Pith. sign in

REVIEW 3 major objections 7 minor 1 cited by

$Lux$: A generative, multi-output, latent-variable model for astronomical data with noisy labels

T0 review · 3 major / 7 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Lux, a generative latent-variable model, treats stellar labels and spectra as two views of one shared latent vector, and the paper shows that a linear realization matches ASPCAP precision and transfers GALAH abundances to APOGEE stars.

desk verdict A useful linear generative model for spectra and labels, honestly validated in-survey; the cross-survey transfer for line-free elements is correlation-based, and the paper needs a baseline comparison and code. read the letter →

arxiv 2502.01745 v2 pith:EEXMOQZN submitted 2025-02-03 astro-ph.IM

Danny Horta , Adrian M. Price-Whelan , David W. Hogg , Melissa K. Ness , Andrew R. Casey This is my paper · ORCID
classification astro-ph.IM
keywords generativelatentvariablemodelstellarlabelinferencemulti-outputspectroscopicsurveysAPOGEEGALAHtransferuncertainty-awaremachinelearning
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

Lux is a generative latent-variable model that models a star's spectrum and its measured stellar labels as two different views of one shared latent vector, with Gaussian noise whose variances are the reported measurement uncertainties. Because the model is generative and multi-output, it can train on stars with missing labels and on label values with known errors, which most data-driven spectral models treat as noiseless ground truth. The paper demonstrates that a purely linear Lux model trained on APOGEE spectra reproduces ASPCAP stellar labels on held-out stars with biases below the reported label uncertainties and RMSE values comparable to those uncertainties, across red giants, lower signal-to-noise spectra, and open-cluster benchmark stars. It further shows that the same simple model, trained on APOGEE spectra paired with GALAH labels for overlapping stars, can predict GALAH abundances for APOGEE-only spectra, including elements like [Eu/Fe] and [Li/Fe] that cannot be directly measured from APOGEE's near-infrared lines.

What carries the argument

The central mechanism is a multi-output linear latent-variable decoder: one latent vector $z_n$ per star is projected to labels by $\ell_n = A\,z_n$ and to fluxes by $f_n = B\,z_n$, with Gaussian noise of reported variances on both outputs, plus a learned per-pixel scatter term $s$ on the fluxes. The model maximizes the product of the per-star Gaussian likelihoods over $A$, $B$, the latent vectors, and $s$, with L2 regularization on the latents. Because the same latent vector feeds both decoders, the full spectrum constrains every label, missing labels can be handled by inflating their variances to large values, and a survey-to-survey transfer is trained by pairing spectra from one survey with labels from another on cross-matched stars.

What would settle it

Train Lux on the APOGEE-GALAH overlap sample, then apply it to a set of APOGEE giants with independently measured $[\mathrm{Eu/Fe}]$ from high-resolution optical spectra (e.g., r-process-enhanced stars) and compare the Lux-predicted values to those direct measurements; if Lux fails to track the independent values, the correlation-based transfer is falsified, and if it tracks them, the transfer is at least partly causal.

Watch

Extended reading notes

Core claim

The paper claims that a single shared latent vector per star, projected linearly onto both the label space (through the matrix $A$) and the spectral-flux space (through the matrix $B$), is sufficient for data-driven stellar label inference. Lux places the reported label values and their uncertainties directly into a Gaussian likelihood, so the objective is the joint probability of labels and fluxes; missing labels are entered with a very large variance so they are effectively ignored during training. A trained model reads a new spectrum by optimizing that star's latent vector against the fixed $B$ matrix and then maps the latent vector through $A$ to obtain labels. In held-out APOGEE red-giant tests this recovers ASPCAP labels with biases smaller than the reported ASPCAP uncertainties and RMSE values comparable to those uncertainties; the same procedure, trained on APOGEE spectra paired with GALAH labels for 4,000 overlapping giants, predicts GALAH abundances for the APOGEE-only test giants. The paper flags that for elements without detectable lines in the APOGEE passband (e.g., $[\mathrm{Eu/Fe}]$, $[\mathrm{Li/Fe}]$), the transferred values are likely inherited from label-label correlations rather than from spectral-feature causality.

Load-bearing premise

That a single linear mapping from a shared latent vector, with Gaussian noise, captures the relationship between spectra and labels across stellar types and between surveys.

Editorial extensions

If this is right

  • Lux can emulate a physics-based label pipeline (ASPCAP) on APOGEE spectra with biases smaller than the reported label uncertainties and RMSE comparable to those uncertainties, across red giants, lower-SNR spectra, and open-cluster benchmark stars.
  • A linear Lux model trained on cross-matched APOGEE spectra and GALAH labels predicts GALAH abundances for APOGEE-only giants, including elements like [Eu/Fe], [Y/Fe], and [Li/Fe] that are not directly measurable from APOGEE spectra alone.
  • Because the likelihood folds in label uncertainties and inflated variances for missing labels, Lux can train on partial label sets and still produce labels for stars lacking a measurement, e.g., stars with no GALAH [Li/Fe] measurement receive a median-imputed value rather than failing.
  • Lux is fast enough for large surveys: training on 5,000 stars takes about 30 minutes on one CPU and testing on 10,000 stars about 20 minutes, making it practical for emulating pipelines on millions of spectra.
  • The framework is extensible in principle to additional outputs (e.g., photometry, astrometry, kinematics) and to nonlinear decoders, though the paper only demonstrates the linear version.

Reading between the lines

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

  • If the linear shared-latent assumption holds, Lux's transferred abundances for elements without detectable lines (e.g., [Eu/Fe]) are only as trustworthy as the stability of the label-label correlations in the training set; applying it to chemically unusual populations (e.g., r-process-enhanced stars) would test whether those correlations hold.
  • The same machinery could be pointed the other way: train on GALAH spectra with APOGEE labels to transfer APOGEE-quality labels to GALAH targets, or add Gaia photometry as an extra output block to infer distances and extinctions.
  • A direct spectral-fidelity test comparing visible weak lines of a transferred element in the best-fit spectrum would separate causal line-driven transfer from correlation-driven transfer, a distinction the paper explicitly leaves open.
  • The near-unity reduced chi-squared on held-out spectra suggests the Gaussian-noise model with per-pixel scatter captures the dominant variance; extending Lux to non-Gaussian noise or nonlinear decoders might be required when moving to surveys with different systematics.
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

3 major / 7 minor

Summary. This paper presents Lux, a generative multi-output latent-variable model that links stellar spectra and stellar labels through a shared latent vector z with linear transformations A and B (Eqs. 1-2). The Gaussian likelihood (Eqs. 3-6) includes label and flux uncertainties and an extra per-pixel scatter term; the parameters are fit by maximizing the penalized likelihood (Eq. 7) with a two-part block-coordinate schedule (Section 4.2) and hyperparameters (latent dimensionality P, regularization strength Omega) chosen by K-fold cross-validation (Appendix A). The authors demonstrate emulation of ASPCAP labels from APOGEE spectra for high- and low-SNR RGB stars, open cluster members, and mixed stellar types (Section 4), and label transfer from GALAH to APOGEE including elements without strong APOGEE spectral features (Section 5), with additional tests on synthetic Korg spectra (Appendix C) and latent-space visualization (Appendix D).

Significance. If the reported results hold, Lux is a useful addition to the data-driven stellar spectroscopy toolbox: the likelihood formulation explicitly incorporates label and flux uncertainties plus missing data (Eqs. 3-6), the implementation is clearly specified, and the held-out validations span SNR regimes, open clusters, mixed stellar types, and a cross-survey transfer, each with reported RMSE and bias statistics. The in-survey emulation results in Section 4 are credible and do not depend on the transfer caveats. The paper also provides a K-fold procedure for setting the latent dimensionality and regularization strength, and an explicit statement of the correlation-versus-causation limitation for line-free elements. However, the lack of a quantitative baseline comparison and the overstatement of the transfer and RMSE results currently temper the strength of the claims.

major comments (3)
  1. [Section 5, Figure 17] The headline claim that Lux is successful at performing label transfer between APOGEE and GALAH is not established for elements without spectral features in the APOGEE band, for which the paper itself provides the caveat that predictions may be based on correlation rather than causation. The held-out test (samples G and H, Section 2.2) draws from the same APOGEE-GALAH cross-match as the training set, so the validation only demonstrates interpolation within the label-correlation structure of that single parent sample. For [Li/Fe], [Y/Fe], and [Eu/Fe], the APOGEE wavelength region has no known features, and the Figure 17 RMSE values are several times the reported GALAH uncertainties (e.g., [Li/Fe] RMSE 0.404 vs mean sigma_GALAH 0.08 dex; [Y/Fe] 0.201 vs 0.11 dex). To support the transfer claim, the authors should either validate on a population-disjoint held-out set (e.g., metallicity- or age-disjoint) demonstrating that the learned correlations generalize, or qualify the abstract and conclusions to present these as correlation-based estimates that require follow-up. As written, the abstract's unqualified claim of successful label transfer overstates what is demonstrated.
  2. [Sections 1, 4.5, and 7] The paper motivates Lux by limitations of existing data-driven methods (The Cannon, The Payne), asserting that prior models assume labels are ground truth and can lead to biases, but it provides no quantitative comparison against either method on the same training and test data. The only side-by-side is the qualitative inspection in Figure 8, and the statements that Lux labels appear tighter and show less scatter are not supported by measured metrics. Since the central novelty claim is that Lux properly accounts for uncertainties and missing data, a benchmark against The Cannon (and ideally The Payne) on identical splits would substantiate the claimed improvements and is standard for a methods paper of this kind.
  3. [Section 4.5, Figure 7] The text and the Figure 7 caption state that for most labels the RMSE is comparable to the average ASPCAP uncertainty, but the numbers in the figure contradict this for several important labels: Teff RMSE is 57.8 K versus mean sigma_ASPCAP of 9.44 K, log g RMSE 0.134 versus 0.03, [Fe/H] 0.054 versus 0.01, and [N/Fe] 0.092 versus 0.02. Only a few labels (e.g., [Si/Fe], [Ni/Fe]) lie within roughly a factor of two of the reported uncertainty. The text and caption should report the RMSE-to-uncertainty ratios accurately or reframe the claim in terms of absolute precision rather than comparability to ASPCAP uncertainties.
minor comments (7)
  1. [Section 4.3, Eq. (10)] The test-step objective in Eq. (10) optimizes z using only the flux likelihood and applies no L2 penalty, while the training objective in Eq. (7) applies a sizable penalty (Omega = 10^3); the authors should state whether this asymmetry is intentional and confirm that the reported precision estimates are robust to including the penalty in the test step.
  2. [Figure 3 caption] The caption contains a typo (high-SNR fiend RGB-test) that should be corrected, and the text label Spectra for giant stars appears to be a figure title rather than part of the caption.
  3. [Appendix C, Figure 18] The Korg spectra are attributed to Wheeler et al. (2022) in the Appendix C text and Figure 18 caption but to Wheeler et al. (2023) in Section 1; the citations should be reconciled.
  4. [Section 4.6, Figure 10] The precision values in Figure 10 measure only the conditional label uncertainty under the fitted model (from flux realizations) and do not include model mis-specification or parameter uncertainty; the text should state this limitation when summarizing the precision claims.
  5. [Section 7] No code availability statement or repository link is provided; a public release of the Lux code would materially aid reproducibility of the optimization schedule and hyperparameter choices.
  6. [Section 3.1] The model assumes Gaussian label noise with variances taken from the catalogs; since underestimated label uncertainties could bias the joint likelihood, the authors should briefly report the experiment they mention in Section 3.1 in which per-label scatter terms were found to be unnecessary.
  7. [Section 4.5] Because Lux is trained to reproduce ASPCAP labels, the in-survey validation demonstrates fidelity as an emulator rather than astrophysical accuracy of the labels; a sentence clarifying this distinction would prevent readers from over-interpreting the agreement.

Circularity Check

0 steps flagged · score 2.0 of 10

Lux's held-out label predictions are genuine held-out predictions; the line-free label-transfer caveat is disclosed in the manuscript and does not reduce the central derivation to its inputs by construction.

full rationale

The derivation chain is self-contained as a supervised generative model: Lux specifies linear decoders l_n = A z_n (Eq. 1) and f_n = B z_n (Eq. 2), optimizes A, B, z, and scatter terms on a training sample, and then infers test-set latent vectors from fluxes alone via Eq. 10 before projecting through the trained A. Training and test stars are disjoint by the sample definitions in Section 2.2, so the ASPCAP emulation results (Figures 7, 9, 11, 12) are genuine held-out predictions, not fits to the test labels. The validation against ASPCAP is self-referential only in the intended sense of emulation: the paper explicitly frames the agreement as 'the Lux model is emulating the ASPCAP pipeline well,' not as an independent physical calibration. The APOGEE-to-GALAH label transfer for line-free elements ([Eu/Fe], [Y/Fe], [Li/Fe]) does rely on training-set label correlations, and the paper itself states: 'the prediction may be based on correlation rather than causation' and 'The model may therefore fail to correctly infer these abundances for stars with different label-correlation behaviors.' That is a disclosed extrapolation limitation, not a circular reduction: the held-out test in Figure 17 is still disjoint from training and tests interpolation within the same cross-match population. Self-citations, such as the Cannon initialization procedure (Ness et al. 2015) and cluster-removal catalog (Horta et al. 2020), are procedural and not load-bearing for the claims; no uniqueness theorem or prior result is invoked to force the model choice. No specific circular step can be exhibited from the paper's equations or argument, so the score reflects the minor self-referential validation target and the correlation caveat rather than actual circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central model depends on a small set of free parameters: the latent dimensionality P and regularization strength Omega are chosen by cross-validation, and the per-pixel scatter vector s_f is fitted. The main assumptions are linearity of the label/flux mapping, Gaussian noise, and the validity of survey labels as targets. No new physical entities are postulated.

free parameters (3)
  • P (latent dimensionality) = 4M (M = number of labels)
    Chosen via five-fold cross-validation (Appendix A); controls model flexibility and is not derived from first principles.
  • Omega (L2 regularization strength) = 10^3
    Chosen via five-fold cross-validation (Appendix A); penalizes the latent vectors and affects generalisation.
  • s_f (per-pixel scatter vector) = initialized ln s = -8, then optimized in Agenda 2
    Free parameters in the flux likelihood (Eq. 4) intended to capture intrinsic scatter and uncharacterised systematics; fitted to training data.
assumptions (5)
  • domain assumption Gaussian likelihood for labels and fluxes with known catalog variances (Eqs. 3-4).
    The model assumes the noise on labels and fluxes is Gaussian with variances given by survey pipelines; this is standard but not proven.
  • domain assumption Linear generative mapping from latent z to labels and fluxes (Eqs. 1-2).
    The paper chooses linear transformations and demonstrates they work for the tested regimes; this is the main structural assumption and is load-bearing for the transfer claims.
  • ad hoc to paper Missing labels can be imputed by median value with sigma=9999 (Section 2.1).
    The paper sets missing labels to the median and inflates their uncertainty so they are effectively ignored by the likelihood; this avoids improper gradients but is a modeling choice, not a principled missing-data mechanism.
  • ad hoc to paper L2 regularization on latent vectors z is beneficial (Eq. 7).
    The paper states that L2 regularization improves held-out performance; the choice is motivated empirically, not derived.
  • domain assumption ASPCAP and GALAH catalog labels are reliable enough to serve as training targets and validation references.
    The entire evaluation compares Lux predictions to catalog labels; any systematic errors in those catalogs are inherited by Lux and the validation statistics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of $Lux$: A generative, multi-output, latent-variable model for astronomical data with noisy labels." pith.science (2026). https://pith.science/paper/EEXMOQZN

@misc{pith2026250201745,
  author       = {Pith},
  title        = {Pith review of: $Lux$: A generative, multi-output, latent-variable model for astronomical data with noisy labels},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EEXMOQZN}},
  note         = {Machine review of arXiv:2502.01745}
}
abstract

The large volume of spectroscopic data available now and from near-future surveys will enable high-dimensional measurements of stellar parameters and properties. Current methods for determining stellar labels from spectra use physics-driven models, which are computationally expensive and have limitations in their accuracy due to simplifications. While machine learning methods provide efficient paths toward emulating physics-based pipelines, they often do not properly account for uncertainties and have complex model structure, both of which can lead to biases and inaccurate label inference. Here we present $Lux$: a data-driven framework for modeling stellar spectra and labels that addresses prior limitations. $Lux$ is a generative, multi-output, latent variable model framework built on JAX for computational efficiency and flexibility. As a generative model, $Lux$ properly accounts for uncertainties and missing data in the input stellar labels and spectral data and can either be used in probabilistic or discriminative settings. Here, we present several examples of how $Lux$ can successfully emulate methods for precise stellar label determinations for stars ranging in stellar type and signal-to-noise from the $APOGEE$ surveys. We also show how a simple $Lux$ model is successful at performing label transfer between the $APOGEE$ and $GALAH$ surveys. $Lux$ is a powerful new framework for the analysis of large-scale spectroscopic survey data. Its ability to handle uncertainties while maintaining high precision makes it particularly valuable for stellar survey label inference and cross-survey analysis, and the flexible model structure allows for easy extension to other data types.

Figures

Figures reproduced from arXiv: 2502.01745 by the authors.

Figure 1
Figure 1. Graphical model of Lux. Here, ℓ represents labels, f represents flux, z are the latent variables, A and B represent the matrices that project the latent variables into stellar labels and stellar fluxes, respectively, s is the vector of scatter terms at every pixel to account for underestimated uncertainties in the flux measurements, and σℓ and σf represent the uncertainties in the labels and flux, respectively. See … view at source ↗
Figure 2
Figure 2. A Flowchart summarizing the optimization scheme for our application of the linear Lux model to APOGEE data, described in Section 4.2. with a test set (i.e. data not included in the training) with held out labels or spectra, we must first determine the corresponding latent vectors for the test set stars. For evaluating the performance of Lux on the test data, we have several options. One option is to use the spectra … view at source ↗
Figure 3
Figure 3. A comparison of APOGEE spectra (black lines) and predicted spectra from Lux (navy and cyan lines) for six stars (each panel is for one star). We use Lux to predict stellar spectra in two ways. First, we use the stellar labels for each star to infer latent representations (zn) for the test objects, and then use the model to transform latent vectors to stellar flux (navy lines). Then, as a self-test of the model, we u… view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: Reduced χ 2 value for all stars in the validation/test set using the stellar fluxes. The median value (dotted line) is 1.07. These stellar fluxes are generated from the latent vectors inferred from the stellar spectra themselves. χ 2 r is computed as χ 2 divided by the…
Figure 5
Figure 5. Figure 5: Comparison of portions of the spectrum for two stars (see legend) in the high-SNR field RGB-test sample with similar Teff , log g, and [Fe/H] but different individual chemical abundance ratios. The stellar spectra shown in each panel are the Lux model spectra for those…
Figure 6
Figure 6. Figure 6: Top: Portion of the APOGEE spectrum (black) and inferred Lux spectra for a random star (2M00011399+8408446) from our High-SNR field RGB-test set. The meaning of the cyan and navy spectra shown in the top panel is the same as in [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Validation for the high-SNR field RGB-test set stars using the twelve labels trained/tested on. Each panel shows the Lux-predicted label values plotted against the ASPCAP values. The Lux labels are determined by optimizing the test set latent representations using each…
Figure 8
Figure 8. Figure 8: Kiel (top) and Tinsley-Wallerstein (bottom) diagrams for stars in the high-SNR field RGB-test sample. We show Lux labels (left), the labels computed using the Cannon (Ness et al. 2015) assuming a quadratic relationship in the labels (middle), and the labels from ASPCAP…
Figure 9
Figure 9. Figure 9: Validation results for RGB stars at lower signal-to-noise (low-SNR field RGB-test). Here, we have chosen a signal-to-noise range that is expected for the SDSS-V Galactic Genesis survey (Kollmeier et al. 2017). As in [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: The Lux model label precision as a function of signal-to-noise for effective temperature (left), surface gravity (middle), and element abundance ratios (right). The median (solid) and (16th , 84th) inter-quartile range (shaded) are computed by imputing ten realization…
Figure 11
Figure 11. Figure 11: Replica of [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: Replica of [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]
Figure 13
Figure 13. Figure 13: Six GALAH chemical abundances for 1,000 APOGEE giant stars (GALAH-APOGEE field giants-test set), determined using an Lux label tranfer model trained on 4,000 stars APOGEE DR17 spectra and GALAH DR3 labels ((GALAH-APOGEE field giants-train set)). Here Lux labels are in…
Figure 14
Figure 14. Figure 14: Median resulting total χ 2 values from the K-fold cross validation test, summed over all wavelengths, all labels, and all stars in the high-SNR field RGB-test set. We show the χ 2 metric estimated by computing the test set z latent parameters using each star’s stellar…
Figure 15
Figure 15. Figure 15: Stellar labels determined using Lux for all 10,000 high signal-to-noise RGB stars in the high-SNR field RGB-test set. Overall, the Lux labels look realistic and do not show any unusual trends. Moreover, the scatter around the labels is small, yielding a tight relation…
Figure 16
Figure 16. Figure 16: Validation results for RGB stars at lower SNR (low-SNR field RGB-test). The Lux labels shown are determined by optimizing the latent representations for test stars using each star’s spectral fluxes. Here, we have chosen a SNR range that is expected for the SDSS-V Gala…
Figure 17
Figure 17. Figure 17: Validation results for RGB stars in the GALAH-APOGEE field giants-test set. The Lux labels shown are determined by optimizing the latent representations for test stars using each star’s spectral fluxes. We show the mean reported GALAH label uncertainty, bias, and RMSE…
Figure 18
Figure 18. Figure 18: Validation results on 1,000 test stars with synthetic model Korg (Wheeler et al. 2022) spectra. The Lux labels shown are determined by optimizing the latent representations for test stars using each star’s synthetic (Korg) spectral fluxes. The model used to impute the…
Figure 19
Figure 19. Figure 19: Components of a T-SNE dimensionality reduction (using two components) on the high-SNR field RGB-test set z latent parameters using a perplexity of 25. Each panel is color coded by one of the corresponding twelve stellar labels used to train the model for test set star…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. From stellar light to astrophysical insight: automating variable star research with machine learning

    astro-ph.IM 2025-07 unverdicted

    An invited review of machine learning for automated variable star research, covering data cleaning, variability classification, stellar parameter inference, and foundation models.

Reference graph

Works this paper leans on

75 extracted references · 7 canonical work pages · cited by 1 Pith paper

  1. [1]

    - [1] #1 = = ^ ^ ^ .\!\!^ d .\!\!^ h .\!\!^ m .\!\!^ s .\!\!^ @mss

    thebibliography [1] 20pt to REFERENCES 6pt =0pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key on reference command Each re...

    2019

  2. [2]

    2022, , 259, 35, 10.3847/1538-4365/ac4414

    Abdurro'uf , Accetta , K., Aerts , C., et al. 2022, , 259, 35, 10.3847/1538-4365/ac4414

    doi:10.3847/1538-4365/ac4414 2022

  3. [3]

    C., Wilhelm , R., et al

    Allende Prieto , C., Beers , T. C., Wilhelm , R., et al. 2006, , 636, 804, 10.1086/498131

    doi:10.1086/498131 2006

  4. [5]

    2023 b , , 267, 8, 10.3847/1538-4365/acd53e

    ---. 2023 b , , 267, 8, 10.3847/1538-4365/acd53e

    doi:10.3847/1538-4365/acd53e 2023

  5. [7]

    2021 b , Autoencoders

    ---. 2021 b , Autoencoders. 2003.05991

    arXiv 2021

  6. [8]

    L., Oelkers , R

    Beaton , R. L., Oelkers , R. J., Hayes , C. R., et al. 2021, arXiv e-prints, arXiv:2108.11907. 2108.11907

    arXiv 2021

  7. [9]

    R., Bershady , M

    Blanton , M. R., Bershady , M. A., Abolfathi , B., et al. 2017, , 154, 28, 10.3847/1538-3881/aa7567

    doi:10.3847/1538-3881/aa7567 2017

  8. [10]

    2021, arXiv preprint arXiv:2105.15183

    Blondel, M., Berthet, Q., Cuturi, M., et al. 2021, arXiv preprint arXiv:2105.15183

    arXiv 2021

Show all 75 references
  1. [11]

    S., & Vaughan , A

    Bowen , I. S., & Vaughan , A. H., J. 1973, , 12, 1430, 10.1364/AO.12.001430

    1973 doi

  2. [12]

    2018, JAX : composable transformations of P ython+ N um P y programs, 0.3.13

    Bradbury, J., Frostig, R., Hawkins, P., et al. 2018, JAX : composable transformations of P ython+ N um P y programs, 0.3.13. http://github.com/google/jax

    2018

  3. [13]

    2024, arXiv e-prints, arXiv:2410.16081, 10.48550/arXiv.2410.16081

    Buck , T., & Schwarz , C. 2024, arXiv e-prints, arXiv:2410.16081, 10.48550/arXiv.2410.16081

    2024 doi arXiv

  4. [14]

    2020, arXiv e-prints, arXiv:2011.02505

    Buder , S., Sharma , S., Kos , J., et al. 2020, arXiv e-prints, arXiv:2011.02505. 2011.02505

    2020 arXiv

  5. [15]

    R., Hogg , D

    Casey , A. R., Hogg , D. W., Ness , M., et al. 2016, arXiv e-prints, arXiv:1603.03040, 10.48550/arXiv.1603.03040

    2016 doi arXiv

  6. [16]

    2022, in Machine Learning for Astrophysics, 17, 10.48550/arXiv.2207.02785

    Ciuca , I., & Ting , Y.-S. 2022, in Machine Learning for Astrophysics, 17, 10.48550/arXiv.2207.02785

    2022 doi arXiv

  7. [17]

    P., Koposov, S

    Cooper, A. P., Koposov, S. E., Allende Prieto, C., et al. 2023, The Astrophysical Journal, 947, 37, 10.3847/1538-4357/acb3c0

    2023 doi

  8. [18]

    V., Hasselquist , S., et al

    Cunha , K., Smith , V. V., Hasselquist , S., et al. 2017, , 844, 145, 10.3847/1538-4357/aa7beb

    2017 doi

  9. [19]

    S., Bellido-Tirado , O., Chiappini , C., et al

    de Jong , R. S., Bellido-Tirado , O., Chiappini , C., et al. 2012, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 8446, Ground-based and Airborne Instrumentation for Astronomy IV, ed. I. S. McLean , S. K. Ramsay , & H. Takami , 84460T, 10....

    2012 doi

  10. [20]

    W., Rix , H.-W., & Ness , M

    Eilers , A.-C., Hogg , D. W., Rix , H.-W., & Ness , M. K. 2019, , 871, 120, 10.3847/1538-4357/aaf648

    2019 doi

  11. [21]

    Freeman , K. C. 2012, in Astronomical Society of the Pacific Conference Series, Vol. 458, Galactic Archaeology: Near-Field Cosmology and the Formation of the Milky Way, ed. W. Aoki , M. Ishigaki , T. Suda , T. Tsujimoto , & N. Arimoto , 393

    2012

  12. [22]

    Gaia Collaboration , Brown , A. G. A., Vallenari , A., et al. 2020, arXiv e-prints, arXiv:2012.01533. 2012.01533

    2020 arXiv

  13. [23]

    Gaia Collaboration , Vallenari , A., Brown , A. G. A., et al. 2023, , 674, A1, 10.1051/0004-6361/202243940

    2023 doi

  14. [24]

    E., Allende Prieto , C., Holtzman , J

    Garc \' a P \'e rez , A. E., Allende Prieto , C., Holtzman , J. A., et al. 2016, , 151, 144, 10.3847/0004-6256/151/6/144

    2016 doi

  15. [25]

    2012, The Messenger, 147, 25

    Gilmore , G., Randich , S., Asplund , M., et al. 2012, The Messenger, 147, 25

    2012

  16. [26]

    2024, , 682, A9, 10.1051/0004-6361/202347122

    Guiglion , G., Nepal , S., Chiappini , C., et al. 2024, , 682, A9, 10.1051/0004-6361/202347122

    2024 doi

  17. [27]

    E., Siegmund , W

    Gunn , J. E., Siegmund , W. A., Mannery , E. J., et al. 2006, , 131, 2332, 10.1086/500975

    2006 doi

  18. [28]

    2008, , 486, 951, 10.1051/0004-6361:200809724

    Gustafsson , B., Edvardsson , B., Eriksson , K., et al. 2008, , 486, 951, 10.1051/0004-6361:200809724

    2008 doi

  19. [29]

    J., Davies , G

    Hall , O. J., Davies , G. R., Elsworth , Y. P., et al. 2019, , 486, 3569, 10.1093/mnras/stz1092

    2019 doi

  20. [30]

    2016, , 833, 81, 10.3847/1538-4357/833/1/81

    Hasselquist , S., Shetrone , M., Cunha , K., et al. 2016, , 833, 81, 10.3847/1538-4357/833/1/81

    2016 doi

  21. [31]

    Ho , A. Y. Q., Rix , H.-W., Ness , M. K., et al. 2017 a , , 841, 40, 10.3847/1538-4357/aa6db3

    2017 doi

  22. [32]

    Ho , A. Y. Q., Ness , M. K., Hogg , D. W., et al. 2017 b , , 836, 5, 10.3847/1538-4357/836/1/5

    2017 doi

  23. [33]

    W., Eilers, A.-C., & Rix, H.-W

    Hogg, D. W., Eilers, A.-C., & Rix, H.-W. 2019, The Astronomical Journal, 158, 147, 10.3847/1538-3881/ab398c

    2019 doi

  24. [34]

    P., Mackereth , J

    Horta , D., Schiavon , R. P., Mackereth , J. T., et al. 2020, , 493, 3363, 10.1093/mnras/staa478

    2020 doi

  25. [35]

    Hunter, J. D. 2007, Computing In Science & Engineering, 9, 90, 10.1109/MCSE.2007.55

    2007 doi

  26. [36]

    C., Dalton, G

    Jin, S., Trager, S. C., Dalton, G. B., et al. 2023, Monthly Notices of the Royal Astronomical Society, 530, 2688–2730, 10.1093/mnras/stad557

    2023 doi

  27. [37]

    2014, , 564, A133, 10.1051/0004-6361/201322440

    Jofr \'e , P., Heiter , U., Soubiran , C., et al. 2014, , 564, A133, 10.1051/0004-6361/201322440

    2014 doi

  28. [38]

    2024, SpectraFM: Tuning into Stellar Foundation Models

    Koblischke, N., & Bovy, J. 2024, SpectraFM: Tuning into Stellar Foundation Models. 2411.04750

    2024 arXiv

  29. [39]

    A., Zasowski , G., Rix , H.-W., et al

    Kollmeier , J. A., Zasowski , G., Rix , H.-W., et al. 2017, arXiv e-prints, arXiv:1711.03234, 10.48550/arXiv.1711.03234

    2017 doi arXiv

  30. [40]

    2013, , 146, 134, 10.1088/0004-6256/146/5/134

    Kordopatis , G., Gilmore , G., Steinmetz , M., et al. 2013, , 146, 134, 10.1088/0004-6256/146/5/134

    2013 doi

  31. [41]

    W., & Bovy, J

    Leung, H. W., & Bovy, J. 2018, Monthly Notices of the Royal Astronomical Society, 10.1093/mnras/sty3217

    2018 doi

  32. [42]

    J., Cannon , R

    Lewis , I. J., Cannon , R. D., Taylor , K., et al. 2002, , 333, 279, 10.1046/j.1365-8711.2002.05333.x

    2002

  33. [43]

    Li , J., Wong , K. W. K., Hogg , D. W., Rix , H.-W., & Chandra , V. 2024, , 272, 2, 10.3847/1538-4365/ad2b4d

    2024 doi

  34. [44]

    R., Schiavon , R

    Majewski , S. R., Schiavon , R. P., Frinchaboy , P. M., et al. 2017, , 154, 94, 10.3847/1538-3881/aa784d

    2017 doi

  35. [45]

    K., et al

    Manea , C., Hawkins , K., Ness , M. K., et al. 2024, , 972, 69, 10.3847/1538-4357/ad58d9

    2024 doi

  36. [46]

    L., Sharma , S., Buder , S., et al

    Martell , S. L., Sharma , S., Buder , S., et al. 2017, , 465, 3203, 10.1093/mnras/stw2835

    2017 doi

  37. [47]

    A., Ness, M

    McKinnon, K. A., Ness, M. K., Rockosi, C. M., & Guhathakurta, P. 2024, Data-driven Discovery of Diffuse Interstellar Bands with APOGEE Spectra. 2307.05706

    2024 arXiv

  38. [48]

    E., et al

    M \'e sz \'a ros , S., Holtzman , J., Garc \' a P \'e rez , A. E., et al. 2013, , 146, 133, 10.1088/0004-6256/146/5/133

    2013 doi

  39. [49]

    2022, , 164, 85, 10.3847/1538-3881/ac7ce5

    Myers , N., Donor , J., Spoo , T., et al. 2022, , 164, 85, 10.3847/1538-3881/ac7ce5

    2022 doi

  40. [50]

    W., Rix , H

    Ness , M., Hogg , D. W., Rix , H. W., Ho , A. Y. Q., & Zasowski , G. 2015, , 808, 16, 10.1088/0004-637X/808/1/16

    2015 doi

  41. [51]

    W., Rix , H

    Ness , M., Hogg , D. W., Rix , H. W., et al. 2016, , 823, 114, 10.3847/0004-637X/823/2/114

    2016 doi

  42. [52]

    2016, , 152, 14, 10.3847/0004-6256/152/1/14

    Ness , M., & Lang , D. 2016, , 152, 14, 10.3847/0004-6256/152/1/14

    2016 doi

  43. [53]

    K., Mendel , J

    Ness , M. K., Mendel , J. T., Buder , S., et al. 2024, arXiv e-prints, arXiv:2407.17661, 10.48550/arXiv.2407.17661

    2024 doi arXiv

  44. [54]

    L., Holtzman , J

    Nidever , D. L., Holtzman , J. A., Allende Prieto , C., et al. 2015, , 150, 173, 10.1088/0004-6256/150/6/173

    2015 doi

  45. [55]

    L., Hasselquist, S., Hayes, C

    Nidever, D. L., Hasselquist, S., Hayes, C. R., et al. 2020, The Astrophysical Journal, 895, 88, 10.3847/1538-4357/ab7305

    2020 doi

  46. [56]

    A., Harva , M

    Nolan , L. A., Harva , M. O., Kab \'a n , A., & Raychaudhury , S. 2006, , 366, 321, 10.1111/j.1365-2966.2005.09868.x

    2006

  47. [57]

    2006--, NumPy : A guide to NumPy , USA: Trelgol Publishing

    Oliphant, T. 2006--, NumPy : A guide to NumPy , USA: Trelgol Publishing. http://www.numpy.org/

    2006

  48. [58]

    Piskunov, N., & Valenti, J. A. 2016, Astronomy & Astrophysics, 597, A16, 10.1051/0004-6361/201629124

    2016 doi

  49. [59]

    Price-Whelan , A. M. 2017, The Journal of Open Source Software, 2, 388, 10.21105/joss.00388

    2017 doi

  50. [60]

    2024, arXiv e-prints, arXiv:2407.05751, 10.48550/arXiv.2407.05751

    R \'o \.z a \'n ski , T., Ting , Y.-S., & Jab o \'n ska , M. 2024, arXiv e-prints, arXiv:2407.05751, 10.48550/arXiv.2407.05751

    2024 doi arXiv

  51. [61]

    A., Beaton , R

    Santana , F. A., Beaton , R. L., Covey , K. R., et al. 2021, arXiv e-prints, arXiv:2108.11908. 2108.11908

    2021

  52. [62]

    P., Phillips , S

    Schiavon , R. P., Phillips , S. G., Myers , N., et al. 2024, , 528, 1393, 10.1093/mnras/stad3020

    2024 doi

  53. [63]

    2015, Journal of Astronomical Telescopes, Instruments, and Systems, 1, 035002, 10.1117/1.JATIS.1.3.035002

    Sheinis , A., Anguiano , B., Asplund , M., et al. 2015, Journal of Astronomical Telescopes, Instruments, and Systems, 1, 035002, 10.1117/1.JATIS.1.3.035002

    2015 doi

  54. [64]

    V., Bizyaev , D., Cunha , K., et al

    Smith , V. V., Bizyaev , D., Cunha , K., et al. 2021, , 161, 254, 10.3847/1538-3881/abefdc

    2021 doi

  55. [65]

    2006, , 132, 1645, 10.1086/506564

    Steinmetz , M., Zwitter , T., Siebert , A., et al. 2006, , 132, 1645, 10.1086/506564

    2006 doi

  56. [66]

    2018, , 860, 159, 10.3847/1538-4357/aac6c9

    Ting , Y.-S., Conroy , C., Rix , H.-W., & Asplund , M. 2018, , 860, 159, 10.3847/1538-4357/aac6c9

    2018 doi

  57. [67]

    2019, , 879, 69, 10.3847/1538-4357/ab2331

    Ting , Y.-S., Conroy , C., Rix , H.-W., & Cargile , P. 2019, , 879, 69, 10.3847/1538-4357/ab2331

    2019 doi

  58. [68]

    H., Fragkoudi , F., & Ness , M

    Wheeler , A., Abril-Cabezas , I., Trick , W. H., Fragkoudi , F., & Ness , M. 2022, , 935, 28, 10.3847/1538-4357/ac7da0

    2022 doi

  59. [69]

    J., Abruzzo , M

    Wheeler , A. J., Abruzzo , M. W., Casey , A. R., & Ness , M. K. 2023, , 165, 68, 10.3847/1538-3881/acaaad

    2023 doi

  60. [70]

    C., Hearty , F

    Wilson , J. C., Hearty , F. R., Skrutskie , M. F., et al. 2019, , 131, 055001, 10.1088/1538-3873/ab0075

    2019 doi

  61. [71]

    2017, , 232, 2, 10.3847/1538-4365/aa80e4

    Xiang , M., Liu , X., Shi , J., et al. 2017, , 232, 2, 10.3847/1538-4365/aa80e4

    2017 doi

  62. [72]

    2019, , 245, 34, 10.3847/1538-4365/ab5364

    Xiang , M., Ting , Y.-S., Rix , H.-W., et al. 2019, , 245, 34, 10.3847/1538-4365/ab5364

    2019 doi

  63. [73]

    J., et al

    Yanny , B., Rockosi , C., Newberg , H. J., et al. 2009, , 137, 4377, 10.1088/0004-6256/137/5/4377

    2009 doi

  64. [74]

    A., Frinchaboy , P

    Zasowski , G., Johnson , J. A., Frinchaboy , P. M., et al. 2013, , 146, 81, 10.1088/0004-6256/146/4/81

    2013 doi

  65. [75]

    E., Chojnowski , S

    Zasowski , G., Cohen , R. E., Chojnowski , S. D., et al. 2017, , 154, 198, 10.3847/1538-3881/aa8df9

    2017 doi

  66. [76]

    2008, Machine Learning, 73, 221, 10.1007/s10994-008-5050-1

    Zhang, J., Ghahramani, Z., & Yang, Y. 2008, Machine Learning, 73, 221, 10.1007/s10994-008-5050-1

    2008 doi

  67. [77]

    2012, Research in Astronomy and Astrophysics, 12, 723, 10.1088/1674-4527/12/7/002

    Zhao , G., Zhao , Y.-H., Chu , Y.-Q., Jing , Y.-P., & Deng , L.-C. 2012, Research in Astronomy and Astrophysics, 12, 723, 10.1088/1674-4527/12/7/002

    2012 doi

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.