Pith. sign in

REVIEW 4 major objections 5 minor 70 references

Unified Deep Learning Approach for Estimating the Metallicities of RR Lyrae Stars Using light curves from Gaia Data Release 3

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

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

Pith's one-line read A unified GRU neural network recovers photometric metallicities for both RRab and RRc RR Lyrae stars from Gaia G-band light curves and extends coverage to 258,696 stars.

desk verdict A credible engineering extension of the authors' RRab work to RRc stars, but the validation loop is closed against the same photometric calibrations used as training labels, so the headline metrics measure reproduction of Muraveva et al. (2025), not true [Fe/H]. read the letter →

arxiv 2505.20947 v1 pith:A6RHUGXF submitted 2025-05-27 astro-ph.SR cs.AI

Lorenzo Monti , Tatiana Muraveva , Alessia Garofalo , Gisella Clementini , Maria Letizia Valentini This is my paper
classification astro-ph.SRcs.AI
keywords RRLyraestarsmetallicityestimationGaiaDR3deeplearningGRUphotometricmetallicitiestime-seriesextrinsicregressionlightcurves
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 tries to establish that a single deep-learning model, a gated recurrent unit (GRU) network, can read the shape of a Gaia G-band light curve and return a photometric metallicity estimate for RR Lyrae stars of both pulsation types, fundamental-mode (RRab) and first-overtone (RRc), without needing separate models or intermediate Fourier-parameter steps. On held-out validation, it reports errors around 0.05 to 0.08 dex, with mean absolute errors of 0.0565 and 0.0505 dex, root-mean-square errors of 0.0765 and 0.0720 dex, and $R^2$ values of 0.94 and 0.96 for RRab and RRc stars. Applied to the cleaned Gaia DR3 catalogue, the same model assigns metallicities to 258,696 RR Lyrae stars, lifting coverage by factors of 1.25 for RRab and 4.48 for RRc over the period–Fourier relations that supplied the training labels. The payoff is a scalable, homogeneous way to chemically map the old, metal-poor stellar populations of the Milky Way from photometry alone.

What carries the argument

The load-bearing object is the GRU regressor, a recurrent neural network whose update and reset gates let it carry a hidden state over the 264–265 resampled phase points of the standardized light curve and map the whole sequence to a scalar $δ$ via a final dense linear layer. It is supported by a preprocessing pipeline that phase-folds each light curve to maximum light, smooths it with a spline, resamples to a fixed number of points, standardizes each curve to zero mean and unit variance, and up-weights rare metallicity values through density-dependent sample weights. The reported metrics come from a repeated stratified K-fold cross-validation protocol with dropout, L1/L2 regularization, and early stopping.

What would settle it

Compare the trained model's predictions against high-resolution spectroscopic $δ$ measurements for RR Lyrae stars excluded from training; a systematic offset or scatter well above the reported 0.05–0.08 dex would show the model does not measure true metallicity.

Watch

Extended reading notes

Core claim

The central claim is that a unified GRU-based regressor, trained on phase-folded, spline-smoothed, amplitude-standardized G-band light curves with a density-weighted loss, reproduces the literature photometric metallicities for both RRab and RRc stars with validation mean absolute errors of 0.0565 and 0.0505 dex and $R^2$ of 0.9401 and 0.9625, respectively. The same model, applied to the full cleaned Gaia DR3 catalogue, produces photometric metallicities for 258,696 RR Lyrae stars (169,024 RRab and 89,672 RRc), increasing the number of stars with available metallicities by factors of 1.25 and 4.48 for the two pulsation types. The authors further report that prediction precision improves with the number of observed epochs, and that the recovered sky distribution agrees with expectations from Galactic structure.

Load-bearing premise

The load-bearing premise is that the photometric metallicities used as training targets and as validation ground truth are accurate enough proxies for true stellar metallicity that a model that reproduces them is scientifically useful, and if those labels contain systematic offsets, the model inherits them and every reported metric only measures agreement with the same calibration.

Editorial extensions

If this is right

  • A single model serves both RRab and RRc stars, so no per-pulsation-mode training is needed for new samples.
  • Running the model on the cleaned Gaia DR3 catalogue yields photometric metallicities for 258,696 RR Lyrae stars, expanding coverage by factors of 1.25 for RRab and 4.48 for RRc.
  • Prediction accuracy improves with the number of observed epochs, so the longer baseline of Gaia DR4 should reduce scatter in derived metallicities.
  • Bypassing intermediate period–Fourier calibrations gives a homogeneous metallicity scale across the catalogue.
  • The derived sky distribution, with metal-rich stars in the disk and metal-poor stars in the halo, is consistent with known Galactic structure and supports the method's plausibility.

Reading between the lines

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

  • The reported metrics quantify agreement with the period–Fourier photometric labels, not with true stellar metallicity; establishing real accuracy requires comparison against spectroscopic $δ$ values the model never saw.
  • Because validation stars were required to have at least 50 epochs, catalogue-wide performance on sparsely sampled stars is likely worse than the headline metrics suggest.
  • A natural head-to-head test would be to compare the GRU predictions with the original Jurcsik–Kovacs Fourier metallicities on a common sample, quantifying what the deep-learning approach changes beyond the relations it was trained to reproduce.
  • The same pipeline should transfer to other photometric bands or to Gaia DR4 with only light retraining, since the input representation is band-agnostic.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This paper presents a GRU-based time-series extrinsic regression model that takes phase-folded, spline-smoothed, standardized Gaia DR3 G-band light curves as input and predicts photometric metallicities for both RRab and RRc RR Lyrae stars. Training and validation labels are the Muraveva et al. (2025) photometric metallicities, which were derived from period–Fourier parameter relations calibrated on literature spectroscopic abundances. On cross-validated development samples the model reports R² ≈ 0.94 for RRab and ≈ 0.96 for RRc, with validation MAE ≈ 0.06 dex and ≈ 0.05 dex, respectively. The model is then applied to the cleaned Gaia DR3 sample, producing photometric metallicities for 258,696 RRLs. The paper also reports learning curves and comparisons with prior deep-learning work, and the code is publicly available.

Significance. If the reported accuracy could be demonstrated against independent spectroscopic measurements, the method would be a practical and scalable tool: it extends earlier RRab-only deep-learning work to RRc stars, releases a public implementation, and increases catalogued photometric-metallicity coverage by factors of 1.25 (RRab) and 4.48 (RRc). The learning-curve behavior and the small training-validation gap are encouraging signs of internal consistency and of no severe overfitting. However, the main evidence is circular with respect to physical metallicity, because every accuracy metric is computed against the same Muraveva et al. (2025) calibration used as the training target; the abstract's claim of high predictive accuracy therefore currently applies only to reproduction of that calibration, and an external spectroscopic validation is needed to support the astrophysical claims.

major comments (4)
  1. [§3.4, Table 1; §4, Fig. 7] All training and validation targets are the same Muraveva et al. (2025) photometric metallicities, as stated in §2.1 and conceded in §4. The headline metrics therefore measure how well the GRU reproduces that particular period–Fourier calibration, not how accurately it measures physical [Fe/H]. The paper should add a comparison against an external sample with spectroscopic metallicities (e.g., high-resolution or ΔS-based measurements) and report the resulting bias and scatter; without this, the abstract's claim of high predictive accuracy is unsupported.
  2. [§4, Fig. 7] The text notes a systematic offset for RRab stars, with predicted values systematically lower than the Muraveva et al. targets, but it neither quantifies the offset nor discusses its effect on the 258,696-value catalogue. Because the model is trained on the same labels, any systematic error in the Muraveva et al. calibration is inherited, and a quantified offset against independent data is needed before the catalogue can be used for Galactic-archaeology applications.
  3. [§1, §3.4] The paper's central 'unified' claim is ambiguous: the abstract says the model handles RRab and RRc 'without requiring separate models,' but §3.4 and Figs. 5 and 6 report separate performance for the two classes and never state explicitly whether a single network was trained on the combined sample or two separate networks were trained. If the latter, the 'unified' claim should be revised to 'a common pipeline and architecture,' and the text should state the actual training configuration.
  4. [§4] The large-scale validation in Fig. 7 (108,766 RRab and 13,388 RRc stars) uses stars deliberately excluded from training by the stricter selection criteria; the visible scatter, especially at low Nepochs, and the RRab offset indicate that Table 1's validation metrics do not transfer to the full catalogue. The paper should report quantitative error metrics for this full-sample comparison, including bias, scatter, and their dependence on Nepochs and [Fe/H], before deploying the model to produce the final catalogue.
minor comments (5)
  1. [§2.3, Eq. (2)] Equation (2) is typeset incorrectly: the summation and integral are malformed, and the smoothing parameter λ is not defined. Please fix the display and define the parameter choice.
  2. [§2.4, Eq. (4)] Equation (4) contains a typo ('Ph·P t') and does not clearly define the standardization; it should read (m(t) − mean)/std, and the meaning of 'μ_m' and 'σ_m' should be stated explicitly.
  3. [§3.3, Eq. (13)] The statement that 'R² ranges from 0 to 1' is incorrect for a general regression model; R² can be negative. The text should say that values near 1 indicate a strong fit.
  4. [§2.1 and §4] The number of RRab stars with Muraveva et al. (2025) metallicities is given as 114,468 in §2.1 and as 114,768 in §4; these numbers should be reconciled.
  5. [§2.4, §3.3] The description of 'Repeated Stratified K-Fold Cross-Validation' does not explain how stratification is applied to a continuous target variable, and the Gaussian KDE bandwidth used to compute the sample weights is not reported. Both choices affect reproducibility and should be specified.

Circularity Check

1 steps flagged · score 4.0 of 10

Validation of 'accurate metallicity prediction' is closed-loop: Table 1 and Fig. 7 compare against the same Muraveva et al. (2025) photometric metallicities used as training targets, so the reported metrics measure reproduction of that photometric calibration, not true [Fe/H].

  1. fitted input called prediction [Sections 2.1, 3.4, 4 (Table 1, Eq. 11, Eq. 13, Fig. 7)]
    "In this study, we use the time-series photometry of RRLs in the G-band ... for a cleaned sample of RRLs with available photometric metallicities from Muraveva et al. (2025). ... The goal is to assess how accurately the model predicts the photometric [Fe/H] values derived by Muraveva et al. (2025) ... It is important to recall that these target values, while serving as the ground truth for training and evaluating our specific regression model, are themselves photometric estimates calibrated using spectroscopically measured metallicities from the literature (Crestani et al. 2021; Liu et al."

    The training labels (Section 2.1) and the validation ground truth (Sections 3.4 and 4) are the same quantity: photometric [Fe/H] from Muraveva et al. (2025), a calibration produced by the same group from the same Gaia DR3 light curves. The GRU is trained to minimize MSE (Eq. 11) against these labels, and the headline metrics (R^2 in Eq. 13, Table 1, Fig. 7) compare predictions against those same labels. Thus the reported 'high predictive accuracy' only establishes that the model reproduces the Muraveva et al. photometric calibration; it does not validate [Fe/H] against independent spectroscopy. Any systematic error in the calibration is inherited and invisible, so the central claim of accurate metallicity estimation reduces to agreement with the fitted input calibration.

full rationale

The paper is self-contained as a supervised regression: the GRU learns a nontrivial mapping from G-band light curves to the Muraveva et al. (2025) photometric metallicities, and held-out cross-validation shows good generalization of that mapping. This is a legitimate technical result. However, the scientific claim 'estimating metallicities' is only validated against the same photometric estimates used as training targets; no external high-resolution spectroscopic comparison is performed in this paper. The paper itself concedes the targets 'are themselves photometric estimates calibrated using spectroscopically measured metallicities from the literature.' Since Muraveva et al. (2025) shares four co-authors with the present work, the validation chain is closed with respect to the scientifically relevant quantity [Fe/H]. The external spectroscopic anchoring (Crestani et al. 2021; Liu et al. 2020) enters only through the prior calibration, not through direct validation here, so the coverage-extension claim (258,696 RRLs) inherits any systematics of that calibration. This is a partial circularity of the validation, not a by-construction equivalence; the model could have failed to reproduce the labels, and the unified RRab/RRc architecture has independent technical content.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central mapping is an empirical regression, not a derivation. Its behavior is fixed by the Muraveva et al. (2025) training labels, hand-chosen preprocessing constants, and hyperparameters selected on validation data. No new physical entities, particles, forces, or conserved quantities are introduced, so the invented-entity ledger is empty.

free parameters (6)
  • neural network weights and biases = not released
    The entire mapping from light curves to metallicity is determined by trained network parameters; no trained weights are provided.
  • smoothing spline parameter lambda (or s) = not reported
    The smoothing strength in Equation (2) controls how closely splines follow noisy Gaia photometry and affects all input sequences.
  • resampled phase-grid length = 264 points (RRab), 265 points (RRc)
    Chosen by hand to standardize input sequence length; no sensitivity analysis is given.
  • Gaussian KDE bandwidth for sample weights = not reported
    The bandwidth sets the density estimate rho-hat(x) in Equation (3) and therefore all training sample weights.
  • selection thresholds for the development sample = sigma[Fe/H] <= 0.4 dex, AmpG <= 1.4 mag, Nepochs >= 50, sigma_phi31 <= 0.10
    The quality cuts in Section 2.1 define the training distribution and are applied by hand before model fitting.
  • GRU hyperparameters = learning rate 0.01, batch size 256 reported; dropout rate, L1/L2 strengths, and hidden units not fully reported
    The final configuration after Hyperband search is not given in full, leaving the exact model underspecified.
assumptions (4)
  • standard math The GRU update equations and the spline objective are correct as stated.
    Equations (2), (6)-(9) are standard formulations taken from the cited deep-learning and interpolation literature.
  • domain assumption Photometric metallicities from Muraveva et al. (2025) are reliable training labels for true stellar metallicity.
    All training and validation use these labels, and Section 4 explicitly compares against them. The calibration has external spectroscopic anchoring, but systematic offsets in that calibration would propagate into the model.
  • domain assumption Standardizing each light curve by subtracting mean magnitude and dividing by standard deviation does not remove metallicity-sensitive morphology.
    Section 2.3 removes absolute amplitude information, which is motivated by distance and extinction independence but discards a quantity that correlates with metallicity in Figure 1.
  • ad hoc to paper A single GRU without an explicit pulsation-mode indicator can separate RRab and RRc morphologies well enough for one unified regressor.
    The paper claims a unified model without separate models, but it does not state whether one shared network was trained on the combined sample or two mode-specific networks were trained.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unified Deep Learning Approach for Estimating the Metallicities of RR Lyrae Stars Using light curves from Gaia Data Release 3." pith.science (2026). https://pith.science/paper/A6RHUGXF

@misc{pith2026250520947,
  author       = {Pith},
  title        = {Pith review of: Unified Deep Learning Approach for Estimating the Metallicities of RR Lyrae Stars Using light curves from Gaia Data Release 3},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A6RHUGXF}},
  note         = {Machine review of arXiv:2505.20947}
}
read the original abstract

RR Lyrae stars (RRLs) are old pulsating variables widely used as metallicity tracers due to the correlation between their metal abundances and light curve morphology. With ESA Gaia DR3 providing light curves for about 270,000 RRLs, there is a pressing need for scalable methods to estimate their metallicities from photometric data. We introduce a unified deep learning framework that estimates metallicities for both fundamental-mode (RRab) and first-overtone (RRc) RRLs using Gaia G-band light curves. This approach extends our previous work on RRab stars to include RRc stars, aiming for high predictive accuracy and broad generalization across both pulsation types. The model is based on a Gated Recurrent Unit (GRU) neural network optimized for time-series extrinsic regression. Our pipeline includes preprocessing steps such as phase folding, smoothing, and sample weighting, and uses photometric metallicities from the literature as training targets. The architecture is designed to handle morphological differences between RRab and RRc light curves without requiring separate models. On held-out validation sets, our GRU model achieves strong performance: for RRab stars, MAE = 0.0565 dex, RMSE = 0.0765 dex, R^2 = 0.9401; for RRc stars, MAE = 0.0505 dex, RMSE = 0.0720 dex, R^2 = 0.9625. These results show the effectiveness of deep learning for large-scale photometric metallicity estimation and support its application to studies of stellar populations and Galactic structure.

Figures

Figures reproduced from arXiv: 2505.20947 by the authors.

Figure 1
Figure 1. Distributions of RRab and RRc stars from our selected development datasets on the amplitude in the [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Normalized splined G-band light curves of 6002 RRab stars (a) and 6613 RRc stars (b) from our development datasets. The sample weights (wd) were computed using a Gaussian kernel density estimation (Gaussian KDE) approach: wd = 1 ρˆ(x) , (3) where ˆρ(x) is the normalized density estimate of the metallicity distribution at a given point x. Gaussian kernels were used to es￾timate the density ˆρ, ensuring that areas wit… view at source ↗
Figure 3
Figure 3. Photometric metallicity distributions and corresponding sample weights (black lines) for RRab ( [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Schematic overview of the GRU-based neural network architecture used for predicting stellar metallicity ([Fe/ [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Training (red) and validation (green) loss curves across epochs for each of the five cross-validation folds.The left panel (a) [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: True versus predicted photometric metallicity values from the [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Comparison of true photometric metallicity values from Muraveva et al. (2025) with the metallicity predicted by the GRU [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Sky distribution of 258,696 RRLs from the cleaned [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

70 extracted references · 47 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archiveprefix author booktitle chapter edition editor howpublished institution eprint journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 ...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in " " * FUNCTION format....

  3. [3]

    2018, , 477, 1472

    Belokurov, V., Deason, A., Koposov, S., et al. 2018, , 477, 1472

    work page 2018

  4. [4]

    W., & Du, Y

    Belokurov, V., Evans, N. W., & Du, Y. L. 2003, , 341, 1373

    work page 2003

  5. [5]

    W., & Le Du, Y

    Belokurov, V., Evans, N. W., & Le Du, Y. 2004, , 352, 233

    work page 2004

  6. [6]

    1907, Astronomische Nachrichten, 175, 325

    Bla z ko , S. 1907, Astronomische Nachrichten, 175, 325

    work page 1907

  7. [7]

    2024, , 527, 12196

    Bobrick , A., Iorio , G., Belokurov , V., et al. 2024, , 527, 12196

    2024

  8. [8]

    Chadid , M., Sneden , C., & Preston , G. W. 2017, , 835, 187

    2017

Show all 70 references
  1. [9]

    2014, arXiv preprint arXiv:1406.1078

    Cho, K., Van Merri \"e nboer, B., Gulcehre, C., et al. 2014, arXiv preprint arXiv:1406.1078

    2014 arXiv

  2. [10]

    1995, , 110, 2319

    Clementini , G., Carretta , E., Gratton , R., et al. 1995, , 110, 2319

    1995

  3. [11]

    2012, , 756, 108

    Clementini , G., Cignoni , M., Contreras Ramos , R., et al. 2012, , 756, 108

    2012

  4. [12]

    2023, , 674, A18

    Clementini, G., Ripepi, V., Garofalo, A., et al. 2023, , 674, A18

    2023

  5. [13]

    2019, , 622, A60

    Clementini, G., Ripepi, V., Molinaro, R., et al. 2019, , 622, A60

    2019

  6. [14]

    T., Martin, R

    Connor, J. T., Martin, R. D., & Atlas, L. E. 1994, IEEE transactions on neural networks, 5, 240

    1994

  7. [15]

    F., et al

    Crestani, J., Fabrizio, M., Braga, V. F., et al. 2021, , 908, 20

    2021

  8. [16]

    2006, , 653, L109

    Dall'Ora , M., Clementini , G., Kinemuchi , K., et al. 2006, , 653, L109

    2006

  9. [17]

    2007, , 475, 1159

    Debosscher, J., Sarro, L., Aerts, C., et al. 2007, , 475, 1159

    2007

  10. [18]

    & Grebel, E

    D \'e k \'a ny, I. & Grebel, E. K. 2022, , 261, 33

    2022

  11. [19]

    K., & Pojma \'n ski, G

    D \'e k \'a ny, I., Grebel, E. K., & Pojma \'n ski, G. 2021, , 920, 33

    2021

  12. [20]

    R., et al

    D'Orazi , V., Storm , N., Casey , A. R., et al. 2024, , 531, 137

    2024

  13. [21]

    2013, , 763, 32

    Drake, A., Catelan, M., Djorgovski, S., et al. 2013, , 763, 32

    2013

  14. [22]

    F., Crestani , J., et al

    Fabrizio , M., Braga , V. F., Crestani , J., et al. 2021, , 919, 118

    2021

  15. [23]

    Gaia Collaboration , Prusti , T., de Bruijne , J. H. J., et al. 2016, , 595, A1

    2016

  16. [24]

    Gaia Collaboration , Vallenari , A., Brown , A. G. A., et al. 2023, , 674, A1

    2023

  17. [25]

    2013, , 767, 62

    Garofalo , A., Cusano , F., Clementini , G., et al. 2013, , 767, 62

    2013

  18. [26]

    2021, , 916, 10

    Garofalo, A., Tantalo, M., Cusano, F., et al. 2021, , 916, 10

    2021

  19. [27]

    K., Chaboyer , B., Marengo , M., et al

    Gilligan , C. K., Chaboyer , B., Marengo , M., et al. 2021, , 503, 4719

    2021

  20. [28]

    2016, ., 1

    Goodfellow, I., Bengio, Y., Courville, A., & Bengio, Y. 2016, ., 1

    2016

  21. [29]

    K., & Jurcsik, J

    Hajdu, G., D \'e k \'a ny, I., Catelan, M., Grebel, E. K., & Jurcsik, J. 2018, , 857, 55

    2018

  22. [30]

    Hoerl, A. E. & Kennard, R. W. 1970, Technometrics, 12, 55

    1970

  23. [31]

    & Belokurov , V

    Iorio , G. & Belokurov , V. 2019, , 482, 3868

    2019

  24. [32]

    & Belokurov , V

    Iorio , G. & Belokurov , V. 2021, , 502, 5686

    2021

  25. [33]

    & Kovacs, G

    Jurcsik, J. & Kovacs, G. 1996, , 312, 111

    1996

  26. [34]

    Kingma, D. P. & Ba, J. 2014, arXiv preprint arXiv:1412.6980

    2014 arXiv

  27. [35]

    2018, Journal of Machine Learning Research, 18, 1

    Li, L., Jamieson, K., DeSalvo, G., Rostamizadeh, A., & Talwalkar, A. 2018, Journal of Machine Learning Research, 18, 1

    2018

  28. [36]

    C., & Zhang, H.-W

    Li, X.-Y., Huang, Y., Liu, G.-C., Beers, T. C., & Zhang, H.-W. 2023, , 944, 88

    2023

  29. [37]

    C., Huang , Y., Zhang , H

    Liu , G. C., Huang , Y., Zhang , H. W., et al. 2020, , 247, 68

    2020

  30. [38]

    2008, Astronomische Nachrichten: Astronomical Notes, 329, 288

    Mahabal, A., Djorgovski, S., Turmon, M., et al. 2008, Astronomische Nachrichten: Astronomical Notes, 329, 288

    2008

  31. [39]

    2015, , 812, 2

    Moln \'a r, L., P \'a l, A., Plachy, E., et al. 2015, , 812, 2

    2015

  32. [40]

    2024, Sensors, 24, 5203

    Monti, L., Muraveva, T., Clementini, G., & Garofalo, A. 2024, Sensors, 24, 5203

    2024

  33. [41]

    M., Wahl , J

    Morgan , S. M., Wahl , J. N., & Wieckhorst , R. M. 2007, , 374, 1421

    2007

  34. [42]

    2020, , 499, 4040

    Muraveva, T., Clementini, G., Garofalo, A., & Cusano, F. 2020, , 499, 4040

    2020

  35. [43]

    2025, , 536, 2749

    Muraveva, T., Giannetti, A., Clementini, G., Garofalo, A., & Monti, L. 2025, , 536, 2749

    2025

  36. [44]

    M., Cohen, J

    Nemec, J. M., Cohen, J. G., Ripepi, V., et al. 2013, , 773, 181

    2013

  37. [45]

    2015, , 447, 2404

    Pancino , E., Britavskiy , N., Romano , D., et al. 2015, , 447, 2404

    2015

  38. [46]

    2011, the Journal of machine Learning research, 12, 2825

    Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, the Journal of machine Learning research, 12, 2825

    2011

  39. [47]

    Preston, G. W. 1959, Astrophysical Journal, vol. 130, p. 507, 130, 507

    1959

  40. [48]

    W., Starr, D

    Richards, J. W., Starr, D. L., Butler, N. R., et al. 2011, , 733, 10

    2011

  41. [49]

    1996, in Automatic Speech and Speaker Recognition: Advanced Topics (Springer), 233--258

    Robinson, T., Hochberg, M., & Renals, S. 1996, in Automatic Speech and Speaker Recognition: Advanced Topics (Springer), 233--258

    1996

  42. [50]

    Rodriguez, P., Wiles, J., & Elman, J. L. 1999, Connection Science, 11, 5

    1999

  43. [51]

    R., Cohen, J

    Sesar, B., Banholzer, S. R., Cohen, J. G., et al. 2014, , 793, 135

    2014

  44. [52]

    M., Soszy \'n ski, I., Udalski, A., et al

    Skowron, D. M., Soszy \'n ski, I., Udalski, A., et al. 2016, , 66, 269

    2016

  45. [53]

    Smith, H. A. 2004, RR Lyrae Stars (Cambridge University Press)

    2004

  46. [54]

    2005, , 55, 59

    Smolec , R. 2005, , 55, 59

    2005

  47. [55]

    2014, The journal of machine learning research, 15, 1929

    Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., & Salakhutdinov, R. 2014, The journal of machine learning research, 15, 1929

    2014

  48. [56]

    W., Bergmeir, C., Petitjean, F., & Webb, G

    Tan, C. W., Bergmeir, C., Petitjean, F., & Webb, G. I. 2021, Data Mining and Knowledge Discovery, 35, 1032

    2021

  49. [57]

    1996, Journal of the Royal Statistical Society Series B: Statistical Methodology, 58, 267

    Tibshirani, R. 1996, Journal of the Royal Statistical Society Series B: Statistical Methodology, 58, 267

    1996

  50. [58]

    E., et al

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261

    2020

  51. [59]

    & Eyer, L

    Willemsen, P. & Eyer, L. 2007, arXiv preprint arXiv:0712.2898

    2007 arXiv

  52. [60]

    Williams, R. J. & Zipser, D. 1989, Neural computation, 1, 270

    1989

  53. [61]

    2004, , 128, 2965

    Wo \'z niak, P., Williams, S., Vestrand, W., & Gupta, V. 2004, , 128, 2965

    2004

  54. [62]

    1966, in Stellar Evolution, ed.\ R

    Baker, N. 1966, in Stellar Evolution, ed.\ R. F. Stein,& A. G. W. Cameron (Plenum, New York) 333

    1966

  55. [63]

    1988, A&A, 200, 58

    Balluch, M. 1988, A&A, 200, 58

    1988

  56. [64]

    Cox, J. P. 1980, Theory of Stellar Pulsation (Princeton University Press, Princeton) 165

    1980

  57. [65]

    N.,& Stewart, J

    Cox, A. N.,& Stewart, J. N. 1969, Academia Nauk, Scientific Information 15, 1

    1969

  58. [66]

    1980, Prog

    Mizuno H. 1980, Prog. Theor. Phys., 64, 544

    1980

  59. [67]

    Tscharnuter W. M. 1987, A&A, 188, 55

    1987

  60. [68]

    1992, in ASP Conf

    Terlevich, R. 1992, in ASP Conf. Ser. 31, Relationships between Active Galactic Nuclei and Starburst Galaxies, ed. A. V. Filippenko, 13

    1992

  61. [69]

    Yorke, H. W. 1980a, A&A, 86, 286

  62. [70]

    F., Tytler, D

    Zheng, W., Davidsen, A. F., Tytler, D. & Kriss, G. A. 1997, preprint

    1997

Pith tools

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