Pith. sign in

REVIEW 2 major objections 6 minor 66 references

Deep Transfer Learning for Star Cluster Classification: I. Application to the PHANGS-HST Survey

T0 review · 2 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Pretrained image networks can sort Hubble star clusters into four classes with human-level accuracy.

desk verdict A solid proof-of-concept for deep transfer classification of HST star clusters, with an honest headline claim that rests on a wobbly human-consistency yardstick. read the letter →

arxiv 1909.02024 v2 pith:SR6HA3TU submitted 2019-09-04 astro-ph.GA astro-ph.IMphysics.data-anphysics.ins-det

Wei Wei , E. A. Huerta , Bradley C. Whitmore , Janice C. Lee , Stephen Hannon , Rupali Chandar , Daniel A. Dale , Kirsten L. Larson
show 8 more authors
David A. Thilker Leonardo Ubeda Médéric Boquien Mélanie Chevance J. M. Diederik Kruijssen Andreas Schruba Guillermo Blanc Enrico Congiu
This is my paper · ORCID
classification astro-ph.GAastro-ph.IMphysics.data-anphysics.ins-det
keywords deeptransferlearningstarclusterclassificationPHANGS-HSTLEGUSHubbleSpaceTelescopemorphologicalconvolutionalneuralnetworksResNet
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

This proof-of-concept paper asks whether deep networks trained on ordinary photographs can be repurposed to classify compact star clusters in Hubble Space Telescope images of nearby galaxies. The authors fine-tune two standard networks, originally trained on the ImageNet object-recognition dataset, on a few thousand human-labelled star cluster postage stamps from the LEGUS survey. When the networks are applied to a galaxy they have never seen, NGC 1559, they recover roughly 70% of the symmetric compact clusters, about 40% of the asymmetric compact clusters, 40-50% of the compact associations, and 50-75% of the non-clusters. The paper argues that this matches the agreement typically found between different human classifiers, and therefore that automated classification is ready for production-scale application to the much larger PHANGS-HST survey.

What carries the argument

The central mechanism is deep transfer learning: starting from deep convolutional neural networks (ResNet18 and VGG19-BN) pretrained on the ImageNet object-recognition dataset and re-training only the later layers on the small star-cluster dataset. The networks are fed 5-channel postage stamps (F275W, F336W, F438W, F555W, F814W) from HST imaging, and the architecture is augmented with a final softmax layer that outputs a probability distribution over the four cluster classes. The argument runs through a series of robustness checks: results are largely unchanged whether the network is trained on classifications by a single expert (BCW) or the consensus of three LEGUS classifiers, whether the crops are 25, 50, or 100 pixels on a side, and which of two architectures is used. This resilience to curation choices is what justifies the claim that the method can be applied at scale to the PHANGS-HST survey.

What would settle it

A direct test would be to have several independent expert classifiers label the same NGC 1559 candidate set used here, and then compare the neural network's predictions against each expert. If the network agrees with the consensus of several experts noticeably less well than the experts agree among themselves, and if the disagreement is not concentrated in the intrinsically ambiguous class 2 and 3 objects, the claim of human-competitive accuracy would be refuted. A simpler check would be to test the network on a second unseen galaxy with multi-expert labels and ask whether the class-by-class accuracies against each expert fall within the published human consistency ranges.

Watch

Extended reading notes

Core claim

The paper's central claim is that deep transfer learning provides a viable route to automating the four-way morphological classification of star cluster candidates in HST UV-optical imaging. Using only a few thousand labelled examples, much smaller than the millions typically required for deep learning, the authors adapt the ImageNet-pretrained networks ResNet18 and VGG19-BN to classify cluster images into class 1 (compact symmetric cluster), class 2 (compact asymmetric cluster), class 3 (compact association), and class 4 (non-cluster). On previously unseen cluster candidates in the galaxy NGC 1559, the networks achieve prediction accuracies of ~70%, ~40%, 40-50%, and 50-75% for the four classes respectively, depending on architecture and training set. The paper asserts that these accuracies are competitive with the 70-80%, 40-50%, 40-50%, and 60-70% consistency levels documented for human classification of the same types of objects, and concludes that the method lays the foundation for automating classification of tens of thousands of star cluster candidates expected from PHANGS-HST.

Load-bearing premise

The claim that the networks perform at human level depends on the assumption that the human-consistency percentages quoted from earlier studies are the right yardstick, and that the single expert's labels used for testing on NGC 1559 are an unbiased gold standard.

Editorial extensions

If this is right

  • If the central claim holds, the PHANGS-HST survey can automate the first pass of star cluster classification for several tens of thousands of candidates, delivering class labels far faster than human eyes can.
  • Automating the classification removes a source of subjectivity and inter-observer scatter from cluster catalogs, and the network output includes an entropy-based confidence estimate for every object, which human labels do not carry natively.
  • Because the networks already achieve human-level consistency on a galaxy at roughly twice the distance of most training galaxies, the method can likely be extended across the full distance range of the survey with modest re-training.
  • A standardized, expert-agreed training dataset would likely push the accuracy of the networks beyond human consistency levels, since the networks currently inherit the disagreements baked into their single-expert or consensus training labels.
  • The finding that accuracy does not depend strongly on image crop size (from 16 pc to 360 pc physical scales) means the same pretrained network can be applied to surveys with different pixel scales or galaxy distances without re-tuning.

Reading between the lines

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

  • The pipeline can probably be extended to a binary first-pass filter (true cluster vs. non-cluster) with higher accuracy than the four-way task, since class 1 and class 4 are the most reliably learned categories and dominate the clean separation of signal from contamination.
  • The network's reliance on the F555W (V-band) image, which the human classifiers also use most, suggests the network has learned morphology rather than colour information; this could make it robust to the different filter sets of future surveys, though this is an extrapolation from the paper's ablation experiment.
  • A natural, testable next step is to use the predicted probability distributions and Shannon entropies to guide triage: objects with confident predictions can enter the catalog directly, while low-confidence objects are sent to human review, which would make the model useful even where its aggregate accuracy is imperfect.
  • If trained on a consensus dataset of multiple expert classifiers, the method could in principle define a more consistent 'silver standard' than any single human annotator, since the network averages over noise in the labels rather than reproducing one person's biases.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper presents a proof-of-concept application of deep transfer learning to morphological classification of compact star cluster candidates in HST imaging. The authors fine-tune ImageNet-pretrained ResNet18 and VGG19-BN models on LEGUS star cluster postage stamps labeled either by a single expert (BCW) or by the mode of three LEGUS classifiers, and evaluate on a held-out validation split and on the PHANGS-HST galaxy NGC 1559, which was not used in training. Per-class accuracies on NGC 1559 are approximately 70%, 40%, 40-50%, and 50-70% for classes 1-4, which the paper argues are competitive with previously published human and automated classification consistency (70-80%, 40-50%, 40-50%, 60-70%). Additional experiments examine the dependence on architecture, label source, image cropping size, and filter set.

Significance. The experimental protocol is careful and is the main strength of the paper: ten independent trainings with reported standard deviations, a genuinely held-out galaxy, two network architectures, and two label sources. The technical conclusion that transfer-learned CNNs can reproduce single-expert star cluster classifications at rates comparable to published human agreement is well supported by the confusion matrices and uncertainty estimates. The paper is also honest in acknowledging the self-consistency effect in the class 4 comparison. The central caveat is that the human-consistency benchmark is assembled from heterogeneous comparisons and is not measured on the same objects as the network evaluation, so the headline claim of being 'competitive with humans' is not yet established at the strength stated; this is addressable with reframing or a modest amount of additional label data.

major comments (2)
  1. [Section 2.1 and Tables 6-7] The adopted human-consistency yardstick (70-80%, 40-50%, 40-50%, 60-70%) is not measured on the same objects or under the same protocol as the NGC 1559 evaluation. The ranges are assembled from catalog-overlap fractions that mix human classification with automated concentration-index cuts, from a same-expert repeat classification of NGC 3351, and from a single-expert versus three-person-mode comparison for NGC 4656; none of these is a same-object inter-expert agreement for NGC 1559. Because the abstract's central claim is that the network performance is 'competitive with consistency achieved in previously published human ... classification,' the comparison should either be supplemented with independent expert labels for a subsample of NGC 1559 or explicitly reframed as agreement with a single expert at rates similar to published single-expert agreement.
  2. [Section 4.2, Tables 6-7] The NGC 1559 evaluation labels are BCW's own classifications, and BCW labeled most of the primary training set (Table 1). The paper itself attributes the higher class-4 accuracy of BCW-trained models (67-75%) over LEGUS-consensus-trained models (52-62%) to self-consistency. This confound means that the headline accuracies for classes 1 and 4 partly measure agreement with a single expert rather than with an independent human consensus. Please quantify the effect by reporting model performance against independent labels for a subset of NGC 1559, or by clearly presenting the BCW-trained and LEGUS-trained results as upper and lower bounds on the human-level claim.
minor comments (6)
  1. [Section 3.1, Table 1] The count of ten BCW-classified fields in Table 1 is not transparent from the text, which says that classifications for 4 of 8 BCW-primary fields are in the LEGUS public archive and that two additional fields were independently classified by BCW; please add a column or footnote specifying the source of each field.
  2. [Section 3.3] Please specify the mapping of the five filters (F275W, F336W, F438W, F555W, F814W) to the three input channels of each concatenated ImageNet-pretrained copy, and justify the choice of a zero-filled sixth channel.
  3. [Section 4.2.1] The sentence 'the network network in this case is 100% certain' contains a duplicated word; additionally, the entropy analysis would be more informative if it showed classification accuracy as a function of an entropy threshold.
  4. [Section 2.1] The phrase 'same data sets' is imprecise for the catalog-overlap comparisons, because those studies combine human classification with automated concentration-index cuts; consider 'same imaging data but different selection procedures.'
  5. [Figure 5] The normalized entropy histograms would be easier to interpret with object counts per bin or a twin axis; the current y-axis label is vague.
  6. [End of manuscript] Please add a data and code availability statement; the paper relies on public LEGUS catalogs but does not state whether the trained models or training scripts will be released.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the network predictions on NGC 1559 are held-out and not fitted to the NGC 1559 labels; the human-consistency yardstick is an external benchmark, so the derivation chain is not self-referential.

full rationale

The paper is an empirical machine-learning study, not a derivation, and its central claimed result is a held-out generalization test: networks are trained on LEGUS samples (Tables 1 and 2), then applied to NGC 1559 candidates that were not in the training set. The reported NGC 1559 accuracies (Tables 6 and 7, and Figure 4) are measured against BCW's labels for that galaxy, and those labels were not used to fit any parameter of the models. There is therefore no fitted-input-called-prediction relation and no equation-level reduction of the result to its inputs. The human-consistency ranges in Section 2.1 are assembled from previously published catalog comparisons and BCW-versus-other-classifier comparisons; they are external benchmarks, not outputs of the networks, and the paper does not derive its network accuracies from them. The main self-reference is that BCW labelled most of the training galaxies and also labelled the NGC 1559 test set, which the paper explicitly discloses in Section 4.2: "the classifications for NGC 1559 were also performed by BCW, and may be due to a higher level of self-consistency in the training and testing classification datasets." That acknowledged effect is a limitation on external validity of the class-4 comparison, but it is not circularity: the test accuracies are genuine agreement rates with a held-out expert's labels, not quantities forced by construction. Similarly, the use of ImageNet pretrained weights and previously published LEGUS classification catalogs is ordinary independent support, not a load-bearing self-citation chain. No uniqueness theorem, ansatz-by-citation, or renaming of a known result is present. The competitive-with-humans interpretation could be challenged because the human-consistency yardstick was not measured on the same objects under the same protocol, but that is a correctness/interpretation risk, not circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim is empirical and rests on the reliability of human labels, the transferability of ImageNet features, the adopted human-consistency benchmark, and the candidate selection pipeline. No new physical entities are introduced. The network hyperparameters are standard and the paper tests robustness to some of them, but the benchmark ranges and label assumptions are the least externally grounded elements.

free parameters (4)
  • Human consistency benchmark ranges = 70-80%, 40-50%, 40-50%, 60-70% for classes 1-4
    Adopted in Section 2.1 as the target accuracy to be achieved or surpassed; assembled from heterogeneous human-vs-human and sample-overlap comparisons. The competitive with humans claim is evaluated against these hand-chosen ranges.
  • Training hyperparameters = learning rate 1e-4, batch size 32 (ResNet18) / 16 (VGG19-BN), ~10,000 batches
    Chosen in Section 3.3 without systematic tuning; the paper shows results are robust to some choices, but these specific values underpin all reported accuracies.
  • Image cropping size = 50x50 pixels fiducial; 25x25 and 100x100 variants
    Section 3.2 and 4.3: the physical scale of input stamps is a curation choice; robustness to it is tested, but the fiducial size is hand-selected.
  • Magnitude limits for candidate selection = -6 mag F555W (LEGUS), -7.5 mag V (NGC1559)
    Used to build the candidate samples from which training and test images are drawn; different limits for training and test galaxies could bias the comparison.
assumptions (5)
  • domain assumption ImageNet-pretrained features transfer to HST star cluster images.
    Section 3.3 and Appendix B assume that low-level features learned on natural images are useful for astronomical images; this is the premise of transfer learning and is tested indirectly by the accuracy results, not proven independently.
  • domain assumption Human visual classifications (BCW and LEGUS modes) are a valid ground truth for the four morphological classes.
    The accuracy metrics measure agreement with human labels (Section 4); the paper does not validate these classes against physical properties such as binding, age, or mass, so the network inherits any subjectivity in the labels.
  • domain assumption The four-class LEGUS system (compact symmetric, compact asymmetric, association, non-cluster) is the appropriate taxonomy.
    Adopted in Section 2 from Adamo et al. 2017 and Cook et al. 2019; alternative taxonomies exist, and the network's performance is only meaningful within this scheme.
  • domain assumption The candidate selection pipeline (SExtractor, concentration index, photometric error cuts, magnitude limits) yields an unbiased sample of cluster candidates.
    Sections 3.1: pre-selection discards unresolved point sources and some artifacts; any bias in this step propagates into training and test sets, and the network cannot recover rejected objects.
  • standard math Standard statistical learning framework for classification.
    Appendix A: cross-entropy minimization, empirical risk, and function classes; standard and uncontroversial.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Transfer Learning for Star Cluster Classification: I. Application to the PHANGS-HST Survey." pith.science (2026). https://pith.science/paper/SR6HA3TU

@misc{pith2026190902024,
  author       = {Pith},
  title        = {Pith review of: Deep Transfer Learning for Star Cluster Classification: I. Application to the PHANGS-HST Survey},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SR6HA3TU}},
  note         = {Machine review of arXiv:1909.02024}
}
read the original abstract

We present the results of a proof-of-concept experiment which demonstrates that deep learning can successfully be used for production-scale classification of compact star clusters detected in HST UV-optical imaging of nearby spiral galaxies (D < 20 Mpc) in the PHANGS-HST survey. Given the relatively small nature of existing, human-labelled star cluster samples, we transfer the knowledge of state-of-the-art neural network models for real-object recognition to classify star clusters candidates into four morphological classes. We perform a series of experiments to determine the dependence of classification performance on: neural network architecture (ResNet18 and VGG19-BN); training data sets curated by either a single expert or three astronomers; and the size of the images used for training. We find that the overall classification accuracies are not significantly affected by these choices. The networks are used to classify star cluster candidates in the PHANGS-HST galaxy NGC 1559, which was not included in the training samples. The resulting prediction accuracies are 70%, 40%, 40-50%, 50-70% for class 1, 2, 3 star clusters, and class 4 non-clusters respectively. This performance is competitive with consistency achieved in previously published human and automated quantitative classification of star cluster candidate samples (70-80%, 40-50%, 40-50%, and 60-70%). The methods introduced herein lay the foundations to automate classification for star clusters at scale, and exhibit the need to prepare a standardized dataset of human-labelled star cluster classifications, agreed upon by a full range of experts in the field, to further improve the performance of the networks introduced in this study.

Figures

Figures reproduced from arXiv: 1909.02024 by the authors.

Figure 1
Figure 1. Examples of each of the four cluster classifications illustrated with HST/WFC3 imaging. The top four rows show star clusters from NGC 4656, which are part of the training set, while the bottom four rows show clusters from recent PHANGS-HST observations of the spiral galaxy NGC 1559, which form our proof-of-concept test sample, and are not used for training. The first two columns show false-color RGB images for conte… view at source ↗
Figure 2
Figure 2. Comparisons between star cluster candidate classifi￾cations made by BCW and the mode of classifications made by three other LEGUS team members (trained by BCW, A. Adamo, and H. Kim) provided in the LEGUS public star cluster catalog for NGC 4656. Each panel shows the distribution of classifica￾tions given in the LEGUS catalog for BCW labelled class 1 (top, symmetric compact clusters), class 2 (upper middle, asymmetri… view at source ↗
Figure 3
Figure 3. Top panels: Prediction, averaged over 10 models, of ResNet18 (left) and VGG19-BN (right) trained on 80% of the data described in [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Top panels: Same as [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: The uncertainty in the neural network’s prediction is quantified by the entropy of the predicted probability distribution over the 4 star cluster image classes considered in this analysis. For a random guess over the 4 classes, the entropy is ln4 ≈ 1.39. The lower the …
Figure 6
Figure 6. Figure 6: Left column: VGG19-BN model classification results for cropping size 25 × 25, 50 × 50 and 100 × 100. Right column: as before, but now for ResNet. 5 DISCUSSION & CONCLUSIONS Using homogeneous datasets of human-labeled star cluster images from the Hubble Space Telescope,…
Figure 7
Figure 7. Figure 7: Left column: ResNet model classification results when the indicated filter is removed from the composite image. Right column: as before, but now for VGG19-BN. The greatest drop in the accuracies occurs when the V-band filter is removed. MNRAS 000, 1–17 (2019) [PITH_FU…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 16 canonical work pages

  1. [1]

    Abbott T., et al., 2016, @doi [Mon. Not. Roy. Astron. Soc.] 10.1093/mnras/stw641 , 460, 1270

    doi:10.1093/mnras/stw641 2016

  2. [2]

    K., Turp M

    Ackermann S., Schawinski K., Zhang C., Weigel A. K., Turp M. D., 2018, @doi [ ] 10.1093/mnras/sty1398 , 479, 415

    doi:10.1093/mnras/sty1398 2018

  3. [3]

    Adamo A., et al., 2017, @doi [ ] 10.3847/1538-4357/aa7132 , http://adsabs.harvard.edu/abs/2017ApJ...841..131A 841, 131

    doi:10.3847/1538-4357/aa7132 2017

  4. [4]

    M., Brunner R

    Ball N. M., Brunner R. J., Myers A. D., Tcheng D., 2006, @doi [ ] 10.1086/507440 , 650, 497

    doi:10.1086/507440 2006

  5. [5]

    M., Brunner R

    Ball N. M., Brunner R. J., Myers A. D., Strand N. E., Alberts S. L., Tcheng D., 2008, @doi [ ] 10.1086/589646 , 683, 12

    doi:10.1086/589646 2008

  6. [6]

    Banerji M., et al., 2010, @doi [ ] 10.1111/j.1365-2966.2010.16713.x , 406, 342

    arXiv 2010

  7. [7]

    Machine and Deep Learning Applied to Galaxy Morphology -- A Comparative Study

    Barchi P. H., de Carvalho R. R., Rosa R. R., Sautter R., Soares-Santos M., Marques B. A. D., Clua E., 2019, arXiv e-prints, p. arXiv:1901.07047

    work page Pith review arXiv 2019

  8. [8]

    Bastian N., et al., 2012, @doi [ ] 10.1111/j.1365-2966.2011.19909.x , http://adsabs.harvard.edu/abs/2012MNRAS.419.2606B 419, 2606

    arXiv 2012

Show all 66 references
  1. [9]

    pp 17--37

    Bengio Y., 2011, in Proceedings of the 2011 International Conference on Unsupervised and Transfer Learning workshop-Volume 27. pp 17--37

    2011

  2. [10]

    Bertin E., Arnouts S., 1996a, @doi [ ] 10.1051/aas:1996164 , https://ui.adsabs.harvard.edu/abs/1996A

    doi

  3. [11]

    Bertin E., Arnouts S., 1996b, , http://adsabs.harvard.edu/abs/1996A

  4. [12]

    Calzetti D., et al., 2015a, @doi [ ] 10.1088/0004-6256/149/2/51 , http://adsabs.harvard.edu/abs/2015AJ....149...51C 149, 51

    doi

  5. [13]

    Calzetti D., et al., 2015b, @doi [ ] 10.1088/0004-637X/811/2/75 , http://adsabs.harvard.edu/abs/2015ApJ...811...75C 811, 75

    doi

  6. [14]

    J., Pickering E

    Cannon A. J., Pickering E. C., 1912, Annals of Harvard College Observatory, https://ui.adsabs.harvard.edu/abs/1912AnHar..56...65C 56, 65

    1912

  7. [15]

    J., Pickering E

    Cannon A. J., Pickering E. C., 1918, Annals of Harvard College Observatory, https://ui.adsabs.harvard.edu/abs/1918AnHar..91....1C 91, 1

    1918

  8. [16]

    J., 2013, @doi [ ] 10.1093/mnras/stt574 , 432, 1483

    Carrasco Kind M., Brunner R. J., 2013, @doi [ ] 10.1093/mnras/stt574 , 432, 1483

    2013 doi

  9. [17]

    Chandar R., et al., 2010, @doi [ ] 10.1088/0004-637X/719/1/966 , http://adsabs.harvard.edu/abs/2010ApJ...719..966C 719, 966

    2010 doi

  10. [18]

    C., Calzetti D., O'Connell R., 2014, @doi [ ] 10.1088/0004-637X/787/1/17 , https://ui.adsabs.harvard.edu/abs/2014ApJ...787...17C 787, 17

    Chandar R., Whitmore B. C., Calzetti D., O'Connell R., 2014, @doi [ ] 10.1088/0004-637X/787/1/17 , https://ui.adsabs.harvard.edu/abs/2014ApJ...787...17C 787, 17

    2014 doi

  11. [19]

    C., Dinino D., Kennicutt R

    Chandar R., Whitmore B. C., Dinino D., Kennicutt R. C., Chien L. H., Schinnerer E., Meidt S., 2016, @doi [ ] 10.3847/0004-637X/824/2/71 , https://ui.adsabs.harvard.edu/abs/2016ApJ...824...71C 824, 71

    2016 doi

  12. [20]

    O., et al., 2019, @doi [ ] 10.1093/mnras/stz331 , http://adsabs.harvard.edu/abs/2019MNRAS.484.4897C 484, 4897

    Cook D. O., et al., 2019, @doi [ ] 10.1093/mnras/stz331 , http://adsabs.harvard.edu/abs/2019MNRAS.484.4897C 484, 4897

    2019 doi

  13. [21]

    http://www.image-net.org/

    Deng J., Dong W., Socher R., Li L.-J., Li K., Fei-Fei L., 2009, in CVPR09. http://www.image-net.org/

    2009

  14. [22]

    W., Dambre J., 2015, @doi [ ] 10.1093/mnras/stv632 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.450.1441D 450, 1441

    Dieleman S., Willett K. W., Dambre J., 2015, @doi [ ] 10.1093/mnras/stv632 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.450.1441D 450, 1441

    2015 doi

  15. [23]

    arXiv:1807.00807 ( @eprint arXiv 1807.00807 )

    Dom \' nguez S \'a nchez H., Huertas-Company et al., 2018, preprint, p. arXiv:1807.00807 ( @eprint arXiv 1807.00807 )

    2018 arXiv

  16. [24]

    Everingham M., Eslami S. M. A., Van Gool L., Williams C. K. I., Winn J., Zisserman A., 2015, International Journal of Computer Vision, 111, 98

    2015

  17. [25]

    W., Willman B., 2012, @doi [ ] 10.1088/0004-637X/760/1/15 , 760, 15

    Fadely R., Hogg D. W., Willman B., 2012, @doi [ ] 10.1088/0004-637X/760/1/15 , 760, 15

    2012 doi

  18. [26]

    A., 2017, arXiv e-prints, p

    George D., Shen H., Huerta E. A., 2017, arXiv e-prints, p. arXiv:1711.07468

    2017 arXiv

  19. [27]

    A., 2018, @doi [ ] 10.1103/PhysRevD.97.101501 , 97, 101501

    George D., Shen H., Huerta E. A., 2018, @doi [ ] 10.1103/PhysRevD.97.101501 , 97, 101501

    2018 doi

  20. [28]

    MIT Press

    Goodfellow I., Bengio Y., Courville A., 2016, Deep Learning. MIT Press

    2016

  21. [29]

    A., 2018, @doi [ ] 10.1088/1538-3873/aac1fd , https://ui.adsabs.harvard.edu/abs/2018PASP..130g2001G 130, 072001

    Gouliermis D. A., 2018, @doi [ ] 10.1088/1538-3873/aac1fd , https://ui.adsabs.harvard.edu/abs/2018PASP..130g2001G 130, 072001

    2018 doi

  22. [30]

    Grasha K., et al., 2019, @doi [ ] 10.1093/mnras/sty3424 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.483.4707G 483, 4707

    2019 doi

  23. [31]

    arXiv:1512.03385

    He K., Zhang X., Ren S., Sun J., 2015, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2015arXiv151203385H p. arXiv:1512.03385

    2015 arXiv

  24. [32]

    pp 770--778, https://www.cv-foundation.org/openaccess/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html

    He K., Zhang X., Ren S., Sun J., 2016, in Proceedings of the IEEE conference on computer vision and pattern recognition. pp 770--778, https://www.cv-foundation.org/openaccess/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html

    2016

  25. [33]

    A., et al., 1992, @doi [ ] 10.1086/116094 , https://ui.adsabs.harvard.edu/abs/1992AJ....103..691H 103, 691

    Holtzman J. A., et al., 1992, @doi [ ] 10.1086/116094 , https://ui.adsabs.harvard.edu/abs/1992AJ....103..691H 103, 691

    1992 doi

  26. [34]

    P., 1926, @doi [ ] 10.1086/143018 , 64, 321

    Hubble E. P., 1926, @doi [ ] 10.1086/143018 , 64, 321

    1926 doi

  27. [35]

    P., 1936, Realm of the Nebulae

    Hubble E. P., 1936, Realm of the Nebulae

    1936

  28. [36]

    Ishak B., 2017, @doi [Contemporary Physics] 10.1080/00107514.2016.1246478 , 58, 99

    2017

  29. [37]

    Kamdar H., Turk M., Brunner R., 2016, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stv2310 , 455, 642

    2016 doi

  30. [38]

    A., Wang S., Gruendl R., Jennings E., Zheng H., 2019, Phys

    Khan A., Huerta E. A., Wang S., Gruendl R., Jennings E., Zheng H., 2019, Phys. Lett., B795, 248

    2019

  31. [39]

    J., Brunner R

    Kim E. J., Brunner R. J., 2017, @doi [ ] 10.1093/mnras/stw2672 , 464, 4463

    2017 doi

  32. [40]

    P., Ba J., 2014, Adam: A Method for Stochastic Optimization ( @eprint arXiv 1412.6980 )

    Kingma D. P., Ba J., 2014, Adam: A Method for Stochastic Optimization ( @eprint arXiv 1412.6980 )

    2014 arXiv

  33. [41]

    E., 2012, in Proceedings of the 25th International Conference on Neural Information Processing Systems - Volume 1

    Krizhevsky A., Sutskever I., Hinton G. E., 2012, in Proceedings of the 25th International Conference on Neural Information Processing Systems - Volume 1. NIPS'12. Curran Associates Inc., USA, pp 1097--1105, http://dl.acm.org/citation.cfm?id=2999134.2999257

    2012

  34. [42]

    R., McKee C

    Krumholz M. R., McKee C. F., Bland -Hawthorn J., 2018, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2018arXiv181201615K p. arXiv:1812.01615

    2018 arXiv

  35. [43]

    LSST Science Collaboration et al., 2009, preprint ( @eprint arXiv 0912.0201 )

    2009 arXiv

  36. [44]

    S., 2002, @doi [ ] 10.1086/342381 , http://adsabs.harvard.edu/abs/2002AJ....124.1393L 124, 1393

    Larsen S. S., 2002, @doi [ ] 10.1086/342381 , http://adsabs.harvard.edu/abs/2002AJ....124.1393L 124, 1393

    2002 doi

  37. [45]

    LeCun Y., Bengio Y., Hinton G., 2015, nature, 521, 436

    2015

  38. [46]

    Ma ek K., et al 2013, @doi [ ] 10.1051/0004-6361/201321447 , 557, A16

    2013 doi

  39. [47]

    Messa M., et al., 2018, @doi [ ] 10.1093/mnras/stx2403 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.473..996M 473, 996

    2018 doi

  40. [48]

    Paszke A., et al., 2017, in NIPS-W

    2017

  41. [49]

    F., McMillan S

    Portegies Zwart S. F., McMillan S. L. W., Gieles M., 2010, @doi [ ] 10.1146/annurev-astro-081309-130834 , https://ui.adsabs.harvard.edu/abs/2010ARA&A..48..431P 48, 431

    2010 doi

  42. [50]

    Russakovsky O., et al., 2015, @doi [International Journal of Computer Vision (IJCV)] 10.1007/s11263-015-0816-y , 115, 211

    2015 doi

  43. [51]

    E., et al., 2014, @doi [ ] 10.1088/0004-6256/148/2/33 , http://adsabs.harvard.edu/abs/2014AJ....148...33R 148, 33

    Ryon J. E., et al., 2014, @doi [ ] 10.1088/0004-6256/148/2/33 , http://adsabs.harvard.edu/abs/2014AJ....148...33R 148, 33

    2014 doi

  44. [52]

    E., et al., 2017, @doi [The Astrophysical Journal] 10.3847/1538-4357/aa719e , 841, 92

    Ryon J. E., et al., 2017, @doi [The Astrophysical Journal] 10.3847/1538-4357/aa719e , 841, 92

    2017 doi

  45. [53]

    W., Whitmore B

    Schweizer F., Miller B. W., Whitmore B. C., Fall S. M., 1996, @doi [ ] 10.1086/118146 , https://ui.adsabs.harvard.edu/abs/1996AJ....112.1839S 112, 1839

    1996 doi

  46. [54]

    Sevilla-Noarbe I., Etayo-Sotos P., 2015, @doi [Astronomy and Computing] 10.1016/j.ascom.2015.03.010 , 11, 64

    2015 doi

  47. [55]

    E., 1948, Bell system technical journal, 27, 379

    Shannon C. E., 1948, Bell system technical journal, 27, 379

    1948

  48. [57]

    arXiv:1409.1556

    Simonyan K., Zisserman A., 2014b, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2014arXiv1409.1556S p. arXiv:1409.1556

    arXiv

  49. [58]

    Solarz A., Bilicki M., Gromadzki M., Pollo A., Durkalec A., Wypych M., 2017, @doi [ ] 10.1051/0004-6361/201730968 , 606, A39

    2017 doi

  50. [59]

    A., Hanisch R

    Suchkov A. A., Hanisch R. J., Margon B., 2005, @doi [ ] 10.1086/497363 , 130, 2439

    2005 doi

  51. [60]

    arXiv:1409.4842

    Szegedy C., et al., 2014, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2014arXiv1409.4842S p. arXiv:1409.4842

    2014 arXiv

  52. [61]

    C., de Carvalho R

    Vasconcellos E. C., de Carvalho R. R., Gal R. R., LaBarbera F. L., Capelato H. V., Frago Campos Velho H., Trevisan M., Ruiz R. S. R., 2011, @doi [ ] 10.1088/0004-6256/141/6/189 , 141, 189

    2011 doi

  53. [62]

    M., Djorgovski S., 1995, @doi [ ] 10.1086/117459 , 109, 2401

    Weir N., Fayyad U. M., Djorgovski S., 1995, @doi [ ] 10.1086/117459 , 109, 2401

    1995 doi

  54. [63]

    C., Sparks W

    Whitmore B. C., Sparks W. B., Lucas R. A., Macchetto F. D., Biretta J. A., 1995, @doi [ ] 10.1086/309788 , https://ui.adsabs.harvard.edu/abs/1995ApJ...454L..73W 454, L73

    1995 doi

  55. [64]

    C., et al., 2014, @doi [ ] 10.1088/0004-637X/795/2/156 , http://adsabs.harvard.edu/abs/2014ApJ...795..156W 795, 156

    Whitmore B. C., et al., 2014, @doi [ ] 10.1088/0004-637X/795/2/156 , http://adsabs.harvard.edu/abs/2014ApJ...795..156W 795, 156

    2014 doi

  56. [65]

    C., et al., 2016, @doi [ ] 10.3847/0004-6256/151/6/134 , https://ui.adsabs.harvard.edu/abs/2016AJ....151..134W 151, 134

    Whitmore B. C., et al., 2016, @doi [ ] 10.3847/0004-6256/151/6/134 , https://ui.adsabs.harvard.edu/abs/2016AJ....151..134W 151, 134

    2016 doi

  57. [66]

    W., et al., 2013, @doi [ ] 10.1093/mnras/stt1458 , https://ui.adsabs.harvard.edu/abs/2013MNRAS.435.2835W 435, 2835

    Willett K. W., et al., 2013, @doi [ ] 10.1093/mnras/stt1458 , https://ui.adsabs.harvard.edu/abs/2013MNRAS.435.2835W 435, 2835

    2013 doi

  58. [67]

    de Vaucouleurs G., 1963, @doi [ ] 10.1086/190084 , https://ui.adsabs.harvard.edu/abs/1963ApJS....8...31D 8, 31

    1963 doi

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.