsabs.harvard.edu/abs/2006MNRAS.367..113P 367, 113

    Pfrommer C., Springel V., En lin T. A., Jubelgas M., 2006, @doi [ ] 10.1111/j.1365-2966.2005.09953.x , http://adsabs.harvard.edu/abs/2006MNRAS.367..113P 367, 113

    2006

  26. [34]

    V., et al., 2024, @doi [ ] 10.1051/0004-6361/202451529 , https://ui.adsabs.harvard.edu/abs/2024A&A...691A..99P 691, A99

    Pignataro G. V., et al., 2024, @doi [ ] 10.1051/0004-6361/202451529 , https://ui.adsabs.harvard.edu/abs/2024A&A...691A..99P 691, A99

    2024 doi

  27. [35]

    Radford A., Narasimhan K., Salimans T., Sutskever I., 2018, OpenAI Technical Report

    2018

  28. [36]

    Radovic A., et al., 2018, Nature, 560, 41

    2018

  29. [37]

    Rajpurohit K., et al., 2024, @doi [ ] 10.3847/1538-4357/ad29fa , https://ui.adsabs.harvard.edu/abs/2024ApJ...966...38R 966, 38

    2024 doi

  30. [38]

    Rolnick D., et al., 2022, ACM Computing Surveys (CSUR), 55, 1

    2022

  31. [39]

    Ronneberger O., Fischer P., Brox T., 2015, Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015

    2015

  32. [40]

    W., 2003, @doi [ ] 10.1086/376723 , http://adsabs.harvard.edu/abs/2003ApJ...593..599R 593, 599

    Ryu D., Kang H., Hallman E., Jones T. W., 2003, @doi [ ] 10.1086/376723 , http://adsabs.harvard.edu/abs/2003ApJ...593..599R 593, 599

    2003 doi

  33. [41]

    Ryu D., Kang H., Cho J., Das S., 2008, @doi [Science] 10.1126/science.1154923 , https://ui.adsabs.harvard.edu/abs/2008Sci...320..909R 320, 909

    2008 doi

  34. [42]

    Sanvitale N., Gheller C., Bowman E., 2022, @doi [Granular Matter] 10.1007/s10035-022-01222-w , 24

    2022 doi

  35. [43]

    Shimwell T., et al., 2022a, Astronomy & astrophysics, 659, A1

  36. [44]

    W., et al., 2022b, @doi [ ] 10.1051/0004-6361/202142484 , https://ui.adsabs.harvard.edu/abs/2022A&A...659A...1S 659, A1

    Shimwell T. W., et al., 2022b, @doi [ ] 10.1051/0004-6361/202142484 , https://ui.adsabs.harvard.edu/abs/2022A&A...659A...1S 659, A1

    doi

  37. [45]

    Stuardi C., Gheller C., Vazza F., Botteon A., 2024, @doi [ ] 10.1093/mnras/stae2014 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.533.3194S 533, 3194

    2024 doi

  38. [46]

    J., et al., 2013, Publications of the Astronomical Society of Australia, 30, e007

    Tingay S. J., et al., 2013, Publications of the Astronomical Society of Australia, 30, e007

    2013

  39. [47]

    Vamathevan J., et al., 2019, Nature reviews Drug discovery, 18, 463

    2019

  40. [48]

    Van Weeren R., de Gasperin F., Akamatsu H., Br \"u ggen M., Feretti L., Kang H., Stroe A., Zandanel F., 2019, Space Science Reviews, 215, 1

    2019

  41. [49]

    N., Kaiser L., Polosukhin I., 2017, arXiv preprint arXiv:1706.03762, 10, S0140525X16001837

    Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A. N., Kaiser L., Polosukhin I., 2017, arXiv preprint arXiv:1706.03762, 10, S0140525X16001837

    2017 arXiv

  42. [50]

    P., Br \"u ggen M., 2021, @doi [ ] 10.1093/mnras/staa3532 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.500.5350V 500, 5350

    Vazza F., Paoletti D., Banfi S., Finelli F., Gheller C., O'Sullivan S. P., Br \"u ggen M., 2021, @doi [ ] 10.1093/mnras/staa3532 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.500.5350V 500, 5350

    2021 doi

  43. [51]

    arXiv:2501.19041

    Vazza F., Gheller C., Zanetti F., Tsizh M., Carretti E., Mtchedlidze S., Brueggen M., 2025, @doi [arXiv e-prints] 10.48550/arXiv.2501.19041 , https://ui.adsabs.harvard.edu/abs/2025arXiv250119041V p. arXiv:2501.19041

    2025 doi arXiv

  44. [52]

    C., Greisen E

    Wells D. C., Greisen E. W., Harten R. H., 1981, , https://ui.adsabs.harvard.edu/abs/1981A&AS...44..363W 44, 363

    1981

  45. [53]

    Wittor D., 2023, Universe, 9, 319

    2023

  46. [54]

    P., et al., 2013, Astronomy & astrophysics, 556, A2

    van Haarlem M. P., et al., 2013, Astronomy & astrophysics, 556, A2

    2013

  47. [55]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.stat...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.