Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Optimising Automatic Morphological Classification of Galaxies with Machine Learning and Deep Learning using Dark Energy Survey Imaging

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read With pixel input alone, a convolutional neural network outperforms nine other machine-learning methods at separating ellipticals from spirals on Dark Energy Survey images, reaching an average accuracy above 0.99 once roughly 2.5 per cent…

desk verdict The method comparison is useful and likely correct; the headline accuracy after 'correcting' Galaxy Zoo labels is circular and should not be trusted. read the letter →

arxiv 1908.03610 v2 pith:G6723AUC submitted 2019-08-09 astro-ph.GA astro-ph.IM

classification astro-ph.GAastro-ph.IM
keywords galaxymorphologymachinelearningconvolutionalneuralnetworksDarkEnergySurveyZoo1pixelinputlenticulargalaxiesclassificationaccuracy
topics Dark Energy
open problems Dark Energy
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

At stake is which supervised machine-learning method should be trusted to classify millions of galaxy images automatically. Using about 2,800 Dark Energy Survey galaxies with visual labels from Galaxy Zoo 1, the paper pits ten methods against one another on a two-way problem: elliptical versus spiral. It argues that the convolutional neural network is the most successful of these when fed raw image pixels, and that its mistakes are diagnostic: high-confidence mismatches are usually cases where the visual label itself is wrong, while low-confidence objects are mostly lenticular (S0) galaxies. With the flagged label errors corrected, the CNN reaches an average test accuracy above 0.99, with 0.994 as the best single run. The practical payoff would be a pre-trained model able to sort DES-level images of millions of galaxies without new human labelling.

What carries the argument

The load-bearing object is a convolutional neural network with three convolutional layers (32, 64, and 128 filters), each followed by max pooling, two fully connected hidden layers of 1024 units with dropout, and a two-class softmax output. A distinctive input mode, called the combination input, stacks the raw linearly-scaled $50 \times 50$ stamp together with its Histogram of Oriented Gradients (HOG) feature image so the CNN reads both at once; this outperforms either input alone. Training uses rotated copies with added Gaussian noise, balanced elliptical/spiral counts, and a classification criterion $p \geq 0.8$ that separates confident galaxies from uncertain ones. The uncertainty class is where the paper's discovery of lenticulars emerges.

What would settle it

Have at least three independent experienced classifiers, blind to both the CNN probabilities and the Galaxy Zoo labels, inspect the DES images of the 22 high-confidence mismatches and the 8 suspected misclassifications shown in the paper's Figures 11 and 15. If a substantial fraction of the 'confirmed' relabelled galaxies do not show unambiguous structures agreeing with the CNN, the corrected-label accuracy of 0.994 overstates the method.

Watch

Extended reading notes

Core claim

The paper's central claim is that, for binary elliptical/spiral classification from image pixels alone on Dark Energy Survey stamps, a convolutional neural network outperforms K-nearest neighbours, logistic regression, support vector machines, random forests, multi-layer perceptrons, and their restricted-Boltzmann-machine variants. With about 2,800 Galaxy Zoo 1 labelled galaxies rotated into roughly 100,000 training samples, the CNN reaches an accuracy near 0.95 with balanced data and combined raw-plus-HOG input. Raising the classification threshold to $p \geq 0.8$ lifts accuracy to about 0.987 by marking low-confidence objects as uncertain, and most of those uncertain objects look like lenticulars on DES images. The paper also claims that about 2.5 per cent of the Galaxy Zoo 1 labels in this sample are wrong, visible when DES's sharper, deeper images expose structure that SDSS lacked; after correcting those labels and retraining, the average accuracy over five runs exceeds 0.99.

Load-bearing premise

The argument assumes that the galaxies the CNN repeatedly flags as 'confirmed misclassified' truly have wrong Galaxy Zoo 1 labels, so that relabelling them and then reporting accuracy is a fair test rather than circular self-confirmation; this enters where the model's own repeated failures decide the corrected truth (Section 5.2.4).

Editorial extensions

If this is right

  • A pre-trained CNN on DES-style pixel stamps can be applied directly to millions of galaxies, giving binary morphology without additional human labels.
  • Balancing the training set by class is necessary when using pixel input; unbalanced sets systematically depress elliptical recall in most methods.
  • The uncertainty channel, $p < 0.8$, is an inexpensive way to flag objects that need human inspection, and in practice those objects are dominated by lenticulars.
  • Galaxy Zoo 1 labels carry measurable contamination, about 2.5 per cent in this sample, and machine-learning triage can flag suspects for reclassification in future surveys.
  • HOG features help most pixel-based methods but not K-nearest neighbours, and their benefit largely disappears when a restricted Boltzmann machine already compresses the features.

Reading between the lines

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

  • Because the same CNN defines which labels are 'confirmed misclassifications' and then measures its own accuracy on the corrected labels, the 0.994 figure is best read as an upper bound until independent visual confirmation of those roughly 2.5 per cent of objects.
  • If the lenticular finding generalizes, the probability gap between the two softmax outputs can be used as a rough morphological axis: S0 candidates, mergers, and edge-on discs may all collect at intermediate probabilities, giving three effective classes from a binary trainer.
  • A direct extension would ask whether the corrected labels also improve the non-CNN methods, since the paper retrains only the CNN after purification; the ranking of the ten methods could shift if all of them received the cleaner training set.
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

3 major / 5 minor

Summary. This paper presents a systematic comparison of ten supervised machine learning methods (including CNN, KNN, LR, SVM, RF, MLPC, and their variants with RBMs) for binary morphological classification of galaxies into ellipticals and spirals using DES imaging with Galaxy Zoo 1 (GZ1) visual labels. The authors evaluate pixel inputs, HOG features, and a CNN-specific combination input, and consistently find CNN to be the best method. The paper further investigates misclassifications, identifies lenticular galaxies as low-confidence objects, and claims that after purifying the training and test labels of ~2.5% of galaxies misclassified by GZ1, CNN reaches an average accuracy of 0.991 (best 0.994). The central claims are the CNN ranking and the >0.99 accuracy figure.

Significance. If the >0.99 accuracy claim were valid, this would be a strong benchmark for automated morphological classification on DES-like imaging and a useful guide for method selection. The paper deserves credit for a systematic, consistent comparison: a fixed 1,000-galaxy test set, three to five reruns, and ROC curves with uncertainty bands are used throughout. The rediscovery of lenticulars as uncertain objects is an interesting empirical result. However, the headline accuracy is not an unbiased measurement as presented; the method comparison and the accuracy claim must be evaluated separately, and the latter currently rests on a circular relabeling procedure and on discarding uncertain galaxies.

major comments (3)
  1. [Section 5.2.4, Table 8, Table 9] The reported accuracies of 0.991 and 0.994 after 'correcting' GZ1 labels are not independent measurements. The confirmed misclassifications are defined by the CNN's own repeated failures: a galaxy is 'confirmed' if it appears at least four times in total failures and at least once among high-probability failures (Table 8). These same labels are then changed in the test set and used to compute the accuracy. Because the model's disagreements define the ground truth, the evaluation is circular, and the abstract's claim of approximately 0.99 accuracy is not supported as an unbiased estimate. I recommend either independently validating the relabeled objects (e.g., by expert visual classification of the full confirmed list, not only the three unanimous-disagreement galaxies in Section 4.5) or clearly reframing the abstract and conclusions to present the corrected-label accuracy as a post-hoc consistency check rather than the model's predictive accuracy.
  2. [Section 5.1, Table 5, Table 9] The headline accuracy is computed only for the N_classifiable subset (p >= 0.8) and excludes 'uncertain' galaxies. For example, Table 5 reports accuracy 0.974 on 912 classifiable galaxies, while 88 of the 1,000 test galaxies are excluded; Table 9 similarly reports 0.991 on 976 galaxies with 16 uncertain. The abstract's '~0.99' is therefore conditional on discarding the hardest cases. If the uncertain galaxies are counted as errors, the full-sample accuracy is substantially lower (e.g., in Table 5, dataset 2: 912 * 0.974 / 1000 is about 0.889). The abstract and conclusions should state explicitly that the accuracy applies to the high-confidence subsample, not to the full test sample.
  3. [Section 5.2.4] The claimed ~2.5% misclassification rate in GZ1 is derived from the CNN-based frequency criteria, but the paper does not describe an independent verification procedure for the full 'confirmed' list (Fig. 14). The authors visually confirm three unanimous-disagreement galaxies in Section 4.5, but no such external verification is reported for the 25 or so confirmed objects. Without independent labels or a described expert inspection protocol, the misclassification rate is also a product of the CNN's behavior and is not independently established.
minor comments (5)
  1. [Abstract] The phrase 'or a investigation' contains a grammar error; it should be 'or an investigation'.
  2. [Table 5 and Table 9] The column header 'Nuncetain' appears to be a typo for 'Nuncertain'.
  3. [Section 2.1.1] The sentence describing the added Gaussian noise is garbled: 'it is big enough to make a detectable but change of pixel values' should be rewritten for clarity.
  4. [Section 5.2.4] The description of the purification procedure would benefit from a clearer account of the number of iterations: the text states 'After carrying out this purification twice,' but the preceding description of rerunning five times on each new training set is ambiguous about how the two iterations relate to the five reruns.
  5. [Table 3] The table footnote refers to 'the sixth method' when describing 'CNN (GPU)', but CNN (GPU) is not the sixth row; this should be corrected.

Circularity Check

2 steps flagged · score 6.0 of 10

Headline 0.99+ accuracy is partly circular: 'corrected' test labels are generated by the CNN's own failure criteria, and uncertain galaxies are excluded from the denominator.

  1. fitted input called prediction [Section 5.2.4, Table 8 and Table 9 (with Section 5.2.1)]
    "We purify our training set by excluding the suspected misclassified galaxies then use the criteria shown in Table 8 to confirm or dismiss our suspected misclassifications... We then change the label of the confirmed misclassified galaxies in the testing set as well. ... We obtain the accuracy of 0.994 for the best model within five reruns, and the average accuracy of five reruns is 0.991."

    The 'confirmed' GZ1 misclassifications are selected by the CNN's own disagreement frequency: Table 8 requires appearing at least 4 times in total failures and at least once in high-probability failures, where 'failures' mean disagreement with the GZ1 label. The new 'corrected' test labels are therefore essentially the labels the CNN tends to output. Retraining with those labels, changing the same labels in the test set, and then reporting 0.991/0.994 accuracy measures how well the CNN agrees with a ground truth partly manufactured from the CNN's own predictions. The paper presents only a few example images (Fig. 12), not an independent expert reclassification of the full confirmed list, so the corrected labels are not externally grounded.

  2. other [Section 5.1, Table 5 and Table 9 caption; compare with abstract]
    "The definition of accuracy (Equation 5) and the meaning of R01 are same as in Table 4. Nclassifiable and Nuncertain are the number of testing data which are classifiable (namely p≥ 0.8) and uncertain (probabilities of both types(p) < 0.8), respectively."

    The ~0.99 accuracy is computed only over the Nclassifiable subset after applying the CNN's own confidence threshold p>=0.8; the 'uncertain' galaxies (e.g. 42–88 of the 1,000 test galaxies in Table 5, and 16–19 in Table 9) are removed from the denominator. This makes the headline accuracy conditional on the classifier's own probability estimates, so the hardest cases, by the model's own judgment, do not count as errors. If those uncertain galaxies were counted as failures, the full-sample accuracy would be lower than the claimed ~0.99. The metric is therefore partly self-selected rather than a full-sample measurement.

full rationale

The comparison of the ten methods on the original GZ1 labels is a legitimate, self-contained benchmark: all methods are trained on the same augmented data and tested on the same 1,000 objects, so the conclusion that CNN outperforms the other methods is not itself circular. However, the paper's most public-facing claims—accuracy of ~0.99 and best accuracy of 0.994—come from Section 5.2.4, where test labels are changed to the CNN's own predicted classes for objects selected by the CNN's disagreement frequency, with no independent expert relabeling of the complete confirmed list. The accuracy measured after that relabeling is an internal consistency check, not an independent accuracy measurement on the original labels. The additional restriction to p>=0.8 'classifiable' galaxies removes the uncertain cases from the denominator, so the reported figure is also conditional on the model's own confidence threshold. The 0.99+ numbers should therefore be re-validated on a blind, independently labeled test sample before being accepted as evidence of near-perfect classification. No load-bearing self-citation chain or imported uniqueness theorem appears in the paper; the circularity is confined to the relabeled evaluation and the conditional accuracy reporting.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central accuracy claim depends on a chain of labeled data: GZ1 consensus labels, augmented copies, and finally relabeled galaxies chosen by the CNN's own failure frequency. The hyperparameters are standard but numerous, and the relabeling criteria in Table 8 are the most consequential free choice because they directly determine the reported 0.99+ accuracy.

free parameters (8)
  • KNN k = 5
    Selected by plotting accuracy versus k (Section 3.2); affects the decision boundary for the KNN baseline.
  • Random Forest n_estimators = 200
    Selected by plotting accuracy versus n_estimators (Section 3.5); affects the ensemble baseline.
  • SVM nu = not stated
    Set via GridSearchCV (Section 3.4); the final value is not reported, so the SVM result is not fully specified.
  • RBM hidden units and learning rate = 1024 units, lr 0.001, 500 iterations
    Fixed values for all RBM-augmented methods (Section 3.1); iterations chosen where training log-likelihood peaks.
  • MLPC architecture = 2 hidden layers x 1024 units, lr 0.001
    Architecture chosen for the neural network baseline (Section 3.6); no sensitivity analysis is given.
  • CNN architecture = 3 conv layers (32/64/128, filters 3,3,2), pooling 2, 2 hidden x1024, dropout 0.5, lr 0.001
    Architecture follows Dieleman et al. (2015) with modifications (Section 3.7, Fig. 4); this is the central classifier.
  • Probability threshold p = 0.8
    Used to separate classifiable from uncertain galaxies (Section 5.1, Table 5); the 0.97 and 0.99 accuracy numbers depend on this choice.
  • Confirmed misclassified criteria = appears >=4 times in total failures and >=1 in high-p failures
    These frequency thresholds in Table 8 define which Galaxy Zoo labels are changed before the final accuracy is computed (Section 5.2.4).
assumptions (4)
  • domain assumption Galaxy Zoo 1 labels with over 80 percent volunteer agreement and Bamford et al. (2009) debiasing are valid ground truth for initial training and testing.
    Section 2 states that the matching uses only high-agreement, bias-corrected GZ1 classifications; the paper later argues that 2.5 percent of these are wrong.
  • domain assumption Rotated and Gaussian-noised copies of a galaxy image retain the same morphological class as the original.
    Section 2.1.1 applies 36 rotations and added noise to augment the sample to about 100,000 images; if rotation changes apparent morphology, the labels leak.
  • domain assumption DES i-band coadd images are sufficient to visually confirm the true morphology of galaxies, including cases where SDSS-based GZ1 labels are wrong.
    Section 5.2.4 uses visual inspection of DES images as the arbiter for confirmed GZ1 misclassifications; this assumes human judgment on DES is more reliable than the original GZ1 consensus.
  • ad hoc to paper The frequency criteria in Table 8 identify genuine GZ1 misclassifications rather than mere CNN instabilities.
    The thresholds (four or more appearances, at least one high-probability failure) were chosen by the authors to define confirmed suspects; no independent validation of these criteria is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Optimising Automatic Morphological Classification of Galaxies with Machine Learning and Deep Learning using Dark Energy Survey Imaging." pith.science (2026). https://pith.science/paper/G6723AUC

@misc{pith2026190803610,
  author       = {Pith},
  title        = {Pith review of: Optimising Automatic Morphological Classification of Galaxies with Machine Learning and Deep Learning using Dark Energy Survey Imaging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G6723AUC}},
  note         = {Machine review of arXiv:1908.03610}
}
abstract

There are several supervised machine learning methods used for the application of automated morphological classification of galaxies; however, there has not yet been a clear comparison of these different methods using imaging data, or a investigation for maximising their effectiveness. We carry out a comparison between several common machine learning methods for galaxy classification (Convolutional Neural Network (CNN), K-nearest neighbour, Logistic Regression, Support Vector Machine, Random Forest, and Neural Networks) by using Dark Energy Survey (DES) data combined with visual classifications from the Galaxy Zoo 1 project (GZ1). Our goal is to determine the optimal machine learning methods when using imaging data for galaxy classification. We show that CNN is the most successful method of these ten methods in our study. Using a sample of $\sim$2,800 galaxies with visual classification from GZ1, we reach an accuracy of $\sim$0.99 for the morphological classification of Ellipticals and Spirals. The further investigation of the galaxies that have a different ML and visual classification but with high predicted probabilities in our CNN usually reveals an the incorrect classification provided by GZ1. We further find the galaxies having a low probability of being either spirals or ellipticals are visually Lenticulars (S0), demonstrating that supervised learning is able to rediscover that this class of galaxy is distinct from both Es and Spirals. We confirm that $\sim$2.5\% galaxies are misclassified by GZ1 in our study. After correcting these galaxies' labels, we improve our CNN performance to an average accuracy of over 0.99 (accuracy of 0.994 is our best result).

Figures

Figures reproduced from arXiv: 1908.03610 by the authors.

Figure 1
Figure 1. Pre-processing procedure pipeline. The pipeline starts from the initial coadd images, then we chop the coadd images into different sizes according to the size of galaxies. After rotation, we chop and downsize the images to the required sizes: 50 by 50 pixels. The details of the procedure is in Section 2.1 (y) direction of stamps. The magnitude and orientation of the gradient are calculated as below, |G| = q G2 x + G… view at source ↗
Figure 2
Figure 2. Examples of images from Histogram Oriented Gradient (HOG) with the cell size of 2 by 2 pixels. Left: HOG images. Right: original images in linear scale. Top: Spirals. Bottom: Ellipticals. algorithms when defining the boundaries between different classes. Therefore, we rescale the pixel values of each image (raw and HOG images) to the range between 0 and 1 through normalising by the maximum and minimum pixel value of… view at source ↗
Figure 3
Figure 3. Illustration of a neural networks. This structure is for illustration only and this includes one hidden layer, and two hidden units. Two input features, X1 and X2, work with the ac￾tivation functions, f 1 and f 2, then obtain the outputs, Y1 and Y2. layers which are able to extract notable features from the in￾put images by applying several filter matrices, and the other difference is the dimension of the input. Mos… view at source ↗
Figures from the paper (15 more)
Figure 4
Figure 4. Figure 4: The schematic overview of the architecture of CNN. The architecture starts from an input image with size 50 by 50 pixels, then three convolutional layers (filter: 32, 64, and 128). Each convolutional layer is followed a pooling layer. Two hidden layers with 1,024 hidde…
Figure 5
Figure 5. Figure 5: The confusion matrix. The x-axis label is the predicted label and the y-axis label is the true label. The ‘0’ means negative as well as Ellipticals type while ‘1’ represents positive signal and Spirals type in this study. training data, so the shape of their ROC curve …
Figure 6
Figure 6. Figure 6: The ROC curve of each method and each dataset using the raw images input (i). The abbreviation of the methods are the same as [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: The recalls of the Ellipticals and Spirals for all meth￾ods and the different types of the input data used. The colours represent the different datasets, while the different shape markers are the different methods. The different types of filled-points rep￾resent the di…
Figure 8
Figure 8. Figure 8: The ROC curve for different types of input within each method. Different colours are for different input types of data. Cyan, orange, and blue are for raw images (i), HOG images (ii), and combination input (iii), respectively. The lighter colour shadings show the scatt…
Figure 9
Figure 9. Figure 9: The average accuracy (Equation 5) of the three reruns versus each method with the different datasets and the different types of input shown. The y-axis is from 0.5 to 1.0. Colours represent different datasets such that yellow, orange, cyan, blue represents dataset 1, 2…
Figure 10
Figure 10. Figure 10: The accuracy versus the number of training data with different types of input. Different colours show different types of input such that cyan, orange, blue are for the raw images (i), the HOG images (ii), and the combination input (iii), respectively. The lighter colo…
Figure 11
Figure 11. Figure 11: The misclassified galaxies with high probabilities (p ≥ 0.8) comparing the classification of Galaxy Zoo 1 and our CNN. On the top of the images shows the probabilities of being Ellipticals, E(0) and Spirals, S(1) by our CNN. The line below the image shows the ID numbe…
Figure 12
Figure 12. Figure 12: Examples of the incorrect label from GZ1 with SDSS imaging. The figures under each number show the galaxy images of DES and SDSS, and their ID numbers. The label of ‘CNN’ shows the predicted label from our method, and which of ’GZ’ shows the label from the Galaxy Zoo …
Figure 13
Figure 13. Figure 13: Examples of the galaxies with low probabilities of classification as either spiral or elliptical. Top 1-12: these objects are turned out to be lenticular galaxies (S0) in cluster inspection. Bottom 1-12: the other types of galaxies. MNRAS 000, 1–17 (2020) [PITH_FULL_…
Figure 14
Figure 14. Figure 14: The confirmed list of the misclassified galaxies in the Galaxy Zoo 1 catalogue. The first row underneath the images is the ID numbers of galaxies, and the second row shows the classification by Galaxy Zoo (GZ) and our CNN (CNN). MNRAS 000, 1–17 (2020) [PITH_FULL_IMAG…
Figure 15
Figure 15. Figure 15: The suspected list of the misclassified galaxies in the Galaxy Zoo 1 catalogue. The first row underneath the images is the ID numbers of galaxies, and the second row shows the classification by Galaxy Zoo (GZ) and our CNN (CNN) [PITH_FULL_IMAGE:figures/full_fig_p022_…
Figure 16
Figure 16. Figure 16: The best testing result which we changed the label of the confirmed misclassified galaxies and excluded the suspected misclassified galaxies in both training and testing set. Top: Con￾fusion matrix. The ‘0’ means Ellipticals and ‘1’ represents Spirals. The colour bar …
Figure 17
Figure 17. Figure 17: Successful examples of classified Ellipticals. The ‘prob’ on the top of the images show the predicted probability of being Ellipticals [PITH_FULL_IMAGE:figures/full_fig_p023_17.png]
Figure 18
Figure 18. Figure 18: Successful examples of the classified Spirals. The ‘prob’ on the top of the images show the predicted probability of being Spirals. MNRAS 000, 1–17 (2020) [PITH_FULL_IMAGE:figures/full_fig_p023_18.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Deblending and Classifying Astronomical Sources with Mask R-CNN Deep Learning

    astro-ph.IM 2019-08 conditional novelty 6.0 of 10

    A Mask R-CNN network simultaneously detects, classifies, and deblends stars and galaxies in simulated and real DECam images, achieving 92% and 98% precision at 80% recall for stars and galaxies respectively.

  2. Uncertainty-Aware Deep Learning for the Ly$\alpha$ Forest: CNN-Based Absorber Detection and Characterization

    astro-ph.GA 2026-07 conditional novelty 5.5 of 10

    A sliding-window CNN recovers Lyα absorber locations and Voigt parameters from spectra, reproducing CDDF and b–N relations on mocks and, more weakly, on UVES data.

Reference graph

Works this paper leans on

71 extracted references · 32 canonical work pages · cited by 2 Pith papers

  1. [1]

    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.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTION or pop #1...

  2. [2]

    Abbott T. M. C., et al., 2018, @doi [ ] 10.3847/1538-4365/aae9f0 , https://ui.adsabs.harvard.edu/abs/2018ApJS..239...18A 239, 18

  3. [3]

    G., van den Bergh S., Nair P., 2003, @doi [ ] 10.1086/373919 , https://ui.adsabs.harvard.edu/abs/2003ApJ...588..218A 588, 218

    Abraham R. G., van den Bergh S., Nair P., 2003, @doi [ ] 10.1086/373919 , https://ui.adsabs.harvard.edu/abs/2003ApJ...588..218A 588, 218

  4. [4]

    E., Luo W., 2019, @doi [ ] 10.3847/1538-4357/ab16d9 , https://ui.adsabs.harvard.edu/abs/2019ApJ...877...58A 877, 58

    Avestruz C., Li N., Zhu H., Lightman M., Collett T. E., Luo W., 2019, @doi [ ] 10.3847/1538-4357/ab16d9 , https://ui.adsabs.harvard.edu/abs/2019ApJ...877...58A 877, 58

  5. [6]

    M., Brunner R

    Ball N. M., Brunner R. J., Myers A. D., Tcheng D., 2006, @doi [ ] 10.1086/507440 , https://ui.adsabs.harvard.edu/abs/2006ApJ...650..497B 650, 497

  6. [9]

    R., et al., 2018, @doi [ ] 10.1093/mnras/sty503 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.476.5516B 476, 5516

    Beck M. R., et al., 2018, @doi [ ] 10.1093/mnras/sty503 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.476.5516B 476, 5516

  7. [10]

    M., 2006, Pattern Recognition and Machine Learning (Information Science and Statistics)

    Bishop C. M., 2006, Pattern Recognition and Machine Learning (Information Science and Statistics). Springer-Verlag, Berlin, Heidelberg

  8. [11]

    P., 1997, @doi [Pattern Recognition] https://doi.org/10.1016/S0031-3203(96)00142-2 , 30, 1145

    Bradley A. P., 1997, @doi [Pattern Recognition] https://doi.org/10.1016/S0031-3203(96)00142-2 , 30, 1145

Show all 71 references
  1. [12]

    Learn.] 10.1023/A:1010933404324 , 45, 5–32

    Breiman L., 2001, @doi [Mach. Learn.] 10.1023/A:1010933404324 , 45, 5–32

  2. [13]

    Chopra P., Yadav S., 2017, @doi [Complex & Intelligent Systems] 10.1007/s40747-017-0054-8 , pp 1--11

  3. [14]

    J., 2003, @doi [ ] 10.1086/375001 , https://ui.adsabs.harvard.edu/abs/2003ApJS..147....1C 147, 1

    Conselice C. J., 2003, @doi [ ] 10.1086/375001 , https://ui.adsabs.harvard.edu/abs/2003ApJS..147....1C 147, 1

  4. [15]

    pp 273--297

    Cortes C., Vapnik V., 1995, in Machine Learning. pp 273--297

  5. [16]

    Cover T., Hart P., 1967, @doi [IEEE Transactions on Information Theory] 10.1109/TIT.1967.1053964 , 13, 21

  6. [17]

    J., 2007, k-Nearest Neighbour Classifiers

    Cunningham P., Delany S. J., 2007, k-Nearest Neighbour Classifiers

  7. [18]

    pp 886--893 vol

    Dalal N., Triggs B., 2005, in 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05). pp 886--893 vol. 1, @doi 10.1109/CVPR.2005.177

  8. [19]

    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

  9. [20]

    L., 2018, @doi [ ] 10.1093/mnras/sty338 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.476.3661D 476, 3661

    Dom \' nguez S \'a nchez H., Huertas-Company M., Bernardi M., Tuccillo D., Fischer J. L., 2018, @doi [ ] 10.1093/mnras/sty338 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.476.3661D 476, 3661

  10. [21]

    Drlica-Wagner A., et al., 2018, @doi [ ] 10.3847/1538-4365/aab4f5 , https://ui.adsabs.harvard.edu/abs/2018ApJS..235...33D 235, 33

  11. [22]

    Dubath P., et al., 2011, @doi [ ] 10.1111/j.1365-2966.2011.18575.x , https://ui.adsabs.harvard.edu/abs/2011MNRAS.414.2602D 414, 2602

  12. [23]

    M., Elyan E., 2014, @doi [Systems Science & Control Engineering] 10.1080/21642583.2014.956265 , 2, 602

    Fawagreh K., Gaber M. M., Elyan E., 2014, @doi [Systems Science & Control Engineering] 10.1080/21642583.2014.956265 , 2, 602

  13. [24]

    Fawcett T., 2006, @doi [Pattern Recognition Letters] https://doi.org/10.1016/j.patrec.2005.10.010 , 27, 861

  14. [25]

    L., 1989, International Statistical Review / Revue Internationale de Statistique, 57, 238

    Fix E., Hodges J. L., 1989, International Statistical Review / Revue Internationale de Statistique, 57, 238

  15. [26]

    Flaugher B., et al., 2015, @doi [ ] 10.1088/0004-6256/150/5/150 , https://ui.adsabs.harvard.edu/abs/2015AJ....150..150F 150, 150

  16. [27]

    Fukushima K., 1975, Biological Cybernetics, 20, 121

  17. [28]

    Fukushima K., 1980, @doi [Biological Cybernetics] 10.1007/bf00344251 , 36, 193

  18. [29]

    Fukushima K., Miyake S., Ito T., 1983, @doi [IEEE Transactions on Systems, Man, and Cybernetics] 10.1109/TSMC.1983.6313076 , SMC-13, 826

  19. [31]

    N., Lolling S

    Goderya S. N., Lolling S. M., 2002, @doi [ ] 10.1023/A:1015193432240 , https://ui.adsabs.harvard.edu/abs/2002Ap&SS.279..377G 279, 377

  20. [32]

    E., 2002, @doi [Neural Comput.] 10.1162/089976602760128018 , 14, 1771–1800

    Hinton G. E., 2002, @doi [Neural Comput.] 10.1162/089976602760128018 , 14, 1771–1800

  21. [33]

    E., Sun Y., Davey N., 2018, @doi [ ] 10.1093/mnras/stx2351 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.473.1108H 473, 1108

    Hocking A., Geach J. E., Sun Y., Davey N., 2018, @doi [ ] 10.1093/mnras/stx2351 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.473.1108H 473, 1108

  22. [34]

    Department of Computer Science, National Taiwan University, http://www.csie.ntu.edu.tw/ cjlin/papers.html

    Hsu C.-W., Chang C.-C., Lin C.-J., 2003, Technical report, A Practical Guide to Support Vector Classification, http://www.csie.ntu.edu.tw/ cjlin/papers.html . Department of Computer Science, National Taiwan University, http://www.csie.ntu.edu.tw/ cjlin/papers.html

  23. [35]

    P., 1926, @doi [ ] 10.1086/143018 , https://ui.adsabs.harvard.edu/abs/1926ApJ....64..321H 64, 321

    Hubble E. P., 1926, @doi [ ] 10.1086/143018 , https://ui.adsabs.harvard.edu/abs/1926ApJ....64..321H 64, 321

  24. [36]

    Huertas-Company M., Rouan D., Tasca L., Soucail G., Le F \`e vre O., 2008, @doi [ ] 10.1051/0004-6361:20078625 , https://ui.adsabs.harvard.edu/abs/2008A&A...478..971H 478, 971

  25. [37]

    Huertas-Company M., et al., 2009, @doi [ ] 10.1051/0004-6361/200811255 , https://ui.adsabs.harvard.edu/abs/2009A&A...497..743H 497, 743

  26. [38]

    Huertas-Company M., Aguerri J. A. L., Bernardi M., Mei S., S \'a nchez Almeida J., 2011, @doi [ ] 10.1051/0004-6361/201015735 , https://ui.adsabs.harvard.edu/abs/2011A&A...525A.157H 525, A157

  27. [39]

    Huertas-Company M., et al., 2015, @doi [ ] 10.1088/0067-0049/221/1/8 , https://ui.adsabs.harvard.edu/abs/2015ApJS..221....8H 221, 8

  28. [40]

    Huertas-Company M., et al., 2018, @doi [ ] 10.3847/1538-4357/aabfed , https://ui.adsabs.harvard.edu/abs/2018ApJ...858..114H 858, 114

  29. [41]

    M., Hogg D

    Huppenkothen D., Heil L. M., Hogg D. W., Mueller A., 2017, @doi [ ] 10.1093/mnras/stw3190 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.466.2364H 466, 2364

  30. [42]

    M., Hegadi R

    Kamble P. M., Hegadi R. S., 2015, @doi [Procedia Computer Science] https://doi.org/10.1016/j.procs.2015.03.137 , 45, 266

  31. [43]

    Kov \'a cs A., Szapudi I., 2015, @doi [ ] 10.1093/mnras/stv063 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.448.1305K 448, 1305

  32. [44]

    D., Polsterer K., Hoecker M., 2015, @doi [ ] 10.1051/0004-6361/201424801 , https://ui.adsabs.harvard.edu/abs/2015A&A...576A.132K 576, A132

    K \"u gler S. D., Polsterer K., Hoecker M., 2015, @doi [ ] 10.1051/0004-6361/201424801 , https://ui.adsabs.harvard.edu/abs/2015A&A...576A.132K 576, A132

  33. [45]

    J., Storrie-Lombardi M

    Lahav O., Naim A., Sodr \'e L. J., Storrie-Lombardi M. C., 1996, @doi [ ] 10.1093/mnras/283.1.207 , https://ui.adsabs.harvard.edu/abs/1996MNRAS.283..207L 283, 207

  34. [46]

    Lecun Y., Bottou L., Bengio Y., Haffner P., 1998, @doi [Proceedings of the IEEE] 10.1109/5.726791 , 86, 2278

  35. [47]

    J., et al., 2008, @doi [ ] 10.1111/j.1365-2966.2008.13689.x , https://ui.adsabs.harvard.edu/abs/2008MNRAS.389.1179L 389, 1179

    Lintott C. J., et al., 2008, @doi [ ] 10.1111/j.1365-2966.2008.13689.x , https://ui.adsabs.harvard.edu/abs/2008MNRAS.389.1179L 389, 1179

  36. [48]

    Lintott C., et al., 2011, @doi [ ] 10.1111/j.1365-2966.2010.17432.x , https://ui.adsabs.harvard.edu/abs/2011MNRAS.410..166L 410, 166

  37. [49]

    H., Hakala P

    Maehoenen P. H., Hakala P. J., 1995, @doi [ ] 10.1086/309697 , https://ui.adsabs.harvard.edu/abs/1995ApJ...452L..77M 452, L77

  38. [50]

    Chapman and Hall/CRC Monographs on Statistics and Applied Probability Series, Chapman & Hall, http://books.google.com/books?id=h9kFH2\_FfBkC

    McCullagh P., Nelder J., 1989, Generalized Linear Models, Second Edition. Chapman and Hall/CRC Monographs on Statistics and Applied Probability Series, Chapman & Hall, http://books.google.com/books?id=h9kFH2\_FfBkC

  39. [51]

    B., et al., 2019, @doi [ ] 10.1051/0004-6361/201832797 , https://ui.adsabs.harvard.edu/abs/2019A&A...625A.119M 625, A119

    Metcalf R. B., et al., 2019, @doi [ ] 10.1051/0004-6361/201832797 , https://ui.adsabs.harvard.edu/abs/2019A&A...625A.119M 625, A119

  40. [52]

    J., Storrie-Lombardi M

    Naim A., Lahav O., Sodre L. J., Storrie-Lombardi M. C., 1995, @doi [ ] 10.1093/mnras/275.3.567 , https://ui.adsabs.harvard.edu/abs/1995MNRAS.275..567N 275, 567

  41. [53]

    C., Stockwell E

    Odewahn S. C., Stockwell E. B., Pennington R. L., Humphreys R. M., Zumach W. A., 1992, @doi [ ] 10.1086/116063 , https://ui.adsabs.harvard.edu/abs/1992AJ....103..318O 103, 318

  42. [54]

    Orr M. J. L., Science C. F. C., 1996, Technical report, Introduction to radial basis function networks

  43. [55]

    arXiv:1201.0490

    Pedregosa F., et al., 2012, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2012arXiv1201.0490P p. arXiv:1201.0490

  44. [56]

    L., Gieseke F., Kramer O., 2012, Galaxy Classification without Feature Extraction

    Polsterer K. L., Gieseke F., Kramer O., 2012, Galaxy Classification without Feature Extraction . p. 561

  45. [57]

    Powers D. M. W., 2011, Journal of Machine Learning Technologies, 2, 37

  46. [58]

    Rosenblatt F., 1958, Psychological Review, pp 65--386

  47. [59]

    E., Hinton G

    Rumelhart D. E., Hinton G. E., Williams R. J., 1986, @doi [ ] 10.1038/323533a0 , https://ui.adsabs.harvard.edu/abs/1986Natur.323..533R 323, 533

  48. [60]

    ICML ’07

    Salakhutdinov R., Mnih A., Hinton G., 2007, in Proceedings of the 24th International Conference on Machine Learning. ICML ’07. Association for Computing Machinery, New York, NY, USA, p. 791–798, @doi 10.1145/1273496.1273596 , https://doi.org/10.1145/1273496.1273596

  49. [61]

    J., 2001, Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond

    Scholkopf B., Smola A. J., 2001, Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond. MIT Press, Cambridge, MA, USA

  50. [62]

    Shamir L., 2009, @doi [ ] 10.1111/j.1365-2966.2009.15366.x , https://ui.adsabs.harvard.edu/abs/2009MNRAS.399.1367S 399, 1367

  51. [63]

    D., Fukunaga K., 1981, IEEE Trans

    Short R. D., Fukunaga K., 1981, IEEE Trans. Information Theory, 27, 622

  52. [64]

    Shu C., Ding X., Fang C., 2011, @doi [Tsinghua Science and Technology] 10.1016/S1007-0214(11)70032-3 , 16, 216

  53. [65]

    MIT Press, Cambridge, MA, USA, p

    Smolensky P., 1986, Information Processing in Dynamical Systems: Foundations of Harmony Theory. MIT Press, Cambridge, MA, USA, p. 194–281

  54. [66]

    Sreejith S., et al., 2018, @doi [ ] 10.1093/mnras/stx2976 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.474.5232S 474, 5232

  55. [67]

    C., Lahav O., Sodre L

    Storrie-Lombardi M. C., Lahav O., Sodre L. J., Storrie-Lombardi L. J., 1992, @doi [ ] 10.1093/mnras/259.1.8P , https://ui.adsabs.harvard.edu/abs/1992MNRAS.259P...8S 259, 8P

  56. [68]

    M., 1950, @doi [Mind] 10.1093/mind/LIX.236.433 , LIX, 433

    TURING A. M., 1950, @doi [Mind] 10.1093/mind/LIX.236.433 , LIX, 433

  57. [69]

    N., 1995, The Nature of Statistical Learning Theory

    Vapnik V. N., 1995, The Nature of Statistical Learning Theory. Springer-Verlag, Berlin, Heidelberg

  58. [70]

    M., Djorgovski S., 1995, @doi [ ] 10.1086/117459 , https://ui.adsabs.harvard.edu/abs/1995AJ....109.2401W 109, 2401

    Weir N., Fayyad U. M., Djorgovski S., 1995, @doi [ ] 10.1086/117459 , https://ui.adsabs.harvard.edu/abs/1995AJ....109.2401W 109, 2401

  59. [71]

    Werbos P., John P., 1974

  60. [72]

    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

  61. [73]

    Y., 2003, in In ICML 2003 Workshop on Learning from Imbalanced Data Sets

    Wu G., Chang E. Y., 2003, in In ICML 2003 Workshop on Learning from Imbalanced Data Sets. pp 49--56

  62. [74]

    Zanaty E., 2012, @doi [Egyptian Informatics Journal] https://doi.org/10.1016/j.eij.2012.08.002 , 13, 177

  63. [75]

    de la Calleja J., Fuentes O., 2004, @doi [ ] 10.1111/j.1365-2966.2004.07442.x , https://ui.adsabs.harvard.edu/abs/2004MNRAS.349...87D 349, 87

Pith tools

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