Pith. sign in

REVIEW 4 major objections 4 minor 77 references

Multiple machine-learning as a powerful tool for the star clusters analysis

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

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

Pith's one-line read A machine-learning ensemble shows NGC 1605 is really two star clusters merging into one.

desk verdict The ML pipeline is real, but the two clusters are never actually separated; the binary-cluster conclusion is built on a selection that guarantees the two samples overlap, and the paper contradicts itself on which cluster is which age. read the letter →

arxiv 2506.13951 v1 pith:HKL6V7Y4 submitted 2025-06-16 astro-ph.GA astro-ph.IMastro-ph.SR

Denilso Camargo This is my paper · ORCID
classification astro-ph.GAastro-ph.IMastro-ph.SR
keywords openstarclustersbinaryclustermergermachinelearningGaiaEDR3NGC1605stellarmembershipassignmentisochronefitting
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 paper claims that running several established machine-learning clustering algorithms together on Gaia-EDR3 astrometry can separate star clusters that overlap so heavily that they look like a single object. Applied to NGC 1605, the combined pipeline recovers two genuine open clusters — NGC1605a at about 2 Gyr and NGC1605b at about 600 Myr, both at roughly 2.75 kpc — whose probable member stars overlap widely in position and motion. The paper reads that overlap as the signature of an old binary cluster in an advanced merger, whose end product would be a single cluster coherent in space and kinematics but holding two different ages. If the claim holds, the same tool could be used to find and dissect other merged or entangled clusters across the Galaxy, and the result would strengthen the case that some old open clusters are not single objects.

What carries the argument

The engine of the analysis is the MMLM pipeline, a combination of established algorithms rather than any single method: pyUPMASK and GMM supply membership probabilities, ASteCA fits synthetic color-magnitude diagrams to derive parameters, HDBSCAN isolates overdensities in the 5D astrometric space, Kmeans (classical and fuzzy) partitions the data into a preset number of groups, and a KNN-smoothing step builds heatmaps that make faint stellar sequences and overdensities visible. Membership is assigned only to stars that pass the 50% probability threshold in all algorithms, so a feature counts only if several independent methods agree. Ages and distances come from fitting PARSEC isochrones to the combined color-magnitude diagrams, distances are checked against zero-point-corrected Gaia parallaxes, and structural parameters come from King-profile fits to the radial density profiles.

What would settle it

Run the whole pipeline a second time on the same field without the 1 mas/yr proper-motion pre-filter and without the 50% membership threshold, and see whether the two clusters, the two isochrones, and the six subclusters still appear; or take spectra across the region to check whether two genuinely distinct age populations really share one set of kinematics.

Watch

Extended reading notes

Core claim

The central claim, stated as the author would state it, is that NGC1605a and NGC1605b are genuine open clusters that form an old binary pair in an advanced stage of merging, with ages of 600±100 Myr (NGC1605b) and 2±0.2 Gyr (NGC1605a) at a heliocentric distance of 2.75±0.40 kpc. The evidence is the agreement of independent algorithms: pyUPMASK, ASteCA, Kmeans, GMM, and HDBSCAN each assign membership in the Gaia 5D astrometric space, and the stars that all of them rate above 50% produce color-magnitude diagrams that require two isochrones, proper-motion diagrams with two kinematic components, and radial density profiles with tidal radii above 40 arcminutes. The paper also shows that the probable member lists of the two clusters are mostly the same stars, since the pair differs from the field population but not from each other, and that a combination of the elbow method, t-SNE, Kmeans, and GMM groups the normalized data into six clusters — two clusters plus four possible merger fragments — matching the earlier study (C21). Because the clusters differ in age but not in space or kinematics, the predicted end product is a single cluster with two age populations.

Load-bearing premise

The analysis assumes that cutting the data to stars within 1 mas/yr of each cluster's proper-motion center, and then keeping only stars that every algorithm rates above 50% membership, exposes the true member populations instead of creating the two kinematic groups the analysis then finds.

Editorial extensions

If this is right

  • If NGC1605a and NGC1605b are truly merging, the final product will be a single open cluster that is coherent in position and motion but contains two distinct age populations, so surveys should expect some old clusters to show two turnoffs in their color-magnitude diagrams.
  • Because most of NGC1605a's main-sequence stars are fainter than the usual G=18 magnitude cut, the pipeline shows that relaxing that cut can recover members of old, distant clusters that standard quality filters would remove.
  • The agreement of several independent clustering algorithms on the same member stars offers a template for validating membership and derived parameters in other complex stellar structures.
  • The survival of a binary pair to an advanced merger stage implies that at least some multiples of evolved open clusters can be long-lived, contrary to the usual expectation that they merge or disrupt early.

Reading between the lines

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

  • A testable extension the paper does not run: apply the same ensemble to other old open clusters with unusually broad main sequences; if the merger picture is right, some of them should resolve into two age populations.
  • The six groups found by the elbow method could be compared with N-body simulations of a close encounter between two clusters with the derived ages and distances, which would predict whether four satellite fragments are a plausible outcome.
  • Running the pipeline on synthetic single-cluster fields would measure how easily the 1 mas/yr proper-motion pre-filter alone can create a two-component structure, which would directly probe the main selection worry.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. This paper proposes a multiple machine-learning method (MMLM) that combines pyUPMASK, ASteCA, K-means, GMM, HDBSCAN, KNN smoothing, and t-SNE on Gaia EDR3 astrometry and photometry, and applies it to the NGC 1605 field. The stated goal is to test MMLM by reanalyzing the binary open cluster candidate NGC 1605a/b from Camargo (2021). The paper claims that the method confirms two genuine open clusters with ages of about 600 Myr and 2 Gyr at a heliocentric distance of about 2.75 kpc, in an advanced stage of merger, and that a combination of the elbow method, t-SNE, KMeans, and GMM groups the data into six clusters, following C21. The conclusion is that NGC 1605a and NGC 1605b will eventually merge into a single kinematically coherent cluster that contains two different age populations.

Significance. If the central claim is correct, MMLM would be a useful and relatively inexpensive pipeline for detecting cluster substructures and for validating parameters of previously identified clusters. The paper makes a genuine effort to combine several standard algorithms and to visualize CMD and PM structures with KNN smoothing. It is also transparent about some limitations, explicitly stating that the ML tools cannot separate the two components and that most stars in the two memberlists are the same sources. However, because of that admission and because the two isochrones are fitted to essentially the same blended CMD, the paper does not independently establish the two-age binary-merger claim; the main astrophysical conclusion rests on pre-selected kinematics and on a model that is degenerate with a single older cluster plus unresolved binaries. The methodological contribution would be strengthened by reproducibility details and by quantitative tests of uniqueness of the inferred substructure.

major comments (4)
  1. [Sect. 2 and Sect. 3 (Fig. 2, median PM values)] The working sample is constructed with a 1 mas/yr proper-motion radius cut centered on the PM centroids of the two clusters. The final median PMs are 0.91 vs 0.92 mas/yr in RA and -1.99 vs -1.98 mas/yr in Dec, so the centers are separated by only about 0.01-0.02 mas/yr. Two circles of radius 1 mas/yr with such centers overlap essentially completely, so the two 'memberlists' are nearly identical. The paper explicitly states that most stars in the memberlists of the two OCs are the same sources and that the ML algorithms are unable to efficiently isolate the stars of each cluster. Therefore the two isochrone ages are not derived from independent stellar populations. The central two-age claim is not supported unless the authors demonstrate quantitatively that the two final samples are distinct and that a single-population model with unresolved binaries is rejected.
  2. [Sect. 2 (PM cut) and Sect. 3 (membership threshold)] The proper-motion cut is justified as 'necessary to demonstrate that there are at least two OCs experiencing a merger event, as suggests C21.' This makes the selection conditional on the conclusion the paper seeks to establish. The paper should test the stability of the recovered substructure against the cut radius (e.g., 0.5 to 2 mas/yr), against field contamination, and against the 50% membership-probability threshold applied across all algorithms. Without such sensitivity tests, the recovered two-component and six-component structures may be artifacts of the chosen cuts.
  3. [Sect. 3 (Fig. 3 and isochrone fitting)] The ASteCA binary-probability map shows the bulk of binaries in the CMD region of the older component's upper main sequence. This is exactly where an unresolved binary sequence of a single about 2 Gyr cluster would appear, so Fig. 3 does not discriminate between two physically distinct populations and one older population with a prominent binary sequence. Moreover, the text says the adopted ages are 'the best solutions obtained in C21', which indicates that the ages are not independently derived in this work. The authors should report the isochrone fit statistics and perform a model comparison (e.g., two isochrones plus binaries versus one old isochrone plus binaries) or obtain independent spectroscopic membership to break this degeneracy.
  4. [Sect. 3 (Fig. 11)] The six-cluster decomposition using the elbow method, t-SNE, KMeans, Fuzzy, and GMM is presented as supporting the C21 substructure scenario, but no stability or validation metrics are provided (e.g., silhouette score, bootstrap replicates, variation of t-SNE perplexity, or comparison with a field-only control region). t-SNE is known to create spurious clusters in noisy data, and the elbow method is a heuristic; without such tests, the six-cluster claim is not established and cannot be used as independent support for the merger scenario.
minor comments (4)
  1. [Sect. 4] The Concluding Remarks state 'an age of 2±0.2 Gyr for NGC1605b and 600±100 Myr for NGC1605a', which is the reverse of the assignments in Sect. 3 ('600±100 Myr (NGC1605b)' and '2±0.2 Gyr (NGC1605a)'). Please correct this internal inconsistency.
  2. [Sect. 2] The text describes t-SNE as a 'supervised machine learning' algorithm; t-SNE is an unsupervised dimensionality-reduction technique (Hinton & Roweis 2003; van der Maaten & Hinton 2008).
  3. [Throughout] The manuscript contains numerous typos and grammatical errors, including 'applyed', 'neareast', 'implementatin', 'efficently', 'flutuations', 'garantee', and 'outliers' used as an adjective. A thorough language edit is needed.
  4. [Sect. 3] The phrase 'the best solutions obtained in C21' should be clarified: if the MMLM pipeline derives the ages, please report the fitting statistic and uncertainty budget; if the ages are adopted from Camargo (2021), state this explicitly in the methodology.

Circularity Check

2 steps flagged · score 6.0 of 10

The two-age 'confirmation' is a double isochrone fit to a single blended CMD, and the merger scenario is imported from C21 rather than independently tested.

  1. fitted input called prediction [Section 3, discussion of Figs. 2 and 3 and the derived ages]
    "Figs. 2 and 3 unveil that most stars in the memberlists of the two OCs are the same sources, which is expected since these stars differ from the Galactic field stars, which are the outliers removed by the algorithms. Further, as proposed by C21, NGC1605a and NGC1605b possibly make up an old binary cluster at an advanced stage of merger... these tools are unable to efficiently isolate the stars from each cluster, since they are too close to each other in the 5D phase space."

    The two 'cluster' CMDs are built from the same dominant set of stars, as the paper admits. Fitting 600 Myr and 2 Gyr PARSEC isochrones to those near-identical CMDs does not establish two co-existing stellar populations; it is the same blended sample fitted twice. The conclusion of an eventual single cluster 'coherent spatially and kinematically, but with two different ages' is therefore a restatement of the two isochrone fits, not a derivation from two independently selected populations. The fitted ages are presented as confirming C21's binary-cluster prediction, but the input memberlists already assume the two-cluster structure.

  2. self citation load bearing [Section 2 (PM quality cut) and Section 3 (elbow method / t-SNE / 6 clusters)]
    "Although this procedure certainly affects the clusters' completeness, it is necessary to demonstrate that there are at least two OCs experiencing an merger event, as suggests C21. ... Interestingly, the elbow method provides the same number of clusters found in C21 that suggests an ongoing merger event between two clusters during a close encounter, and four subclusters as possible products of the merger process."

    The proper-motion pre-filter is justified by the very conclusion it is supposed to test ('as suggests C21'), and the 6-cluster grouping is interpreted as C21's 2+4 substructure rather than tested against a null or alternative model. The central 'old binary cluster in an advanced stage of merging' scenario is imported from the author's prior C21 paper and then reported as confirmed by the MMLM pipeline. The confirmation therefore reduces in part to a self-citation chain: C21 supplies the candidate, the filter, and the interpretive frame.

full rationale

The paper is not entirely circular: the Gaia-based reanalysis, the ML membership probabilities, and the PARSEC isochrone fits do constitute independent numerical work, and the distance and structural parameters are derived from the data rather than read off from C21. However, the central claim that NGC1605a and NGC1605b are two genuine clusters with distinct ages reduces, by the paper's own admissions, to fitting two isochrones to essentially the same blended CMD. The paper explicitly states that most stars in the two memberlists are the same sources and that the ML tools are unable to isolate the two clusters in 5D phase space. In addition, the sample construction and the interpretation of the 6-cluster grouping follow C21, a same-author prior work, with the PM cut described as 'necessary to demonstrate' the merger 'as suggests C21.' Thus the 'confirmation' of the binary-cluster merger is partly a fitted-input result and partly a self-citation chain. An alternative single-old-cluster-plus-binary-sequence interpretation is not ruled out by a model comparison, but that is a correctness risk; the circularity score is set by the fitted-input and self-citation structure of the derivation.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The analysis depends on several modeling choices not derived from first principles: a tight PM pre-filter, an arbitrary membership threshold, KNN smoothing parameters, a t-SNE perplexity, and a subjective elbow choice for k=6. The isochrone ages and distance are fitted values, not external constants. These assumptions are common in cluster analysis, but they reduce the strength of the confirmation and make the six-cluster result partly dependent on user choices.

free parameters (7)
  • proper_motion_radius_cut = 1 mas/yr
    A tight PM cut around each cluster centroid used before clustering; it can force the sample to look comoving and is justified by the need to demonstrate the merger (Section 2).
  • membership_probability_threshold = 50% in all algorithms
    Stars are considered members only if all algorithms assign more than 50% probability; the threshold is arbitrary and no sensitivity analysis is given (Section 2).
  • KNN_smoothing_k = 3, 5, 8, 12, 28
    Chosen by inspection to highlight stellar sequences and overdensities; the visualization parameter affects the impression of substructure.
  • tSNE_perplexity = 50
    One value is used for the t-SNE projection; t-SNE maps depend strongly on perplexity and no stability check is reported.
  • number_of_clusters = 6
    The PCA elbow method is used to fix k=6 for KMeans and GMM; elbow selections are subjective and no silhouette or model comparison is given.
  • isochrone_age_pair = 600 Myr and 2 Gyr, with assignment inconsistent between Sections 3 and 4
    Ages are fitted to CMDs with PARSEC isochrones; no fit uncertainty or Bayesian evidence is reported, and the component assignment is contradictory.
  • isochrone_distance = 2.75 kpc
    Distance is fitted from isochrones and parallax; the quoted uncertainty is 0.40 kpc, but no covariance or systematic treatment is given.
assumptions (5)
  • domain assumption Gaia EDR3 astrometric and photometric errors after cuts are small enough not to bias clustering.
    Used throughout; PM error and RUWE cuts are applied, but residual systematics such as the parallax zero point are discussed only qualitatively.
  • standard math PARSEC isochrones and King profiles are valid models for these clusters.
    Used for age, distance, and structural parameters in Sections 2.1 and 3; the paper does not justify their applicability to a pair in the process of merging.
  • ad hoc to paper The strict PM radius cut of 1 mas/yr does not remove a large fraction of true members.
    Section 2 admits the cut affects completeness and may discard ejected members, yet uses it to build the final dataset.
  • ad hoc to paper The two fitted isochrones correspond to two physically distinct stellar populations rather than a single broad population.
    The CMDs show overlapping sequences; the decomposition into two discrete ages is imposed by fitting two isochrones in Fig. 2.
  • ad hoc to paper The six clusters found by elbow, t-SNE, KMeans, and GMM correspond to physical substructures rather than algorithmic artifacts.
    Section 3 advises care with t-SNE interpretation, but still interprets the six groups as supporting C21's merger scenario.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multiple machine-learning as a powerful tool for the star clusters analysis." pith.science (2026). https://pith.science/paper/HKL6V7Y4

@misc{pith2026250613951,
  author       = {Pith},
  title        = {Pith review of: Multiple machine-learning as a powerful tool for the star clusters analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HKL6V7Y4}},
  note         = {Machine review of arXiv:2506.13951}
}
read the original abstract

This work proposes a multiple machine learning method (MMLM) aiming to improve the accuracy and robustness in the analysis of star clusters. The MMLM performance is evaluated by applying it to the reanalysis of the old binary cluster candidate - NGC 1605a and NGC 1605b - found by Camargo (2021) (hereafter C21). The binary cluster candidate is analyzed by employing a set of well established machine learning algorithms applied to the Gaia-EDR3 data. Membership probabilities and open clusters (OCs) parameters are determined by using the clustering algorithms pyUPMASK, ASteCA, Kmeans, GMM, and HDBSCAN. In addition, a KNN smoothing algorithm is implemented to enhances the visualization of features like overdensities in the 5D space and intrinsic stellar sequences on the color-magnitude diagrams (CMDs). The method validates the clusters' parameters previously derived, however, suggests that their probable members-stars are distributed over a wider overlapping area. Finally, a combination of the elbow method, t-SNE, kmeans, and GMM algorithms group the normalized data into 6 clusters, following C21. In short, these results confirm NGC1605a and NGC1605b as genuine OCs and reinforce the previous suggestion that they form an old binary cluster in an advanced stage of merging after a tidal capture during a close encounter. Thus, MMLM has proven to be a powerful tool that helps to obtain more accurate and reliable clusters parameters and its application in future studies may contribute to a better characterization of the Galaxy's star cluster system.

Figures

Figures reproduced from arXiv: 2506.13951 by the authors.

Figure 1
Figure 1. Spatial and PM distributions of stars for an extraction area of R = 20′ centered in the coordinates of NGC1605a and color coded by the pyUPMASK membership probabilities. Both dia￾grams reveal multiple substructures, especially the PM distibution. vide valuable memberships, accurate parame￾ters, and to search for unknow clusters, espe￾cially open clusters (Li et al. 2025, and refer￾ences therein). Open clusters are k… view at source ↗
Figure 2
Figure 2. Gaia-EDR3 optical CMDs for the probable member-stars (black circles) within an extraction area of R = 60′ centered in the coordinates of NGC1605a (left panels), NGC1605b (middle panels), and a combination of both (right panels). The heatmaps are built by applying the KNN-smoothing algorithm smoothing on K = 3 (top panels) and K = 8 neareast neighbors (bottom panels) and are color coded by the smoothness level. The t… view at source ↗
Figure 3
Figure 3. Gaia-EDR3 CMDs for the NGC1605a (left panel) and NGC1605b memberlists (right panel). The color bar denotes the binary probability for the cluster members candidates. The bulk of probable binary stars in both CMDs are located in the region corresponding to the NGC1605a upper MS. avoid large uncertainties, a photometric cut is used excluding stars fainter than G = 18 mag. These cuts ensure that the most of the detecte… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Heatmaps of the spatial distribution for stars in the NGC1605b memberlist, built by using the KNN-smoothing algorith smoothing on K = 28 (top panels) and K = 5 (bottom panels) neareast neighbors. The right panels display the schematic distribution of the cluster probab…
Figure 5
Figure 5. Figure 5: Parallax versus G magnitude for the NGC1605b memberlist. The heatmaps are built by the KNN-smoothing algorithm. densed tree are clusters. Such an hierarchical procedure is a powerful tool to isolate outliers. HDBSCAN can identify clusters of varying den￾sity. Kmeans sp…
Figure 6
Figure 6. Figure 6: Parallax and PMs errors versus G mag￾nitude for stars in the NGC1605b memberlist. tended structures, extraction radii of R = 60′ are adopted. Subsequently, the extracted Gaia dataset is submitted to the set of clustering algorithms (pyUPMASK, ASteCA, GMM, and HDBSCAN) …
Figure 7
Figure 7. Figure 7: Gaia-EDR3 PM distribution, for NGC1605b, highlighted by the KNN-smoothing al￾gorithm for K = 12 (top panels) and K = 5 neigh￾bors (bottom panels). The black circles are the stars in the cluster memberlist. the region on the CMD of a star cluster where it is expected a …
Figure 8
Figure 8. Figure 8: Top panel: MST for the NGC1605b memberlist implemented by a combination of KNN and HDBSCAN. Bottom panel: a zoom in the PM diagram built by using the KNN-smoothing algo￾rithm. The [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Gaia CMDs for the NGC1605 memberlist color coded by the HDBSCAN and GMM membership probabilities. along with the concepts of core distance and mutual reachability distance. For comparison, the KNN-smoothing algorithm also provides a zoom in the PM diagram (bottom panel…
Figure 10
Figure 10. Figure 10: RDP for NGC1605b and NGC1605a. Blue region: 1σ King fit uncertainty. terestingly, the elbow method provides the same number of clusters found in C21 that suggests an ongoing merger event between two clusters during a close encounter, and four subclusters as possible p…
Figure 11
Figure 11. Figure 11: In the Top-Left panel is shown the result of applying the PCA elbow method, which found 6 clusters in the NGC 1605 area. The wcss (within-clusters sum-of-squares) measures the sum of squared distances between each star and its assigned cluster center as a function of …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

77 extracted references · 65 canonical work pages

  1. [1]

    Bhattacharya, S., 2021, MNRAS, 502, 2582

    work page 2021

  2. [2]

    2021, A&A, 645, L2

    Anders, F., Cantat-Gaudin, T., Quadrino-Lodoso, I., et al. 2021, A&A, 645, L2

    work page 2021

  3. [3]

    Corradi, W.J.B., MNRAS, 2021, 510, 5695

    work page 2021

  4. [4]

    2023, A&A , 674, A32

    Babusiaux, C., Fabricius, C., Khanna, S., et al. 2023, A&A , 674, A32

    work page 2023

  5. [5]

    A., Tadross, A

    Badawy, W. A., Tadross, A. L., Hendy, Y. H. M., et al. 2024, New Astronomy, 109, 102196. Belokurov V., et al., 2020, MNRAS, 496, 1922 Bica E., Bonatto C., Camargo D., 2008, MNRAS, 385, 349

    work page 2024

  6. [6]

    Bisht, D., Zhu, Q., Yadav, R. K. S., et al. 2021, MNRAS, 503, 5929

    work page 2021

  7. [7]

    V., & Bajkova, A

    Bobylev, V. V., & Bajkova, A. T. 2019, Astron. Lett., 45, 208

    work page 2019

  8. [8]

    2011, A&A, 532, A131

    Borissova, J., Bonatto, C., Kurtev, R., et al. 2011, A&A, 532, A131

    work page 2011

Show all 77 references
  1. [9]

    2012, MNRAS, 427, 127 Buckner A

    Bressan, A., Marigo, P., Girardi, L., Salasnich, B., et al. 2012, MNRAS, 427, 127 Buckner A. S. M., Naylor T., Dobbs C. L., 2024, MNRAS 527, 5448

    2012

  2. [10]

    & Bica, E

    Camargo, D., Bonatto, C. & Bica, E. 2009, A&A, 508, 211

    2009

  3. [11]

    & Bica, E

    Camargo, D., Bonatto, C. & Bica, E. 2010, A&A, 521, 42

    2010

  4. [12]

    & Bica, E

    Camargo, D., Bonatto, C. & Bica, E. 2011, MNRAS, 416, 1522 Camargo D., Bonatto C., Bica E.., MNRAS, 2012, 423, 1940

    2011

  5. [13]

    & Bonatto, C

    Camargo, D., Bica, E. & Bonatto, C. 2013, MNRAS, 432, 3349

    2013

  6. [14]

    & Bonatto, C

    Camargo, D., Bica, E. & Bonatto, C. 2016, MNRAS, 455, 3126

    2016

  7. [15]

    2018, ApJL, 860, 27

    Camargo, D. 2018, ApJL, 860, 27

    2018

  8. [16]

    2019, MNRAS, 484L, 90C

    Camargo, D., & Minniti, D. 2019, MNRAS, 484L, 90C

    2019

  9. [17]

    Camargo, D., 2021, ApJ, 923, 21

    2021

  10. [18]

    Campello, R. J. G. B., Moulavi, D., & Sander, J. 2013, Advances in Knowledge Discovery and Data Mining, 7819, 160

    2013

  11. [19]

    2018, A&A, 618, A93

    Cantat-Gaudin, T., Jordi, C., Vallenari, A., et al. 2018, A&A, 618, A93

    2018

  12. [20]

    2021, Astronomy Reports, 65, 755

    Casado, J. 2021, Astronomy Reports, 65, 755

    2021

  13. [21]

    2020, A&A, 635, A45

    Castro-Ginard, A., Jordi, C., Luri, X., et al. 2020, A&A, 635, A45

    2020

  14. [22]

    2022, A&A, 661, A118

    Castro-Ginard, A., Jordi, C., Luri, X., et al. 2022, A&A, 661, A118

    2022

  15. [23]

    2024, AJ, 167, 44 Chi H., Wei S., Wang F., Li Z., 2023, ApJS, 265, 20

    Chen, J., Li, Z., Zhang, S., Zhao, W., & Wu, Y. 2024, AJ, 167, 44 Chi H., Wei S., Wang F., Li Z., 2023, ApJS, 265, 20

    2024

  16. [24]

    2022, MNRAS, 517, 2925 Conrad C., Scholz R

    Chulkov, D., & Malkov, O. 2022, MNRAS, 517, 2925 Conrad C., Scholz R. D., Kharchenko N. V., Piskunov A. E., R¨ oser S., 2017, A&A, 600, A106 14Denilso Camargo Cover T., Hart P., 1967, IEEE Trans. Inf. Theory, 13, 21 de la Fuente Marcos, R., & de la Fuente Marcos, C., 2009, A&A...

    2022

  17. [25]

    Buckner, A. S. M., Bate, M. R., 2022, MNRAS, 517, 675

    2022

  18. [26]

    P., Laird, N

    Dempster, A. P., Laird, N. M., & Rubin, D. B. 1977, Proceedings of the Royal Statistical Society, 39, 12

    1977

  19. [27]

    2021, A&A, 649, A5

    Fabricius, C., Luri, X., Arenou, F., et al. 2021, A&A, 649, A5

    2021

  20. [28]

    Froebrich, D., Scholz, A., & Raftery, C. L. 2007, MNRAS, 374, 399

    2007

  21. [29]

    2022, A&A, 668, A4 Gaia Collaboration et al., 2021, A&A, 649, A1 Gaia Collaborationet al., 2023, A&A 674, A1

    Fu, X., Bragaglia, A., Liu, C., et al. 2022, A&A, 668, A4 Gaia Collaboration et al., 2021, A&A, 649, A1 Gaia Collaborationet al., 2023, A&A 674, A1. Garro E. R., et al., 2020, A&A, 642, L19

    2022

  22. [30]

    R., Minniti, D., G´ omez, M., et al

    Garro, E. R., Minniti, D., G´ omez, M., et al. 2022, A&A, 662, A95

    2022

  23. [31]

    Gokmen, S., Eker, Z., Yontan, T., et al., 2023, AJ 166, 263

    2023

  24. [32]

    J., Xu, Y., Hou, L

    Hao, C. J., Xu, Y., Hou, L. G., et al. 2021, A&A, 652, A102

    2021

  25. [33]

    Hinton, G. E. & Roweis, S. T. 2003, in Advances in neural information processing systems, 857, 864

    2003

  26. [34]

    Hou, L. G. 2021, Frontiers in Astronomy and Space Sciences, 8, 103

    2021

  27. [35]

    2025, AJ, 169, 98H

    Hu, Qingshun, Li, Yuting, Qin, Mingfeng, et al. 2025, AJ, 169, 98H

    2025

  28. [36]

    C., & Zhang, H

    Huang, Y., Yuan, H., Beers, T. C., & Zhang, H. 2021, ApJ, 910, L5

    2021

  29. [37]

    L., & Reffert, S., 2023, A&A 673, A114

    Hunt, E. L., & Reffert, S., 2023, A&A 673, A114

    2023

  30. [38]

    V., Piskunov A

    Peter, Kharchenko, N. V., Piskunov A. E., et al., 2024, A&A, 686, A225

    2024

  31. [39]

    Jerabkova, T., Boffin, H. M. J., Beccari, G., et al. 2021, A&A, 647, A137

    2021

  32. [40]

    Karam, J., Sills, A., 2022, MNRAS, 513, 6095

    2022

  33. [41]

    I., Miglio, A., Mosser, B., & Elsworth, Y

    Khan, S., Anderson, R. I., Miglio, A., Mosser, B., & Elsworth, Y. P. 2023, A&A, 680, A105

    2023

  34. [42]

    A., Ishchenko, M., Postnikova, E., 2020, A&A, 642, L4 Krone-Martins A., Moitinho A., 2014, A&A, 561, A57

    Kovaleva, D. A., Ishchenko, M., Postnikova, E., 2020, A&A, 642, L4 Krone-Martins A., Moitinho A., 2014, A&A, 561, A57

    2020

  35. [43]

    D., Getman, K

    Feigelson, E. D., Getman, K. V., 2019, ApJ, 870, 32

    2019

  36. [44]

    2022, ApJS, 259, 19

    Li, Z., Deng, Y., Chi, H., et al. 2022, ApJS, 259, 19

    2022

  37. [45]

    Li, S., Casertano, S., & Riess, A. G. 2022, ApJ, 939, 96

    2022

  38. [46]

    Li, Z., Mao, C., 2023, ApJS, 265, 3

    2023

  39. [47]

    Li, Z., Zhu, Z., 2024, arXiv240502530

    2024

  40. [48]

    2024, arXiv:2401.10717

    Li, Z., & Mao, C. 2024, arXiv:2401.10717

    2024 arXiv

  41. [49]

    2025, arXiv e-prints, arXiv:2502.15300

    Li, G., Lu, Z., Wang, J., & Wang, Z. 2025, arXiv e-prints, arXiv:2502.15300

    2025 arXiv

  42. [50]

    A., Hern´ andez, J., et al

    Lindegren, L., Klioner, S. A., Hern´ andez, J., et al. 2021, A&A, 649, A2

    2021

  43. [51]

    2019, ApJS, 245, 32

    Liu, L., & Pang, X. 2019, ApJS, 245, 32

    2019

  44. [52]

    Liu, G., Zhang, Y., Zhong, J. et al. 2025, A&A, 696, 117 Ma ´ ız Apell´ aniz, J., Pantaleoni Gonz´ alez, M., & Barb´ a, R. H. 2021, A&A, 649, A13

    2025

  45. [53]

    2013 Ap&SS, 344, 175

    Majaess, D. 2013 Ap&SS, 344, 175

    2013

  46. [54]

    G., et al

    Majaess, D., Bonatto, C., Turner, D. G., et al. 2025 arXiv250206930M

    2025

  47. [55]

    McInnes, L., Healy, J., & Astels, S., 2017, The Journal of Open Source Software, 2

    2017

  48. [56]

    W., Emerson, J

    Minniti, D., Lucas, P. W., Emerson, J. P., et al. 2010, New A, 15, 433

    2010

  49. [57]

    Minniti, D., Geisler, D., Alonso-Garcia, J. et al. 2017a, ApJ, 849, L24 Minniti D., Fern´ andez-Trincado J. G., G´ omez M., Smith L. C., Lucas P. W., Contreras Ramos R., 2021, A&A, 650, L11

    2021

  50. [58]

    2021, A&A, 652, A129 Molina Lera, J

    Minniti, D., Palma, T., Camargo, D., et al. 2021, A&A, 652, A129 Molina Lera, J. A., Baume, G., & Gamen, R. 2019, MNRAS, 488, 2158

    2021

  51. [59]

    2023, MNRAS, 520, 4154

    Molinaro, R., Ripepi, V., Marconi, M., et al. 2023, MNRAS, 520, 4154

    2023

  52. [60]

    Negueruela, I., & de Burgos, A., 2023, A&A 675, A19

    2023

  53. [61]

    2023, MNRAS, 523, 3538

    Haghi, H. 2023, MNRAS, 523, 3538

    2023

  54. [62]

    W., 2022a, MNRAS, 513, 2437

    Palma, T., Coenda, V., Baume, G., & Feinstein, C., 2025, A&A, 693, 218 Penoyre Z., Belokurov V., Evans N. W., 2022a, MNRAS, 513, 2437

    2025

  55. [63]

    S., Perren, G

    Pera, M. S., Perren, G. I., Moitinho, A., Navone, H. D., & Vazquez, R. A. 2021, A&A, 650, A109

    2021

  56. [64]

    I., V´ azquez, R

    Perren, G. I., V´ azquez, R. A., & Piatti, A. E. 2015, A&A, 576, A6

    2015

  57. [65]

    I., Giorgi, E

    Perren, G. I., Giorgi, E. E., Moitinho, A., et al. 2020, A&A, 637, A95

    2020

  58. [66]

    I., Pera, M

    Perren, G. I., Pera, M. S., Navone, H. D., & V´ azquez, R. A. 2022, A&A, 663, A131

    2022

  59. [67]

    Piatti, A. E. & Malhan, K. 2022, MNRAS, 511, L1

    2022

  60. [68]

    2023, ApJS, 265, 12

    Qin, S., Zhong, J., Tang, T., & Chen, L. 2023, ApJS, 265, 12

    2023

  61. [69]

    2025, A&A, 693, A317

    Qin, S., Zhong, J., Tang, T., et al. 2025, A&A, 693, A317

    2025

  62. [70]

    Wurster, J., 2022, MNRAS, 509, 6155

    2022

  63. [71]

    W., et al

    Riello, M., De Angeli, F., Evans, D. W., et al. 2021, A&A, 649, A3

    2021

  64. [72]

    G., Breuval, L., Yuan, W., et al

    Riess, A. G., Breuval, L., Yuan, W., et al. 2022, ApJ, 938, 36

    2022

  65. [73]

    K., Hempel, M., Alonso-Garc ´ ıa, J

    Saito, R. K., Hempel, M., Alonso-Garc ´ ıa, J. et al. 2024, A&A, 689, A148

    2024

  66. [74]

    F., Cutri, R

    Skrutskie, M. F., Cutri, R. M., Stiening, R., et al. 2006, AJ, 131, 1163

    2006

  67. [75]

    2022, A&A, 666, A75 Ta¸ sdemir, S., & C ¸ ınar, D

    Song, F., Esamdin, A., Hu, Q., et al. 2022, A&A, 666, A75 Ta¸ sdemir, S., & C ¸ ınar, D. C. 2025, CoSka, 55c, 506 van der Maaten, L. & Hinton, G. 2008, The Journal of Machine Learning Research, 9, 85

    2022

  68. [76]

    L., Eisenhardt, P

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

    2010

  69. [77]

    Wu, Y., Chen, J., Zhang, J., et al., 2024, ApJ, 965, 131

    2024

Pith tools

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