Pith. sign in

REVIEW 5 major objections 5 minor 46 references

Photometric Redshift Predictions with a Neural Network for DESI Quasars

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

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

Pith's one-line read Adding GALEX ultraviolet photometry to DESI and WISE bands improves neural-network photometric redshifts for quasars, cutting scatter and catastrophic outliers.

desk verdict The GALEX improvement is real in direction, but the headline numbers compare different training samples, so the magnitude is unproven and there are internal numeric inconsistencies. read the letter →

arxiv 2507.03260 v1 pith:7M6KI3QI submitted 2025-07-04 astro-ph.GA

Jeremy P. Moss , Stephen J. Curran , Yvette C. Perrott This is my paper
classification astro-ph.GA
keywords photometricredshiftsquasarsneuralnetworkGALEXultravioletphotometryDESIWISEmachinelearningactivegalacticnuclei
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 aims to show that a fully connected neural network can estimate photometric redshifts for DESI quasars accurately enough for surveys that cannot obtain spectra for every source, and that accuracy improves markedly when far- and near-ultraviolet fluxes from GALEX are added to the DESI optical and WISE near-infrared bands. On the 24,616 quasars with SDSS and GALEX photometry, the network reaches a correlation coefficient of r = 0.9187 with spectroscopic redshift and a normalised median absolute deviation of 0.197, a 29 per cent reduction in scatter relative to the optical–infrared baseline. The claimed reason is physical: QSO rest-frame UV features, including the Lyman break, shift through the observed bands with redshift, so the FUV-to-W2 wavelength lever arm breaks colour–redshift degeneracies that the narrower DESI/WISE baseline leaves open. Such photometric redshifts would let future wide-area surveys study quasar populations without spectroscopic follow-up of every source.

What carries the argument

The machinery is a three-layer fully connected neural network with 200 neurons per hidden layer, ReLU activations, MSE loss, the Adam optimiser, and early stopping, trained on 80 per cent of the sample over 100 random train/test splits with K-fold cross-validation. The physical mechanism carrying the argument is the wide wavelength baseline from GALEX FUV (λ ≈ 1575 Å) through DESI g, r, z to WISE W1, W2 (λ ≈ 4.6 μm): as redshift increases, rest-frame UV features such as the Lyman break at 1216 Å and the Mg II line at 2800 Å move through these bands, so the added UV bands break colour–redshift degeneracies that a purely optical–infrared set leaves open. The paper deliberately uses raw fluxes and magnitudes rather than colour indices, arguing that this avoids introducing extra feature correlations while retaining the most fundamental observational data.

What would settle it

Train the same neural network on the DxS sample twice, once with g, r, z, W1, W2 only and once with FUV and NUV added, holding all sources and hyperparameters fixed; if the correlation and normalised median absolute deviation do not improve by roughly the amount implied by Table 9b (NMAD from about 0.32 to about 0.27), or if the improvement does not persist when the model is applied to faint, non-SDSS-matched DESI quasars, then the headline gain is largely a sample-selection effect rather than a true benefit of ultraviolet photometry.

Watch

Extended reading notes

Core claim

The central claim is that adding GALEX FUV and NUV photometry to DESI g, r, z and WISE W1, W2 fluxes substantially improves photometric redshift predictions for quasars: for a fully connected three-layer neural network, the correlation with spectroscopic redshift rises from r = 0.8099 on the DESI-only sample to r = 0.9187 on the DESI+GALEX sample, normalised median absolute deviation falls from 0.278 to 0.197, and the maximum and mean absolute errors both improve. The neural network also outperforms a tuned k-nearest-neighbours model on the same features. The authors attribute the gain to the wavelength lever arm from FUV (rest-frame 1575 Å) to W2, which lets the model span the Lyman break and other rest-frame UV features as they redshift through the observed bands. A secondary claim is that the bimodality seen in the optical-only predictions, driven by separation in the g–r versus z–W1 colour plane, is reduced when UV bands are included, and that when DESI and SDSS spectroscopic redshifts disagree, the photometric prediction agrees with DESI for about 73 per cent of the 107 outliers.

Load-bearing premise

The conclusion that GALEX ultraviolet bands cause the improvement assumes that the gain measured on the smaller, brighter SDSS-matched subset would also appear on the full DESI sample; the paper itself notes in Section 4.2 that GALEX-detected quasars are already among the brighter, better-constrained objects.

Editorial extensions

If this is right

  • Photometric redshifts of quasars from a neural network are accurate enough that large surveys without full spectroscopic coverage can statistically use them for quasar samples.
  • Deep ultraviolet photometry is more valuable than additional optical bands for quasar redshift estimation, since the network on DESI+GALEX outperforms the comparable model using SDSS ugriz alone.
  • The bimodal structure in redshift predictions, tied to g–r versus z–W1 colours, means that explicitly modelling two quasar populations could further improve accuracy.
  • When DESI and SDSS spectroscopic redshifts disagree, the nine-band photometric prediction sides with DESI for about 73 per cent of the 107 outliers, suggesting DESI's line associations are generally the more reliable ones.

Reading between the lines

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

  • The comparison that carries the headline gain mixes two changes at once: the added ultraviolet bands and a switch from the full 87,318-source DESI sample to the smaller, brighter, better-constrained 24,616-source DxS subset; a clean test of the GALEX contribution would retrain the same architecture on the DxS sample with and without FUV/NUV, and the paper's own drop-column results on that fixed sa
  • Permutation importance places FUV and NUV near the bottom even though add/remove comparisons show they matter, a known failure mode for correlated features; future redshift-model papers should report drop-column or Shapley-value importances alongside permutation scores.
  • For radio-selected quasar samples expected from next-generation surveys, the same architecture could be retrained, but the selection-bias issue will be more severe because radio-selected samples have different redshift and luminosity distributions.
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

5 major / 5 minor

Summary. This paper trains fully connected neural-network and k-nearest-neighbour regressors to predict photometric redshifts for DESI EDR quasars, using DESI g, r, z and WISE W1, W2 fluxes, optionally augmented with GALEX FUV and NUV fluxes obtained through a cross-match with SDSS DR16Q (the DxS subset of 24,616 sources). The authors report that adding GALEX photometry improves the neural network from r=0.8099 and NMAD=0.2781 on the full DESI sample to r=0.9187 and NMAD=0.1971 on DxS, and interpret this as evidence that ultraviolet coverage helps. They also examine redshift-dependent performance, a bimodality in the predictions, feature importances via permutation and drop-column methods, and use of the model to adjudicate between discrepant DESI and SDSS spectroscopic redshifts.

Significance. If the comparison were controlled, the result would be a useful practical demonstration that adding ultraviolet photometry to optical plus infrared quasar photometry improves photometric redshifts, with potential value for future wide surveys. The paper has genuine strengths: 100 randomized train/test runs, multiple standard metrics, held-out evaluation for the main comparisons, and feature-importance analysis with two complementary methods, including the more reliable drop-column approach. However, the headline quantitative claim is not supported as presented because the main GALEX comparison changes both the feature set and the training sample, and the appendix's same-sample check both shows a smaller effect and is internally inconsistent with the main table. The qualitative conclusion that ultraviolet photometry helps is plausible and consistent with the direction of Table 9, but the magnitude of the improvement needs to be re-established with a controlled and internally consistent comparison.

major comments (5)
  1. [§3.3; Tables 4 and 6 vs Table 9] The central claim that adding GALEX ultraviolet photometry improves the neural network from NMAD 0.278 to 0.197 rests on a comparison across different training samples: Table 4 uses the full 87,318-source DESI sample with (g,r,z,W1,W2), while Table 6 uses the 24,616-source DxS subset with (g,r,z,W1,W2,FUV,NUV). Because Section 4.2 acknowledges that GALEX-detected quasars are brighter and better constrained, the improvement conflates added bands with an easier sample. The same-sample check in Table 9b shows a much smaller effect on DxS, NMAD 0.3171 without ultraviolet photometry versus 0.2699 with it, about 15% rather than 29%, so the headline magnitude is not established. Please re-run the DESI-only model on the same DxS subset used for the GALEX model and report both comparisons.
  2. [Table 9b vs Table 6] Table 9b gives a neural-network NMAD of 0.2699 for the feature set (g,r,z,W1,W2,FUV,NUV) on DxS, whereas Table 6 reports 0.1971 for what appears to be the same feature set and dataset; these two values cannot both describe the same configuration. Table 9a's kNN baseline of 0.2584 matches Table 6's kNN GALEX value, ruling out a simple across-the-board formatting difference. Please clarify whether Table 9b uses a different sample, a single run, a different network, or a different training pipeline, and reconcile the discrepancy.
  3. [Table 7; Eq. (1)] The percentage changes in Table 7 contradict Eq. (1), which defines the improvement relative to the NN DESI baseline. For example, the NN DESI/GALEX NMAD is reported as -41.1%, but Eq. (1) gives (0.1971 - 0.2781)/0.2781 = -29.1%; the correlation coefficient is reported as 11.84% instead of 13.4%. The Section 5 text, which states a 13% increase in correlation and a 29% reduction in NMAD, matches Eq. (1), but Table 7 and the surrounding discussion, which says all metrics improve by about 40%, do not. Correct the table and the related sentences.
  4. [§4.4; Figures 13-14; Conclusion] The outlier-adjudication claim is partly circular. The model whose predictions are used in Figures 13 and 14 was trained with zDESI as the target on the DxS sample after removing the 107 outliers, as described in Section 4.4, so a tendency for zphot to sit closer to zDESI than to zSDSS is expected even if the two spectroscopic measurements are equally accurate. To support the conclusion that DESI redshifts are more reliable in the discrepant cases, train a comparison model on zSDSS or validate the outlier assignments against an independent redshift source.
  5. [§2.6-2.7] Model selection appears to use the test set directly: Section 2.6 states that the neural-network architecture was chosen based on the lowest RMS error on the test set, and Section 2.7 states that the kNN hyperparameters were optimized to minimize RMS error on the test set. If the reported metrics are computed on the same test set used for selection, the absolute performance values are likely optimistic. Please use a separate validation split for hyperparameter selection and report test-set metrics from the final selected configuration.
minor comments (5)
  1. [§2.5] The definition of explained variance is malformed as printed: the expression for sigma-squared of the residuals lacks a division by N and appears to use zspec in the denominator; please restate it consistently with the residual definition Delta-z = zspec - zphot.
  2. [§2.6 / Figure 5] The text says architectures with 1 to 6 hidden layers and 50 to 300 neurons per layer were explored, but the final architecture is only given in Figure 5; please state the number of layers, neurons per layer, activation function, optimizer, and early-stopping details in the text.
  3. [§3.1] The symbol sigma-prime-z is introduced without definition in Section 3.1, where the text reports low scatter of sigma-prime-z = 0.387; define it or replace it with a metric defined in Section 2.5.
  4. [References] There is a typo in the introduction, W olf, in the reference to Wolf et al. 2018, and the reference list contains entries for Duncan 2021 and Duncan 2022 with only one clearly cited in the text; please clean up the bibliography.
  5. [Abstract / §5] The abstract and conclusions present r=0.9187 and NMAD=0.197 without stating that these values are for the DxS subset rather than the full DESI sample; please state the sample explicitly to avoid over-generalization.

Circularity Check

1 steps flagged · score 2.0 of 10

Main photo-z evaluation uses held-out test data and is not circular; one secondary outlier-reliability claim is biased by training on the very redshift catalog it endorses, and self-citations are not load-bearing.

  1. fitted input called prediction [Section 4.4, 'Predicting redshift for outliers in DESI/SDSS crossmatched sources' (Figures 13 and 14)]
    "To address the mismatches in the DESI and SDSS redshifts, we remove the 107 problematic sources and retrain our model on the remaining 24509 sources, using the DESI spectroscopic redshift (zDESI) as the target. Figure 13 shows a diagnostic scatterplot comparing the photometric redshift accuracy and photometric consistency for the outlier QSOs, showing that for 78 out of 107 (∼ 73%) outliers, our predictions most closely match the DESI spectroscopic redshift."

    The model's regression target is zDESI, so the photometric redshifts are fitted to reproduce DESI spectroscopic redshifts on the training sample. Using those same zphot values to adjudicate between zDESI and zSDSS for the 107 outliers therefore has a built-in bias: zphot will tend to lie closer to the catalog on which it was trained, even if the photometry carries little independent redshift information. The claim that 'our predictions most closely match the DESI spectroscopic redshift' is thus partly enforced by the training target, not an independent test of which catalog is more reliable.

full rationale

The central photo-z results are not circular: the NN is trained on 80% of a given sample and evaluated on a held-out 20%, with metrics averaged over 100 shuffled runs (Section 2.5), which is a standard out-of-sample evaluation. The self-citations to Curran et al. (2021, 2022) and Curran (2022) support method choices such as raw fluxes and the network architecture, and they are not load-bearing for the UV-improvement claim. I do not count the GALEX comparison as circular: although the abstract's comparison of Table 4 (full DESI, 87318 sources) with Table 6 (DxS subset, 24616 sources, with GALEX) is confounded by sample selection, and Section 4.2 itself notes that GALEX-detected quasars are 'already among the brighter, better-constrained subset,' this is an experimental-design flaw rather than a reduction of a result to its inputs. Likewise, the inconsistency between Table 6 (NN+DxS+GALEX NMAD = 0.1971) and Table 9b's baseline for the same feature set (NMAD = 0.2699), and the Table 7 percentage changes that contradict Eq. (1), are correctness issues to be fixed by the authors, not circularity. The one mild circular step is the Section 4.4 outlier analysis, where the model is trained with zDESI as the target and then used to argue that zphot favors DESI over SDSS; that claim is biased by construction and is a secondary, partial circularity. Hence the overall score is 2: one minor non-load-bearing self-citation pattern and one secondary claim that is partly fitted-input-called-prediction, while the main photometric-redshift derivation remains self-contained.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central claim rests primarily on standard ML assumptions plus the comparability of the DESI and DxS samples. No new physical entities are introduced. The model hyperparameters are fitted to the test set, which is a genuine free-parameter concern.

free parameters (3)
  • NN architecture (layers, neurons) = 3 hidden layers, 200 neurons each
    Selected by lowest RMS error on the test set (Section 2.6), so the test metrics are optimistic and the hyperparameters are fitted to the test data.
  • kNN k, distance metric, weighting = k=18, Manhattan, distance weighting
    Optimized via grid search over 100 iterations using RMS error on the test set (Section 2.7, Table 3).
  • Redshift outlier threshold = 0.14
    Chosen as the 1-sigma width of the residual distribution (Section 4.4); this selection removes 107 sources from training and affects the outlier analysis.
assumptions (6)
  • domain assumption Photometric redshifts are a function of broadband fluxes; ML can learn this mapping from spectroscopic training data.
    Foundational to the whole study; stated in Section 1 and used throughout.
  • domain assumption The DxS crossmatch with a 1 arcsec radius correctly associates DESI and SDSS sources.
    Section 2.2; if matches are wrong, the training labels are corrupted.
  • domain assumption Standard astronomical data processing (extinction correction, flux standardization) preserves the information needed for redshift inference.
    Section 2.4.
  • domain assumption The held-out test set is representative of the full DESI quasar population.
    This is challenged by the DxS subset selection; Section 4.2 acknowledges GALEX sources are brighter.
  • domain assumption DESI spectroscopic redshifts are the ground truth for training.
    Used as regression target throughout; the paper itself questions some SDSS redshifts but assumes DESI is correct (Section 4.4).
  • standard math Neural network training with MSE loss and early stopping converges to a stable solution.
    Assumed in Section 2.6; no formal convergence guarantee, but standard ML practice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Photometric Redshift Predictions with a Neural Network for DESI Quasars." pith.science (2026). https://pith.science/paper/7M6KI3QI

@misc{pith2026250703260,
  author       = {Pith},
  title        = {Pith review of: Photometric Redshift Predictions with a Neural Network for DESI Quasars},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7M6KI3QI}},
  note         = {Machine review of arXiv:2507.03260}
}
abstract

Accurate redshift measurements are essential for studying the evolution of quasi-stellar objects (QSOs) and their role in cosmic structure formation. While spectroscopic redshifts provide high precision, they are impractical for the vast number of sources detected in large-scale surveys. Photometric redshifts, derived from broadband fluxes, offer an efficient alternative, particularly when combined with machine learning techniques. In this work, we develop and evaluate a neural network model for predicting the redshifts of QSOs in the Dark Energy Spectroscopic Instrument (DESI) Early Data Release spectroscopic catalogue, using photometry from DESI, the Widefield Infrared Survey Explorer (WISE) and the Galactic Evolution Explorer (GALEX). We compare the performance of the neural network model against a k-Nearest Neighbours approach, these being the most accurate and least resource-intensive of the methods trialled herein, optimising model parameters and assessing accuracy with standard statistical metrics. Our results show that incorporating ultraviolet photometry from GALEX improves photometric redshift estimates, reducing scatter and catastrophic outliers compared to models trained only on near infrared and optical bands. The neural network achieves a correlation coefficient with spectroscopic redshift of $0.9187$ with normalised median absolute deviation of $0.197$, representing a significant improvement over other methods. Our work combines DESI, WISE and GALEX measurements, providing robust predictions which address the difficulties in predicting photometric redshift of QSOs over a large redshift range.

Figures

Figures reproduced from arXiv: 2507.03260 by the authors.

Figure 1
Figure 1. , contains 87318 sources spectroscopically identified as QSOs using DESI’s Redrock (RR) template-fitting algorithm and the QuasarNET (QN) deep-learning classifier. In this paper, the term ‘DESI dataset’ encompasses both the imaging data from the Legacy Imaging Surveys’ DR9, which includes g, r, and z fluxes, and the spectroscopic data from the EDR QSO catalogue, which provides redshift information for a subset of th… view at source ↗
Figure 3
Figure 3. The distribution of redshifts for the full DESI sample and the SDSS￾matched subset. The legend in each panel shows the mean redshift, standard deviation and maximum redshift [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. The distribution of magnitudes for the DESI sample as a whole compared to those with a match in SDSS (see Section 3.2). The legend in each panel shows the mean magnitude and the standard deviation. While DESI fluxes are used directly for model training, the comparison in this figure is made in magnitude space to match the SDSS format. 2.5 Model Evaluation and Metrics To evaluate model performance we use the followin… view at source ↗
Figures from the paper (15 more)
Figure 5
Figure 5. Figure 5: Architecture of the neural network used in the NN algorithm. Blue boxes show densely-connected hidden layers, each with 200 neurons and activation functions, and with optimiser and learning rate (lr) indicated. The orange box indicates the loss function (MSE) used to m…
Figure 6
Figure 6. Figure 6: shows a representative sample scatterplot for one of the 100 runs of the NN using the DESI fluxes as training features, with the lower panel showing the residuals normalised by ∆z/(1 + z). In this sample, the main cluster of points is near the 1:1 line. Some bimodality…
Figure 7
Figure 7. Figure 7: shows the distribution of |∆z| with the source separa￾tion. While there is a grouping of outliers at |∆z| ≳ 0.14, there is no sign of any correlation. These outliers were excluded from the training sample in order not to contaminate the sam￾ple. Lastly, to increase our…
Figure 8
Figure 8. Figure 8: Top: the difference in magnitudes versus difference in redshift between DESI and SDSS measurements, with m being g, r or z. Red stars show sources for which zDESI – zSDSS > 0.14. Bottom: distribution of the |∆m| in the top row [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Comparison of neural network photometric redshift predictions for SDSS-only versus SDSS+GALEX fluxes [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: presents a visualisation of the bimodal groups in colour-space. The bimodal groups were identified by assigning each source a label based on its position in colour space, and the centres of these groups were defined as the mean of their respective colour indices [PIT…
Figure 11
Figure 11. Figure 11: Comparison of kNN performance metrics across the DESI-only and DESI+GALEX (DxS) samples over five redshift bins [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: Spectroscopic redshifts from SDSS and DESI for the matched sources in the DxS sample. The background colour-coded scatterplot shows the 24509 QSOs for which zSDSS ≈ zDESI. The green crosses indicate sources classified as outliers which lie outside 1σ ∼ 0.14. The legen…
Figure 13
Figure 13. Figure 13: The sum of the difference in the SDSS and DESI g, r, z magnitudes versus the difference between the predicted and closest spectroscopic red￾shift for the outliers in [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: Photometric redshift predictions from the NN model (red stars) for the outliers identified in [PITH_FULL_IMAGE:figures/full_fig_p014_14.png]
Figure 15
Figure 15. Figure 15: Root Mean Squared Error (RMS) vs. Number of Nearest Neighbours (k) for the kNN, showing the average RMS error for each value of k. The error decreases sharply for small values of k and stabilises around k = 18, indicating an optimal choice for this parameter. Just as …
Figure 16
Figure 16. Figure 16: Example plot showing prediction results after training the kNN model on the DESI fluxes (g, r, z, W1, W2); cf [PITH_FULL_IMAGE:figures/full_fig_p016_16.png]
Figure 17
Figure 17. Figure 17: As for [PITH_FULL_IMAGE:figures/full_fig_p017_17.png]
Figure 19
Figure 19. Figure 19: Feature importances from (a) kNN and (b) neural network models. Each panel shows the mean increase in MSE when omitting each flux in the DESI/GALEX sample [PITH_FULL_IMAGE:figures/full_fig_p018_19.png]
Figure 20
Figure 20. Figure 20: The redshift predictions of a subset of the 107 outliers for which zDESI – zSDSS > 0.14, with 100 runs of the NN trained on the DESI/GALEX (g, r, z, W1, W2, NUV, FUV). The filled black markers show the DESI spectroscopic redshift, the unfilled markers the SDSS spectro…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 37 canonical work pages

  1. [1]

    2016, TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems, arXiv:1603.04467

    Abadi, M., Agarwal, A., Barham, P., et al. 2016, TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems, arXiv:1603.04467

    arXiv 2016

  2. [2]

    2022, The Astronomical Journal, 164, 207

    Abareshi, B., Aguilar, J., Ahlen, S., et al. 2022, The Astronomical Journal, 164, 207

    work page 2022

  3. [3]

    G., Aguilar, J., Ahlen, S., et al

    Adame, A. G., Aguilar, J., Ahlen, S., et al. 2023, arXiv e-prints, doi:10.48550/ARXIV.2306.06308

    arXiv doi:10.48550/arxiv.2306.06308 2023

  4. [4]

    M., Davis, T

    Alexander, D. M., Davis, T. M., Chaussidon, E., et al. 2023, The Astronomical Journal, 165, 124

    work page 2023

  5. [5]

    W., & Gould, A

    Atlee, D. W., & Gould, A. 2007, The Astrophysical Journal, 664, 53

    work page 2007

  6. [6]

    M., Brunner, R

    Ball, N. M., Brunner, R. J., Myers, A. D., et al. 2008, The Astrophysical Journal, 683, 12 —. 2007, The Astrophysical Journal, 663, 774

    work page 2008

  7. [7]

    2023, Bird’s Convention for Diagramatic Neural Networks

    Bird, G. 2023, Bird’s Convention for Diagramatic Neural Networks

    work page 2023

  8. [8]

    D., Hennawi, J

    Bovy, J., Myers, A. D., Hennawi, J. F., et al. 2012, The Astrophysical Journal, 749, 41

    work page 2012

Show all 46 references
  1. [9]

    B., van Dokkum, P

    Brammer, G. B., van Dokkum, P. G., & Coppi, P. 2008, The Astrophysical Journal, 686, 1503

    2008

  2. [10]

    2021, Frontiers in Astronomy and Space Sciences, 8, arXiv:2104.06774

    Brescia, M., Cavuoti, S., Razim, O., et al. 2021, Frontiers in Astronomy and Space Sciences, 8, arXiv:2104.06774

    2021 arXiv

  3. [11]

    2018, doi:10.48550/ARXIV.1808.09955

    Busca, N., & Balland, C. 2018, doi:10.48550/ARXIV.1808.09955

    2018 doi arXiv

  4. [12]

    2004, arXiv e-prints, 1

    Carilli, C., & Rawlings, S. 2004, arXiv e-prints, 1

    2004

  5. [13]

    2007, Astronomical Data Analysis Software and Systems, arXiv:0711.2477v1

    Carliles, S., Heinis, S., & Priebe, C. 2007, Astronomical Data Analysis Software and Systems, arXiv:0711.2477v1

    2007 arXiv

  6. [14]

    2023, The Astrophysical Journal, 944, 107

    Chaussidon, E., Yèche, C., Palanque-Delabrouille, N., et al. 2023, The Astrophysical Journal, 944, 107

    2023

  7. [15]

    J., & Szalay, A

    Connolly, A. J., & Szalay, A. S. 1999, The Astronomical Journal, 117, 2052

    1999

  8. [16]

    Curran, S. J. 2020, Monthly Notices of the Royal Astronomical Society: Letters, 493, L70 —. 2022, Monthly Notices of the Royal Astronomical Society, 12, 1

    2020

  9. [17]

    J., Moss, J

    Curran, S. J., Moss, J. P., & Perrott, Y. C. 2021, Monthly Notices of the Royal Astronomical Society, 503, 2639 —. 2022, Monthly Notices of the Royal Astronomical Society, 514, 1

    2021

  10. [18]

    2021, TensorFlow, Zenodo, doi:10.5281/zenodo.4758419

    Developers, T. 2021, TensorFlow, Zenodo, doi:10.5281/zenodo.4758419

    2021 doi

  11. [19]

    J., Lang, D., et al

    Dey, A., Schlegel, D. J., Lang, D., et al. 2019, The Astronomical Journal, 157, 168 D’Isanto, A., & Polsterer, K. L. 2018, Astronomy and Astrophysics, 609, A111

    2019

  12. [20]

    2010, Astronomische Nachrichten, 331, 535

    Dravins, D. 2010, Astronomische Nachrichten, 331, 535

    2010

  13. [21]

    Duncan, K. J. 2021, 20, 1 —. 2022, Monthly Notices of the Royal Astronomical Society, 512, 3662 Gil de Paz, A., Boissier, S., Madore, B. F., et al. 2007, The Astrophysical Journal Supplement Series, 173, 185

    2021

  14. [22]

    2016, Research in Astronomy and Astrophysics, 16, 005

    Han, B., Ding, H.-P., Zhang, Y.-X., & Zhao, Y.-H. 2016, Research in Astronomy and Astrophysics, 16, 005

    2016

  15. [23]

    2022, 10, 1

    Hong, S., Zou, Z., Luo, A.-l., et al. 2022, 10, 1

    2022

  16. [24]

    2015, Monthly Notices of the Royal Astronomical Society, 449, 1275

    Hoyle, B., Michael Rau, M., Zitlau, R., Seitz, S., & Weller, J. 2015, Monthly Notices of the Royal Astronomical Society, 449, 1275

    2015

  17. [25]

    P., & Ba, J

    Kingma, D. P., & Ba, J. 2017, Adam: A Method for Stochastic Optimization, arXiv:1412.6980 Kügler, S. D., Gianniotis, N., & Polsterer, K. L. 2016, in 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016, Athens, Greece, December 6-9, 2016 (IEEE), 1–8

    2017 arXiv

  18. [26]

    2022, 525, 513

    Li, C., Zhang, Y., Cui, C., et al. 2022, 525, 513

    2022

  19. [27]

    2023, Publications of the Astronomical Society of Australia, 40, 1

    Luken, K., Norris, R., Wang, X., et al. 2023, Publications of the Astronomical Society of Australia, 40, 1

    2023

  20. [28]

    J., Norris, R

    Luken, K. J., Norris, R. P., Park, L. A. F., Wang, X. R., & Filipovic, M. D. 2022, Estimating Galaxy Redshift in Radio-Selected Datasets Using Machine Learning, arXiv:2202.13504

    2022 arXiv

  21. [29]

    W., Higley, A

    Lyke, B. W., Higley, A. N., McLane, J. N., et al. 2020, Astrophysical Journal Supplement Series, 250, 8

    2020

  22. [30]

    C., Fanson, J., Schiminovich, D., et al

    Martin, D. C., Fanson, J., Schiminovich, D., et al. 2005, The Astrophysical Journal, 619, L1

    2005

  23. [31]

    R., Soares, G., et al

    Nakazono, L., Valença, R. R., Soares, G., et al. 2024, Monthly Notices of the Royal Astronomical Society, stae971

    2024

  24. [32]

    D., Jimenez, R., Verde, L., et al

    Niemack, M. D., Jimenez, R., Verde, L., et al. 2009, The Astrophysical Journal, 690, 89

    2009

  25. [33]

    P., Hopkins, A

    Norris, R. P., Hopkins, A. M., Afonso, J., et al. 2011, Publications of the Astronomical Society of Australia, 28, 215

    2011

  26. [34]

    2019, Astronomy and Astrophysics, 621, 1

    Pasquet, J., Bertin, E., Treyer, M., Arnouts, S., & Fouchez, D. 2019, Astronomy and Astrophysics, 621, 1

    2019

  27. [35]

    2018, Astronomy and Astrophysics, 611, 1

    Pasquet-Itam, J., & Pasquet, J. 2018, Astronomy and Astrophysics, 611, 1

    2018

  28. [36]

    2017, Publications of the Astronomical Society of the Pacific, 129, doi:10.1088/1538-3873/aa7843

    Popowicz, A., & Kurek, A. 2017, Publications of the Astronomical Society of the Pacific, 129, doi:10.1088/1538-3873/aa7843

    2017 doi

  29. [37]

    2024, Astronomy & Astrophysics, 690, A365

    Saxena, A., Salvato, M., Roster, W., et al. 2024, Astronomy & Astrophysics, 690, A365

    2024

  30. [38]

    Schilizzi, R. T. 2004, https://doi.org/10.1117/12.551206, 5489, 62

    2004 doi

  31. [39]

    F., & Finkbeiner, D

    Schlafly, E. F., & Finkbeiner, D. P. 2011, The Astrophysical Journal, 737, 103

    2011

  32. [40]

    2021, 237, 235.03 W olf, C., Onken, C

    Schlegel, D., Dey, A., Herrera, D., et al. 2021, 237, 235.03 W olf, C., Onken, C. A., Luvaul, L. C., et al. 2018, Publications of the Astronomical Society of Australia, 35, doi:10.1017/pasa.2018.5

    2021 doi

  33. [41]

    L., Eisenhardt, P

    Wright, E. L., Eisenhardt, P. R. M., Mainzer, A. K., et al. 2010, The Astronomical Journal, 140, 1868

    2010

  34. [42]

    B., Liu, X

    Yuan, H. B., Liu, X. W., & Xiang, M. S. 2013, Monthly Notices of the Royal Astronomical Society, 430, 2188

    2013

  35. [43]

    2010, Monthly Notices of the Royal Astronomical Society, 408, 1283

    Zhang, F., Han, Z., Li, L., Shan, H., & Zhang, Y. 2010, Monthly Notices of the Royal Astronomical Society, 408, 1283

    2010

  36. [44]

    2013, The Astronomical Journal, 146, 22

    Zhang, Y., Ma, H., Peng, N., Zhao, Y., & Wu, X.-b. 2013, The Astronomical Journal, 146, 22

    2013

  37. [45]

    2019, Research in Astronomy and Astrophysics, 19, 175

    Zhang, Y.-X., Zhang, J.-Y., Jin, X., & Zhao, Y.-H. 2019, Research in Astronomy and Astrophysics, 19, 175

    2019

  38. [46]

    A., Mao, Y.-Y., et al

    Zhou, R., Newman, J. A., Mao, Y.-Y., et al. 2021, Monthly Notices of the Royal Astronomical Society, 501, 3309

    2021

Pith tools

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