Pith. sign in

REVIEW 4 major objections 5 minor 58 references

Restoration of contaminated data in an Intensity Mapping survey using deep neural networks

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A deep neural network that inpaints RFI-contaminated pixels before foreground removal lowers the residual noise in 21-cm intensity mapping and moves the measured angular power spectrum closer to the true neutral-hydrogen signal.

desk verdict Uses LaMa to inpaint RFI-flagged 21-cm IM data; the idea is promising but the loss as written trains the network to reproduce the RFI, so the headline results are not yet trustworthy. read the letter →

arxiv 2506.06386 v1 pith:55JX5EL2 submitted 2025-06-05 eess.SP

Lin-Cheng Li , Jia-Yu Lin , Yuan-Gen Wang , Lister Staveley-Smith This is my paper
classification eess.SP
keywords 21-cmintensitymappingradiofrequencyinterferencedeepneuralnetworkrestorationimageinpaintingforegroundremovalangularpowerspectrumLaMa-dilated
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

Radio-frequency interference makes chunks of 21-cm intensity-mapping data unusable, and simply flagging those chunks throws away any chance of recovering the faint cosmological signal hidden underneath. This paper asks whether a deep neural network can fill in the flagged regions before foreground removal, so that the following cleaning steps see a complete data cube. Using real observations to train an inpainting network, it compares the residual root-mean-square after polynomial fitting, SVD, and FastICA with and without restoration, and finds that restoration lowers the residuals, most strongly when the contaminated fraction is large. In a mock observation with a known true HI signal, the restored data yield an angular power spectrum closer to the fiducial spectrum than the unrestored data do. The paper concludes that learned restoration can increase the signal-to-noise ratio available from contaminated intensity-mapping surveys.

What carries the argument

The central object is the LaMa-dilated inpainting network, a convolutional image-inpainting architecture whose residual blocks use dilated convolutions to combine context from different scales when filling masked regions. It is trained with the LaMa loss—a high-receptive-field perceptual loss plus adversarial, feature-matching, and gradient-penalty terms—on 256×256 patches of real observations with realistic RFI masks. The network's job is to replace masked pixels with an estimate of the underlying smooth foreground and HI signal before polynomial, SVD, or ICA foreground removal. A second load-bearing piece is the mock observation built with a public intensity-mapping simulation, which supplies a known true HI angular power spectrum for comparing restored and unrestored analyses.

What would settle it

In the mock observation the clean no-RFI foreground and HI are known in every masked pixel, so one can directly measure the error between the network's restored values and that known truth; if the restored pixels are no closer to the truth than a simple interpolation of surrounding unmasked pixels, the reported RMS and power-spectrum improvements would reflect smoothing rather than genuine signal recovery.

Watch

Extended reading notes

Core claim

The central claim is that a DNN-based inpainting pipeline, the LaMa-dilated network, can restore RFI-contaminated pixels in 21-cm intensity mapping observations well enough that subsequent foreground-removal methods—polynomial fitting, SVD, and FastICA—recover a cleaner residual than they do from the flagged data. The paper further claims that in a mock observation generated by a public intensity-mapping simulation, the angular power spectrum of restored data lies closer to the true HI spectrum than that of unrestored data, especially at low multipoles. The restoration therefore is not just cosmetic; it is claimed to preserve large-scale astrophysical information that flagging would otherwise remove.

Load-bearing premise

The load-bearing assumption is that the network learns to fill masked regions with the true underlying foreground-plus-HI signal rather than reproducing the RFI that was in those pixels, even though the training loss is written against data that still contain the contamination in the masked areas.

Editorial extensions

If this is right

  • Surveys can retain more flagged data: restoring masked pixels before foreground removal lowers residual noise instead of discarding those pixels.
  • The improvement grows with the contaminated fraction, so the method is most valuable in the data that current pipelines handle worst.
  • The low-multipole part of the angular power spectrum shifts toward the true HI spectrum, meaning the large-scale cosmological signal is better preserved.
  • The restoration step is method-agnostic: it improves polynomial fitting, SVD, and ICA foreground removal alike without changing those algorithms.

Reading between the lines

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

  • A natural next test is to train the same network on simulated data with a clean foreground target in masked pixels; measuring per-pixel error there would separate genuine signal recovery from smoothing.
  • Because the selection metric favors correlation with the foreground, the ranking of network architectures could change if the metric also penalized spurious HI-like structure in restored regions.
  • For future large surveys, the practical payoff will depend on whether the restored pixels are unbiased; cross-correlation of restored 21-cm data with optical galaxy surveys is the direct observational check.
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

4 major / 5 minor

Summary. The paper proposes using a LaMa-dilated inpainting neural network to restore RFI-masked pixels in 21-cm intensity mapping data from Parkes P913, and then evaluates the effect of this restoration on foreground removal by polynomial fitting, SVD, and ICA. The authors report lower RMS residuals after restoration (Sections 4.1-4.3), select LaMa-dilated via a Cm/Cu metric (Section 3.3), and report in a CRIME-based mock observation that angular power spectra are closer to the true HI signal (Section 6). The central claim is that DNN restoration before foreground removal improves data quality and preserves the large-scale HI signal.

Significance. If the pipeline works as claimed, restoring flagged pixels before foreground subtraction would be a practically useful pre-processing step for 21-cm IM surveys, and the use of real Parkes data with realistic RFI masks is a valuable feature. The paper includes an ablation of loss terms, patch sizes, and metrics, and reports a mock-observation test; these are appropriate steps and go beyond a purely qualitative demonstration. However, the current manuscript does not establish that the network learns to remove rather than reproduce RFI, and the evaluation metrics are tied to the same foreground model used to demonstrate improvement. The scientific conclusions therefore rest on assumptions that are not yet verified.

major comments (4)
  1. [Sec. 2.2, Eqs. (1)-(5)] The loss functions are written against the full, contaminated data: Eq. (1) compares phi_HRF(x) and phi_HRF(x_hat) over the full image, and Eq. (3) uses a quantity called ground truth without stating that the masked regions are excluded from the loss or that the target in masked regions is a cleaned foreground model. Because the masked regions are precisely where RFI is present, a network trained with this objective can minimize the loss by reproducing the contamination rather than removing it. The RMS reductions in Sections 4.1-4.3 and the angular power spectrum shifts in Section 6 would then not establish genuine restoration of the foreground+HI signal. Please define the ground truth concretely, restrict the loss to unmasked pixels if that is the intention, and retrain and re-evaluate with the corrected objective before interpreting the downstream results.
  2. [Sec. 3.3 and Sec. 5.3] The Cm/Cu metric used for model selection compares predicted data with "foregrounds", but the operational definition of Cm and Cu is never given, and Section 5.3 reveals that foregrounds extracted by the polynomial-fitting pipeline of Li et al. (2021) are used as ground truth in the SSIM/PSNR evaluation. If the same foreground model underlies Cm/Cu, then selecting LaMa-dilated because it best matches those polynomial-fitted foregrounds and then demonstrating improvement over polynomial fitting in Section 4.1 is circular. Please specify how Cm and Cu are computed and evaluate the restoration against a target that is independent of the foreground-removal method used as the baseline.
  3. [Sec. 4] The claim in Section 4 that "a lower root-mean-square (RMS) level of data invariably means cleaner observational data with less contamination" is not justified. Filling masked regions with smooth model predictions can lower the residual RMS after polynomial, SVD, or ICA subtraction even when the HI signal is distorted, so RMS alone cannot certify restoration quality. The mock-observation test in Section 6 is the right idea but currently lacks a quantitative measure of the distance to the fiducial C_l and, in the polynomial-fitting panel, shows a restored curve that lies above the HI curve rather than unambiguously matching it. Please add a signal-preservation diagnostic and quantify the agreement with the fiducial power spectrum, for example with error bars or a chi-square-like statistic.
  4. [Sec. 6.4, Fig. 11] Figure 11 reports angular power spectra for a single CRIME realization without error bars, and the conclusion that the restored curves are "closer" to the HI fiducial is drawn from visual inspection in log space over several orders of magnitude. The simulation also omits observational noise and beam effects and downsamples the frequency axis by averaging 20 channels; these simplifications should be stated as limitations and their effect on the conclusion discussed. Without uncertainties, the central claim that restoration moves the measured C_l toward the true signal is not quantitatively supported.
minor comments (5)
  1. [Sec. 3.2] The sentence "the channels with mean data values outside of the range ... e considered as the contaminated by RFI and are masked" contains a typo: "e considered" should be "are considered", and "the contaminated" should be "contaminated".
  2. [Sec. 3.2] The optimization setup is underspecified: the optimizer, learning rate, learning-rate schedule, and convergence criterion are not reported. Please include these details or a reference to the exact LaMa configuration used.
  3. [Sec. 2.2] The relation between the notation (x, x_hat) in Eq. (1) and (I_hat, I_tilde) in Eqs. (3)-(4) is unclear; please define all symbols consistently.
  4. [Fig. 9 caption] The caption states that the panels show M128, M256, and M256; the third panel should presumably be M512.
  5. [Sec. 6.4, Fig. 11] The label "polinomial fitting" in Figure 11 should be "polynomial fitting".

Circularity Check

1 steps flagged · score 4.0 of 10

Model selection via Cm/Cu and real-data RMS evaluation both rely on the same polynomial foreground model from Li et al. (2021), creating a selection bias; the independent CRIME simulation supports the central claim.

  1. fitted input called prediction [Sections 3.3, 5.3, and 4.1]
    "Considering that our restoration aims to recover the foregrounds as much as possible, we compute the ratio of Cm/Cu as the performance metric... We therefore use the LaMa-dilated model to perform further analysis in the following sections. ... In the work of Li et al. (2021), the foregrounds are fitted with polynomials and subtracted from the observational data for each integration cycle. We take the foregrounds extracted in the pipeline of Li et al. (2021) as the ground-truth of our prediction..."

    The Cm/Cu metric used to select among LaMa architectures compares predictions in masked regions to foregrounds extracted by the polynomial fitting pipeline of Li et al. (2021), a paper by overlapping authors. The selected model is therefore chosen to correlate with that specific polynomial foreground model. The real-data RMS improvement is then measured after applying the same polynomial foreground removal technique. This creates a selection bias: the architecture that best matches the evaluation's ground-truth foreground is the one deemed successful, so the reported RMS reduction is partly a consequence of the selection criterion rather than an independent prediction.

full rationale

The paper's most significant potential circularity lies in the model selection and real-data evaluation. The authors choose the LaMa-dilated architecture using Cm/Cu, a metric that measures correlation between predicted masked regions and foregrounds derived from the polynomial fitting pipeline of their own previous work (Li et al. 2021). They then demonstrate RMS improvement after foreground removal using the same polynomial fitting method. This alignment means the chosen model is selected for consistency with the reference foreground model, so the real-data RMS gains are not fully independent of the selection metric. However, the paper includes a mock-observation test using CRIME, which has a known HI signal and does not depend on the real-data foreground model. In that test, the angular power spectrum after restoration moves closer to the true HI spectrum, providing independent support for the claim that restoration improves signal recovery. The training loss in Section 2.2 targets the contaminated data x, which is a methodological concern but not a construction-level circularity, since the network output is not defined as the foreground model. Given the independent simulation, the central claim retains external validity, but the real-data analysis contains a partial selection-driven circularity.

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

The central claim rests on the assumption that inpainting can recover the clean foreground-plus-signal in masked regions from context, that the polynomial foreground model from prior work is a valid proxy for truth, and that the CRIME simulation is representative. No new physical entities are introduced. The free parameters are hyperparameters and data-selection thresholds that are chosen by hand rather than fitted to the target result, but they affect the training and evaluation.

free parameters (4)
  • Loss weights (kappa, alpha, beta, gamma) = kappa=10, alpha=30, beta=100, gamma=0.001
    Hand-chosen weights for the adversarial, perceptual, discriminator-feature, and gradient-penalty losses; standard LaMa defaults. They affect training dynamics but not the final claim directly.
  • Masked-fraction threshold = 40 percent
    Patches with more than 40 percent masked pixels are excluded from the training set. This choice shapes the training distribution and may affect generalization to highly contaminated data.
  • Number of SVD modes removed = 1 to 5 (varied)
    The number of modes subtracted in SVD foreground removal is varied by hand; results are shown for each value.
  • Number of ICA components = 4
    Four independent components are separated and subtracted in FastICA; chosen by hand without a stated criterion.
assumptions (4)
  • domain assumption Contaminated regions can be faithfully restored from surrounding unmasked data by inpainting.
    The entire method assumes the masked RFI regions carry no unique information and that the underlying foregrounds and HI signal are spatially and spectrally coherent enough to be predicted from context. Invoked throughout Sections 2 and 3.
  • domain assumption The foregrounds fitted by the polynomial pipeline of Li et al. (2021) are a valid ground truth for evaluating restoration quality.
    Used in Section 3.3 for the Cm/Cu metric and in Section 5.3 for SSIM and PSNR. The metric and model selection rely on this prior extraction as the target for masked regions.
  • domain assumption The CRIME simulation with foreground parameters from Santos et al. (2005) is representative of real 21-cm IM observations at 800 to 820 MHz.
    Used in Section 6 to establish the mock observation and the fiducial HI angular power spectrum; the conclusions about power-spectrum recovery depend on this model.
  • ad hoc to paper The neural network trained on real data learns to remove RFI rather than reproduce it, despite the loss being written against the contaminated data.
    The paper does not specify a masked-region target in the loss. If the network minimizes Eq. (1) against the full data x, it would be trained to reconstruct the RFI. The claimed restoration behavior therefore rests on an unstated assumption about the training objective.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Restoration of contaminated data in an Intensity Mapping survey using deep neural networks." pith.science (2026). https://pith.science/paper/55JX5EL2

@misc{pith2026250606386,
  author       = {Pith},
  title        = {Pith review of: Restoration of contaminated data in an Intensity Mapping survey using deep neural networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/55JX5EL2}},
  note         = {Machine review of arXiv:2506.06386}
}
read the original abstract

21-cm Intensity Mapping (IM) is a promising approach to detecting information about the large-scale structure beyond the local universe. One of the biggest challenges for an IM observation is the foreground removal procedure. In this paper, we attempt to conduct the restoration of contaminated data in an IM experiment with a Deep Neural Network (DNN). To investigate the impact of such data restoration, we compare the root-mean-square (RMS) of data with and without restoration after foreground removal using polynomial fitting, singular value decomposition, and independent component analysis, respectively. We find that the DNN-based pipeline performs well in lowering the RMS level of data, especially for data with large contaminated fractions. Furthermore, we investigate the impact of the restoration on the large-scale 21-cm signal in the simulation generated by CRIME. Simulation results show that the angular power spectrum curves from data with restoration are closer to the real one. Our work demonstrates that the DNN-based data restoration approach significantly increases the signal-to-noise ratio compared with conventional ones, achieving excellent potential for IM observations.

Figures

Figures reproduced from arXiv: 2506.06386 by the authors.

Figure 1
Figure 1. The architecture of LaMa-dilated neural network. 𝐸𝑥 ∥∇𝐷(𝑥) ∥2 , and the perceptual loss on the features of discriminator network (𝐿Disc) which is used to stabilize training. As a result, the final loss function for our model can be expressed as 𝐿final = 𝜅𝐿Adv + 𝛼𝐿HRFPL + 𝛽𝐿Disc + 𝛾𝑅1, (5) where 𝜅 = 10, 𝛼 = 30, 𝛽 = 100, and 𝛾 = 0.001 are hyper-parameters. 3 DATA 3.1 Observation The observational data were obtained fr… view at source ↗
Figure 2
Figure 2. An example of the randomly extracted area (red square) from the original data. is the dimension of each sample, 𝑛1 and 𝑛2 are the dimension in cycles and the dimension in channels of the data, respectively. In total, we extracted 23,806 patches of data, from which 5,000 patches are set as the validation set, and the rest are used as the training set. During the training, all models are trained for more than 3×105 it… view at source ↗
Figure 3
Figure 3. An example of the restoration results generated from different models. The leftmost panel is the original data with masks (yellow areas). The other panels are the restoration results generated from big-LaMa, LaMa-Fourier, LaMa-regular, and LaMa-dilated, respectively, indicated by the titles of each panel. 0% 10% 20% 30% 40% 50% Contaminated fraction 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 C m / C u LaMa-dilated LaMa-regular… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: The median 𝐶m/𝐶u as the measurement of model performance for different models. Blue, orange, green, and red data points are the results for LaMa-dilated, LaMa-regular, LaMa-Fourier, and big-LaMa, respectively. Error bars represent the ranges from the 25th to the 75th p…
Figure 7
Figure 7. Figure 7: Top panel: the RMS levels after foreground removal with FastICA in different datasets. The error bars represent the ranges from the 25th to the 75th percentiles in each interval. Bottom panel: the normalized offsets to the values of the dataset without restoration. our…
Figure 8
Figure 8. Figure 8: The values of 𝐶m/𝐶u for models using different loss functions. The blue, yellow, and green points are the results of the LaMa-dilated model, LaMa-dilated model without 𝐿disc, and without 𝐿HRFPL, respectively. The error bars represent the ranges from the 25th to the 75t…
Figure 9
Figure 9. Figure 9: An example of restoration results generated by models trained with samples of different dimensions. Panels from left to right are the original data with masks (yellow areas), the restoration results generated from models trained with samples of dimension 128×128, 256×2…
Figure 10
Figure 10. Figure 10: The total intensity of the HealPix map generated by CRIME in the spectral channel of 810 MHz, in unit of mK. In total, 11500 data patches are extracted from the mock observa￾tional data. Among them, 2000 data are set as validation set and the rest are set as the train…
Figure 11
Figure 11. Figure 11: Angular power spectra of different data after foreground removal using different methods. Blue, and green curves represent the results for mock observational data with and without restoration, respectively. For comparison, the angular power spectrum of real HI signal …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 8 canonical work pages

  1. [1]

    G., Santos M

    Alonso D., Ferreira P. G., Santos M. G., 2014, @doi [ ] 10.1093/mnras/stu1666 , https://ui.adsabs.harvard.edu/abs/2014MNRAS.444.3183A 444, 3183

    doi:10.1093/mnras/stu1666 2014

  2. [2]

    G., Santos M

    Alonso D., Bull P., Ferreira P. G., Santos M. G., 2015, @doi [ ] 10.1093/mnras/stu2474 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.447..400A 447, 400

    doi:10.1093/mnras/stu2474 2015

  3. [3]

    J., et al., 2018, @doi [ ] 10.1093/mnras/sty346 , http://adsabs.harvard.edu/abs/2018MNRAS.476.3382A 476, 3382

    Anderson C. J., et al., 2018, @doi [ ] 10.1093/mnras/sty346 , http://adsabs.harvard.edu/abs/2018MNRAS.476.3382A 476, 3382

    doi:10.1093/mnras/sty346 2018

  4. [5]

    arXiv:2309.00710

    Berti M., Spinelli M., Viel M., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2309.00710 , https://ui.adsabs.harvard.edu/abs/2023arXiv230900710B p. arXiv:2309.00710

    arXiv doi:10.48550/arxiv.2309.00710 2023

  5. [6]

    K., Iliev I

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

    doi:10.1093/mnras/stab1518 2021

  6. [7]

    A., et al., 2015, @doi [ ] 10.1093/mnras/stv2153 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.454.3240B 454, 3240

    Bigot-Sazy M. A., et al., 2015, @doi [ ] 10.1093/mnras/stv2153 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.454.3240B 454, 3240

    doi:10.1093/mnras/stv2153 2015

  7. [9]

    arXiv:2202.01242

    CHIME Collaboration et al., 2022, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2022arXiv220201242C p. arXiv:2202.01242

    arXiv 2022

  8. [10]

    Chakraborty A., Datta A., Mazumder A., 2022, @doi [ ] 10.3847/1538-4357/ac5cc5 , https://ui.adsabs.harvard.edu/abs/2022ApJ...929..104C 929, 104

    doi:10.3847/1538-4357/ac5cc5 2022

Show all 58 references
  1. [11]

    B., 2010, @doi [ ] 10.1038/nature09187 , http://adsabs.harvard.edu/abs/2010Natur.466..463C 466, 463

    Chang T.-C., Pen U.-L., Bandura K., Peterson J. B., 2010, @doi [ ] 10.1038/nature09187 , http://adsabs.harvard.edu/abs/2010Natur.466..463C 466, 463

    2010 doi

  2. [13]

    Chen X., 2015, in IAU General Assembly. p. 2252187

    2015

  3. [14]

    Cimatti A., Scaramella R., 2012, Memorie della Societa Astronomica Italiana Supplementi, http://adsabs.harvard.edu/abs/2012MSAIS..19..314C 19, 314

    2012

  4. [15]

    Colless M., et al., 2001, @doi [ ] 10.1046/j.1365-8711.2001.04902.x , http://adsabs.harvard.edu/abs/2001MNRAS.328.1039C 328, 1039

    2001

  5. [16]

    arXiv:1611.00036

    DESI Collaboration et al., 2016, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2016arXiv161100036D p. arXiv:1611.00036

    2016 arXiv

  6. [17]

    S., White S

    Davis M., Efstathiou G., Frenk C. S., White S. D. M., 1985, @doi [ ] 10.1086/163168 , https://ui.adsabs.harvard.edu/abs/1985ApJ...292..371D 292, 371

    1985 doi

  7. [18]

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

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

    2017 doi

  8. [20]

    arXiv:2203.08056

    Dvorkin C., et al., 2022, @doi [arXiv e-prints] 10.48550/arXiv.2203.08056 , https://ui.adsabs.harvard.edu/abs/2022arXiv220308056D p. arXiv:2203.08056

    2022 doi arXiv

  9. [21]

    Gagnon-Hartman S., Cui Y., Liu A., Ravanbakhsh S., 2021, @doi [ ] 10.1093/mnras/stab1158 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.504.4716G 504, 4716

    2021 doi

  10. [22]

    Gao L.-Y., Li Y., Ni S., Zhang X., 2023, @doi [ ] 10.1093/mnras/stad2646 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.525.5278G 525, 5278

    2023 doi

  11. [23]

    R., 2012, @doi [ ] 10.1111/j.1365-2966.2012.20582.x , https://ui.adsabs.harvard.edu/abs/2012MNRAS.421.3570G 421, 3570

    Guha Sarkar T., Mitra S., Majumdar S., Choudhury T. R., 2012, @doi [ ] 10.1111/j.1365-2966.2012.20582.x , https://ui.adsabs.harvard.edu/abs/2012MNRAS.421.3570G 421, 3570

    2012

  12. [24]

    Hothi I., et al., 2021, @doi [ ] 10.1093/mnras/staa3446 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.500.2264H 500, 2264

    2021 doi

  13. [25]

    Hyvarinen A., 1999, @doi [IEEE Transactions on Neural Networks] 10.1109/72.761722 , 10, 626

    1999 doi

  14. [26]

    Ivezic Z., et al., 2008, preprint, http://adsabs.harvard.edu/abs/2008arXiv0805.2366I ( @eprint arXiv 0805.2366 )

    2008 arXiv

  15. [27]

    Jiang Y.-E., et al., 2023, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/accdc0 , https://ui.adsabs.harvard.edu/abs/2023RAA....23g5003J 23, 075003

    2023 doi

  16. [28]

    H., et al., 2004, @doi [ ] 10.1111/j.1365-2966.2004.08353.x , http://adsabs.harvard.edu/abs/2004MNRAS.355..747J 355, 747

    Jones D. H., et al., 2004, @doi [ ] 10.1111/j.1365-2966.2004.08353.x , http://adsabs.harvard.edu/abs/2004MNRAS.355..747J 355, 747

    2004

  17. [29]

    H., et al., 2009, @doi [ ] 10.1111/j.1365-2966.2009.15338.x , http://adsabs.harvard.edu/abs/2009MNRAS.399..683J 399, 683

    Jones D. H., et al., 2009, @doi [ ] 10.1111/j.1365-2966.2009.15338.x , http://adsabs.harvard.edu/abs/2009MNRAS.399..683J 399, 683

    2009

  18. [30]

    Li W., et al., 2019, Monthly Notices of the Royal Astronomical Society, 485, 2628

    2019

  19. [31]

    Li L.-C., Staveley-Smith L., Rhee J., 2021, @doi [RAA] 10.1088/1674-4527/21/2/30 , https://ui.adsabs.harvard.edu/abs/2021RAA....21...30L 21, 030

    2021 doi

  20. [32]

    Liu A., Tegmark M., 2011, @doi [ ] 10.1103/PhysRevD.83.103006 , http://adsabs.harvard.edu/abs/2011PhRvD..83j3006L 83, 103006

    2011 doi

  21. [33]

    Maino D., et al., 2002, @doi [ ] 10.1046/j.1365-8711.2002.05425.x , https://ui.adsabs.harvard.edu/abs/2002MNRAS.334...53M 334, 53

    2002

  22. [34]

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

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

    2021 doi

  23. [35]

    W., et al., 2013, @doi [ ] 10.1088/2041-8205/763/1/L20 , http://adsabs.harvard.edu/abs/2013ApJ...763L..20M 763, L20

    Masui K. W., et al., 2013, @doi [ ] 10.1088/2041-8205/763/1/L20 , http://adsabs.harvard.edu/abs/2013ApJ...763L..20M 763, L20

    2013 doi

  24. [36]

    Moriwaki K., Nishimichi T., Yoshida N., 2023, @doi [Reports on Progress in Physics] 10.1088/1361-6633/acd2ea , https://ui.adsabs.harvard.edu/abs/2023RPPh...86g6901M 86, 076901

    2023 doi

  25. [37]

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

    2022 doi

  26. [38]

    R., de Bruyn A

    Offringa A. R., de Bruyn A. G., Zaroubi S., Biehl M., 2010, @doi [arXiv e-prints] 10.48550/arXiv.1007.2089 , https://ui.adsabs.harvard.edu/abs/2010arXiv1007.2089O p. arXiv:1007.2089

    2010 doi arXiv

  27. [39]

    Paciga G., et al., 2011, @doi [ ] 10.1111/j.1365-2966.2011.18208.x , https://ui.adsabs.harvard.edu/abs/2011MNRAS.413.1174P 413, 1174

    2011

  28. [40]

    Paciga G., et al., 2013, @doi [ ] 10.1093/mnras/stt753 , https://ui.adsabs.harvard.edu/abs/2013MNRAS.433..639P 433, 639

    2013 doi

  29. [41]

    R., et al., 2010, @doi [ ] 10.1088/0004-6256/139/4/1468 , https://ui.adsabs.harvard.edu/abs/2010AJ....139.1468P 139, 1468

    Parsons A. R., et al., 2010, @doi [ ] 10.1088/0004-6256/139/4/1468 , https://ui.adsabs.harvard.edu/abs/2010AJ....139.1468P 139, 1468

    2010 doi

  30. [42]

    B., Chang T.-C., 2009, @doi [ ] 10.1111/j.1745-3933.2008.00581.x , http://adsabs.harvard.edu/abs/2009MNRAS.394L...6P 394, L6

    Pen U.-L., Staveley-Smith L., Peterson J. B., Chang T.-C., 2009, @doi [ ] 10.1111/j.1745-3933.2008.00581.x , http://adsabs.harvard.edu/abs/2009MNRAS.394L...6P 394, L6

    2009

  31. [43]

    G., Cooray A., Knox L., 2005, , 625, 575

    Santos M. G., Cooray A., Knox L., 2005, , 625, 575

    2005

  32. [44]

    F., et al., 2006, @doi [ ] 10.1086/498708 , https://ui.adsabs.harvard.edu/abs/2006AJ....131.1163S 131, 1163

    Skrutskie M. F., et al., 2006, @doi [ ] 10.1086/498708 , https://ui.adsabs.harvard.edu/abs/2006AJ....131.1163S 131, 1163

    2006 doi

  33. [45]

    Sun H., Deng H., Wang F., Mei Y., Xu T., Smirnov O., Deng L., Wei S., 2022, @doi [ ] 10.1093/mnras/stac570 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.512.2025S 512, 2025

    2022 doi

  34. [46]

    arXiv:2109.07161

    Suvorov R., et al., 2021, @doi [arXiv e-prints] 10.48550/arXiv.2109.07161 , https://ui.adsabs.harvard.edu/abs/2021arXiv210907161S p. arXiv:2109.07161

    2021 doi arXiv

  35. [47]

    R., et al., 2013, @doi [ ] 10.1093/mnrasl/slt074 , http://adsabs.harvard.edu/abs/2013MNRAS.434L..46S 434, L46

    Switzer E. R., et al., 2013, @doi [ ] 10.1093/mnrasl/slt074 , http://adsabs.harvard.edu/abs/2013MNRAS.434L..46S 434, L46

    2013 doi

  36. [48]

    Villaescusa-Navarro F., Alonso D., Viel M., 2017, @doi [ ] 10.1093/mnras/stw3224 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.466.2736V 466, 2736

    2017 doi

  37. [49]

    Visbal E., McQuinn M., 2021, in American Astronomical Society Meeting Abstracts. p. 310.06

    2021

  38. [50]

    G., Knox L., 2006, @doi [ ] 10.1086/506597 , http://adsabs.harvard.edu/abs/2006ApJ...650..529W 650, 529

    Wang X., Tegmark M., Santos M. G., Knox L., 2006, @doi [ ] 10.1086/506597 , http://adsabs.harvard.edu/abs/2006ApJ...650..529W 650, 529

    2006 doi

  39. [51]

    Wang J., et al., 2021, @doi [ ] 10.1093/mnras/stab1365 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.505.3698W 505, 3698

    2021 doi

  40. [52]

    B., et al., 2018, @doi [ ] 10.1017/pasa.2018.37 , https://ui.adsabs.harvard.edu/abs/2018PASA...35...33W 35, e033

    Wayth R. B., et al., 2018, @doi [ ] 10.1017/pasa.2018.37 , https://ui.adsabs.harvard.edu/abs/2018PASA...35...33W 35, e033

    2018 doi

  41. [53]

    B., Blake C., Shaw J

    Wolz L., Abdalla F. B., Blake C., Shaw J. R., Chapman E., Rawlings S., 2014, @doi [ ] 10.1093/mnras/stu792 , https://ui.adsabs.harvard.edu/abs/2014MNRAS.441.3271W 441, 3271

    2014 doi

  42. [54]

    L., et al., 2010, @doi [ ] 10.1088/0004-6256/140/6/1868 , https://ui.adsabs.harvard.edu/abs/2010AJ....140.1868W 140, 1868

    Wright E. L., et al., 2010, @doi [ ] 10.1088/0004-6256/140/6/1868 , https://ui.adsabs.harvard.edu/abs/2010AJ....140.1868W 140, 1868

    2010 doi

  43. [55]

    Yohana E., Ma Y.-Z., Li D., Chen X., Dai W.-M., 2021, @doi [ ] 10.1093/mnras/stab1197 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.504.5231Y 504, 5231

    2021 doi

  44. [56]

    G., et al., 2000, @doi [ ] 10.1086/301513 , http://adsabs.harvard.edu/abs/2000AJ....120.1579Y 120, 1579

    York D. G., et al., 2000, @doi [ ] 10.1086/301513 , http://adsabs.harvard.edu/abs/2000AJ....120.1579Y 120, 1579

    2000 doi

  45. [57]

    Yoshiura S., Shimabukuro H., Hasegawa K., Takahashi K., 2021, @doi [ ] 10.1093/mnras/stab1718 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.506..357Y 506, 357

    2021 doi

  46. [58]

    F., Karakci A., Korotkov A., Sutter P

    Zhang L., Bunn E. F., Karakci A., Korotkov A., Sutter P. M., Timbie P. T., Tucker G. S., Wandelt B. D., 2016, @doi [ ] 10.3847/0067-0049/222/1/3 , https://ui.adsabs.harvard.edu/abs/2016ApJS..222....3Z 222, 3

    2016 doi

  47. [59]

    Zhang M., Li Y., Zhang J.-F., Zhang X., 2023, @doi [ ] 10.1093/mnras/stad2033 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.524.2420Z 524, 2420

    2023 doi

  48. [60]

    Zhou X., Gong Y., Deng F., Zhang M., Yue B., Chen X., 2023, @doi [ ] 10.1093/mnras/stad563 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.521..278Z 521, 278

    2023 doi

  49. [61]

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

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

    2013 doi

  50. [62]

    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 7, 2026 · model on record in the stance chip above.