Pith. sign in

REVIEW 4 major objections 6 minor 68 references

Lyman Break Galaxy selection and redshift measurement with supervised contrastive learning

T0 review · 4 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read Supervised weighted contrastive learning separates DESI Lyman-break galaxies from contaminants better than the current line-detector network, at equal redshift accuracy.

desk verdict Classification win is real; redshift parity is conditional on template-based augmentation that lacks independent high-z validation. read the letter →

arxiv 2608.10080 v1 pith:NNNVGFAW submitted 2026-08-10 astro-ph.CO

classification astro-ph.CO
keywords LymanbreakgalaxiescontrastivelearningredshiftestimationspectralclassificationDESIdeepgalaxycontaminantsK-nearestneighbors
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 supervised weighted contrastive learning can handle both tasks needed for a faint high-redshift galaxy sample: separating Lyman break galaxies (LBGs) from quasars and low-redshift emission line galaxies, and estimating redshifts, using a small, visually inspected training set. The proposed zlbg pipeline is tested on DESI pilot spectra and reports stronger contaminant classification than the current lbgNET line-detector network (AUC 0.997 versus 0.988) with comparable redshift identification (AUC 0.790 versus 0.799). If this result holds, DESI Run 2 could obtain a cleaner LBG sample at equal redshift completeness, which matters for primordial non-Gaussianity measurements, Ly-alpha forest tomography, and void catalogs that rely on high-redshift tracers.

What carries the argument

The load-bearing mechanism is a supervised weighted contrastive loss. For each batch of spectra, a relationship coefficient multiplies the contrastive log-softmax: a class-similarity matrix sets pairwise weights among the five classes (LBGa, LBGme, LBGse, ELG, QSO), and for LBG pairs a Gaussian kernel in redshift, w(z_a,z_i) = exp(-(z_a-z_i)^2 / (2 $sigma_z^{2}$ (1+mean z)^2)) with sigma_z = 0.025, sharpens the embedding by redshift proximity. The encoder is the same Conv1D backbone as lbgNET; the projection head is discarded after training, and downstream tasks use the representation layer: a small MLP for classification and a 50-neighbor KNN on cosine distance for redshift, with a quality flag q_z derived from neighborhood scatter. Data augmentation splices the four LBG stacked templates into observed spectra to generate redshift coverage up to z = 4.5 and SNR variation via different exposure co-adds.

What would settle it

Take newly observed DESI survey-validation LBG spectra with secure visual-inspection redshifts at z > 3.8, run zlbg's KNN redshift and classification head, and compare purity and outlier fraction against the z < 3.8 test set; if high-redshift objects show a sharp drop in redshift accuracy or a rise in contaminant misclassification, the template-splicing augmentation is not representative of real spectra at the redshifts it was designed to cover.

Watch

Extended reading notes

Core claim

The central claim is that replacing the line-finder readout of the DESI LBG network with a contrastively trained embedding changes what the network learns about spectra: instead of reporting confidence in individual spectroscopic lines, it organizes spectra by galaxy type and by redshift, so contaminants separate cleanly while redshift proximity is encoded continuously. On the same visually inspected test set, zlbg reaches AUC 0.997 for contaminant selection versus 0.988 for lbgNET, and AUC 0.790 versus 0.799 for redshift identification, a difference the paper reads as comparable. The redshift information is meant to be consumed as a prior for redrock template fitting, and the paper shows that broadening the prior to a half-width of 0.035(1+z) recovers more sources with zlbg than with lbgNET, evidence that the embedding is less prone to catastrophic line misidentification.

Load-bearing premise

The whole approach rests on the assumption that redshift-augmented training spectra, built by splicing stacked LBG templates into observed spectra to reach z up to 4.5, faithfully represent real LBG spectra at redshifts where visual inspection has little coverage (above about z = 3.8), and that the templates themselves, derived from visually inspected spectra plus lbgNET's high-confidence classifications, carry no systematic error that the contrastive loss will learn.

Editorial extensions

If this is right

  • Switching DESI LBG processing to zlbg, or combining it with lbgNET, would raise LBG sample purity while keeping redshift completeness, since the two confidence scores are weakly correlated.
  • Because zlbg outputs per-class probabilities rather than a single line confidence, it adds information about LBG subtype (LBGa, LBGme, LBGse) and about the nature of contaminants, which is useful for the separate LAE program.
  • The redshift-aligned embedding should be less sensitive to catastrophic outliers caused by misidentified emission lines, and the paper shows performance improves relative to lbgNET as the redrock prior width is increased to 0.035(1+z).
  • The same architecture can be retrained on as-yet-unobserved DESI Run 2 survey validation data; the paper explicitly leaves larger datasets for future work.

Reading between the lines

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

  • If the template-splicing augmentation is faithful at z > 3.8, the same contrastive recipe should transfer to other faint high-redshift populations where visual inspection coverage is sparse, such as DESI's Ly-alpha emitter targets.
  • The learned representation could serve as a reusable prior for other LBG science beyond redrock fitting, such as Ly-alpha forest correlations or void catalogs, since the embedding appears to carry a continuous redshift gradient.
  • A direct testable extension is to retrain with separate high-redshift template realizations rather than one randomly chosen LBGme template, which may reduce the LBGme subtype confusion seen in the confusion matrix.
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 / 6 minor

Summary. The paper proposes zlbg, a supervised weighted contrastive-learning pipeline for classifying DESI Lyman Break Galaxy (LBG) spectra and estimating their redshifts. The encoder is the same backbone as QuasarNET/lbgNET, but the line-finder head is replaced by a projection head trained with a continuously weighted contrastive loss that encodes both galaxy class similarity and redshift proximity. After training, a small MLP provides per-class probabilities and a K-nearest-neighbors regressor provides redshifts, with a redshift-quality flag q_z used to form the final confidence threshold. Training data come from DESI visual-inspection (VI) campaigns, supplemented by SNR and redshift augmentations; the redshift augmentation splices LBG templates into observed spectra to extend coverage to z=4.5. On a held-out test set of 781 spectra, zlbg achieves contaminant-selection AUC 0.997 versus 0.988 for lbgNET and redshift AUC 0.790 versus 0.799 under the fiducial ℓ_z=0.025(1+z) criterion. Appendices explore kernel choices, intra-class weighting, prior-width sensitivity, and treatment of bad spectra.

Significance. If the results hold, the paper offers a genuinely useful alternative to lbgNET for DESI Run 2 LBG processing: stronger contaminant rejection at comparable redshift completeness would improve the purity of the LBG sample without sacrificing redshift yield. The work has clear strengths: the code is publicly released, lbgNET is retrained on the same training set for a fair architecture comparison, the paper performs controlled-seed runs, and the appendices document sensitivity to kernel shape, class weights, prior width, and bad spectra. The main caveat is that the redshift-comparability claim is not yet established independently of the template-based redshift augmentation, and the test set is small enough that statistical uncertainties on the headline AUC differences should be quantified.

major comments (4)
  1. [Section 5.3, Figure 8, and Section 4.3] The headline result of "comparable redshift identification" is load-bearing on the redshift augmentation described in Section 4.3. In Figure 8, the 'Base (no z-aug)' configuration gives zlbg a redshift AUC of 0.737 versus 0.789 for lbgNET, while the fiducial 'Base' configuration gives 0.790 versus 0.799. Since the templates used for augmentation are built from VI stacks plus lbgNET classifications at tau=0.99 (Section 4.1), and since the VI sample has sparse coverage at z>3.8, the high-redshift portion of the augmented training set is largely template rather than observed data. The reported test AUCs therefore do not independently validate the z~3.8-4.5 regime that DESI Run 2 requires. Please add an independent high-z validation set (for example, published LBG spectra or DESI pilot data not used in template construction) or report redshift-binned purity/efficiency and redshift accuracy specifically for z>3.8.
  2. [Section 3.3, Section 5.2, and Appendix C] The redshift kernel width sigma_z=0.025 is set equal to the evaluation success criterion ell_z=0.025(1+z), and the same ell_z enters the quality flag q_z in Equation (3.8). This couples the training objective, the confidence threshold, and the success metric to the same scale. It is not full circularity because evaluation is on held-out spectra, but it weakens the force of the "comparable redshift" claim: the comparison is made under a metric matched to the training kernel. Please report a decoupled evaluation (for example, KNN point-estimate nMAD and 3nMAD/5nMAD outlier fractions as in Figure 13 but applied directly to the KNN redshifts) and, ideally, retrain with several sigma_z values from Figure 10 and evaluate at a fixed, independently chosen ell_z.
  3. [Section 5.1, Section 5.2, and Figure 8] The test set is small (781 spectra, including only 72 ELGs and 63 QSOs), and the reported AUC differences are not accompanied by confidence intervals. The text states that seed-to-seed variation is about 0.2% for contaminant selection and 2% for redshift performance, but this does not characterize the sampling uncertainty of the test set itself. Please provide bootstrap or DeLong confidence intervals for the headline AUC values (0.997 vs 0.988 for classification, 0.790 vs 0.799 for redshift) and for the four configurations in Figure 8, so the reader can judge whether the improvements and the augmentation dependence are statistically significant.
  4. [Section 5.3, Figure 8] The conclusion that zlbg requires redshift augmentation to match lbgNET is not consistent across training configurations: in 'Base+Val (no z-aug)' zlbg actually exceeds lbgNET (0.778 vs 0.759), whereas in 'Base (no z-aug)' it is lower (0.737 vs 0.789). This non-monotonic pattern suggests the comparison is noisy with a single seed and a small test set. Please report multiple seeds and error bars before drawing a firm conclusion about which training configurations are required for parity.
minor comments (6)
  1. [Figure 2 and Section 3.1] The label 'Flatter' in Figure 2 should read 'Flatten', and the text uses 'Replayer' where 'representation layer' or 'Rep layer' would be clearer for readers outside the DESI pipeline.
  2. [Section 2] There are several typographical errors, including 'spectras', 'redshiftidentification', and 'redshiftpredictedredshift'; these should be corrected in a revision.
  3. [Figure 3 and Section 4.3] The caption of Figure 3 says that n_aug=5 is 'the ratio between the number of redshift augmentations and SNR augmentations,' but Section 4.3 defines n_aug=5 as the total number of spectra including the original spectrum; please reconcile these statements.
  4. [Section 5.4] The Pearson correlation coefficients restricted to QSO and ELG subsets are computed over 63 and 72 objects, respectively; the low r values may reflect small-sample noise rather than genuine information complementarity, so the joint-information claim should be phrased more cautiously.
  5. [Section 3.4] The sentence 'While the MLP performs well on contaminants, it performs poorly at identifying the correct subtype of LBG' is followed by 'the MLP offers sufficient classification performance'; please rephrase to avoid the apparent contradiction.
  6. [Appendix C] In the sentence 'redrock is ran on the co-added spectra,' the grammar should be corrected, and it would be helpful to state explicitly that the same redrock templates from Section 4.1 are used for both pipelines so that template choice cannot bias the comparison.

Circularity Check

1 steps flagged · score 4.0 of 10

Redshift comparison is partially self-referential: the training kernel width equals the success tolerance, while the classification claim is independent.

  1. self definitional [Section 3.3 (Eq. 3.4) and Section 5 (redshift purity/efficiency definition)]
    "Here, σz or ∆z are chosen as arbitrary hyperparameters that match with the chosen "refinement" prior ℓz = 0.025(1+zpred) size for redrock ... The fiducial choice presented in Section 5 is the gaussian kernel with σz = 0.025, matching well with the amplitude of ℓz. ... purity is the ratio of true LBGs selected by the network ... that respect |ztrue −z pred| ≤ℓz = 0.025(1+z pred)."

    The redshift relationship weight in the training loss (Eq. 3.4) is a Gaussian whose width σz = 0.025 is explicitly chosen to match the redshift-dependent success tolerance ℓz = 0.025(1+z) that later defines redshift purity, efficiency, and AUC. The network is therefore trained to keep same-redshift pairs within the same relative window that defines a 'correct' redshift identification. The reported redshift AUC (0.790 vs 0.799) is partly a measure of the alignment between the training kernel and the evaluation criterion, not an independent test of redshift accuracy at a scale fixed a posteriori. This coupling does not affect the contaminant-classification claim, which is evaluated with s_zlbg and does not use ℓ_z.

full rationale

The only defensible circular step is the explicit matching of the redshift kernel width σz (Section 3.3, Eq. 3.4) to the ℓz = 0.025(1+z) tolerance that defines redshift success in Section 5. This makes the 'comparable redshift identification' claim partly built into the training objective rather than an independent discovery. It is not a full reduction: redshifts are evaluated on held-out VI spectra, the KNN and threshold selection add nontrivial machinery, and Appendix A shows performance is not sharply peaked at σz = 0.025. The classification result (AUC 0.997 vs 0.988) is independent of this coupling because it uses s_zlbg rather than τ_zlbg and does not involve ℓ_z. The template-based redshift augmentation (Section 4.3) is a genuine validation gap for z ≳ 3.8: the templates are built from VI stacks plus lbgNET high-confidence classifications, and the test set has sparse coverage there, so the high-redshift regime is not independently confirmed. That is a limitation of evidence, not a circular derivation. No load-bearing self-citation or imported uniqueness theorem is present: lbgNET is re-trained on the same data, and the comparison uses an external VI ground truth. Overall score 4 reflects partial circularity in the redshift metric only.

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

The central claim rests on the VI labels, the template-augmented training set, and the choice of redshift kernel. No new physical entities are introduced; the free parameters are hyperparameters jointly shaping the loss and the evaluation convention.

free parameters (7)
  • Gaussian redshift kernel width sigma_z = 0.025
    Chosen to match the redrock prior and the redshift success criterion, coupling the training objective to the evaluation metric.
  • Class similarity matrix C intra-LBG off-diagonal weight = 0.5
    Set empirically after inspecting UMAP embeddings; Appendix B shows alternative choices perform similarly, indicating hand tuning.
  • redrock prior half-width amplitude = 0.025 in units of (1+z)
    Chosen so that z=3 matches the fixed prior half-width of 0.1 used for lbgNET; also defines the redshift success criterion.
  • Contrastive temperature = 0.3
    Chosen empirically because the training data is small and noisy; it controls the sharpness of the cosine similarity distribution.
  • Number of KNN neighbors for redshift inference = 50
    Default choice for the z-KNN regressor; no sensitivity analysis is reported for this value.
  • Redshift augmentation width and count = Uniform over +/-0.8 in z, naug=5
    Arbitrarily chosen to extend redshift coverage up to z=4.5; affects the training distribution and the learned embedding.
  • Spectral binning width = 8 Angstrom
    Chosen to match lbgNET; finer binning showed worse performance in both redshift and classification.
assumptions (5)
  • domain assumption Visual inspection redshifts and class labels with mean quality above 2.5 are accurate ground truth.
    All training, validation, and test labels come from three-reviewer VI campaigns; errors in VI propagate to labels and to template construction (Section 4.2).
  • domain assumption LBG templates built from VI spectra plus lbgNET high-confidence classifications are representative of the LBG population at all redshifts used in augmentations.
    Redshift augmentations splice these templates into observed spectra to cover z up to 4.5 where VI coverage is sparse (Sections 4.1 and 4.3).
  • domain assumption ELG and QSO contaminants in the test set, resampled to roughly 15% density, represent the true contaminant population after DESI LBG target selection.
    The test set is built from VI-0, VI-1, and survey validation contaminants; if the real DESI Run 2 target density or contaminant mix differs, the purity and efficiency curves may not transfer (Section 4.4).
  • domain assumption Redshift proximity in the embedding space implies spectral similarity relevant to redshift inference.
    This is the core premise of the redshift kernel weighting and the KNN redshift estimator (Sections 3.3 and 3.5); if the kernel does not capture the relevant spectral variation, KNN redshifts fail.
  • domain assumption The QuasarNET backbone encoder is an adequate feature extractor for LBG spectra.
    Both pipelines reuse the same convolutional encoder from QuasarNET without modification; the claim is therefore scoped to this backbone (Section 3.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Lyman Break Galaxy selection and redshift measurement with supervised contrastive learning." pith.science (2026). https://pith.science/paper/NNNVGFAW

@misc{pith2026260810080,
  author       = {Pith},
  title        = {Pith review of: Lyman Break Galaxy selection and redshift measurement with supervised contrastive learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NNNVGFAW}},
  note         = {Machine review of arXiv:2608.10080}
}
abstract

Some of the next steps for high-precision cosmology lie within the high-redshift, high-density universe. Spectroscopic survey experiments such as the Dark Energy Spectroscopic Instrument (DESI)'s second phase DESI Run 2 will shift towards probing Lyman Break Galaxy (LBG) populations from z$\sim$2 to z$\sim$4.5. For this faint sample, spectroscopic redshift measurement and sample decontamination remains a challenge, even after target selection. We propose an approach based on supervised weighted contrastive learning, in order to both learn a redshift representation for spectra and decontaminate the sample from quasars and low redshift emission line galaxies. This strategy generalizes the contrastive learning loss approach with continuous relationship weights, such that the network simultaneously learns redshift and classification tasks. The model shows stronger outlier classification and comparable redshift identification performances when compared to the previous network used for DESI (a modified version of QuasarNET) on the same dataset. In particular, contrastive learning is well suited to the small, visually-inspected sample used for training and testing, especially given the multi-task nature of this work.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 34 canonical work pages

  1. [1]

    Schlegel, S

    D.J. Schlegel, S. Ferraro, G. Aldering, C. Baltay, S. BenZvi, R. Besuner et al.,A spectroscopic road map for cosmic frontier: Desi, desi-ii, stage-5, 2022

  2. [2]

    Besuner, A

    R. Besuner, A. Dey, A. Drlica-Wagner, H. Ebina, G.F. Moroni, S. Ferraro et al.,The spectroscopic stage-5 experiment, 2025

  3. [3]

    Schlegel, J.A

    D.J. Schlegel, J.A. Kollmeier, G. Aldering, S. Bailey, C. Baltay, C. Bebek et al.,The MegaMapper: A Stage-5 Spectroscopic Instrument Concept for the Study of Inflation and Dark Energy,arXiv e-prints(2022) arXiv:2209.04322 [2209.04322]

  4. [4]

    Takada, R.S

    M. Takada, R.S. Ellis, M. Chiba, J.E. Greene, H. Aihara, N. Arimoto et al.,Extragalactic science, cosmology, and Galactic archaeology with the Subaru Prime Focus Spectrograph, PASJ 66(2014) R1 [1206.0737]

  5. [5]

    Tamura, N

    N. Tamura, N. Takato, A. Shimono, Y. Moritani, K. Yabe, Y. Ishizuka et al.,Prime focus spectrograph (pfs) for the subaru telescope: overview, recent progress, and future perspectives, in Ground-based and Airborne Instrumentation for Astronomy VI, C.J. Evans, L. Simard and H. Takami, eds., vol. 9908, p. 99081M, SPIE, Aug., 2016, DOI

  6. [6]

    Bacon, V

    R. Bacon, V. Maineiri, S. Randich, A. Cimatti, J.-P. Kneib, J. Brinchmann et al.,Wst – widefield spectroscopic telescope: Motivation, science drivers and top-level requirements for a new dedicated facility, 2024

  7. [7]

    T.M.S. Team, C. Babusiaux, M. Bergemann, A. Burgasser, S. Ellison, D. Haggard et al.,The detailed science case for the maunakea spectroscopic explorer, 2019 edition, 2019

  8. [8]

    de Jong, O

    R.S. de Jong, O. Agertz, A.A. Berbel, J. Aird, D.A. Alexander, A. Amarsi et al.,4MOST: Project overview and information for the First Call for Proposals,The Messenger175(2019) 3 [1903.02464]

Show all 68 references
  1. [9]

    Steidel, K.L

    C.C. Steidel, K.L. Adelberger, M. Dickinson, M. Giavalisco and M. Pettini,Lyman Break Galaxies at z~3 and Beyond,arXiv e-prints(1998) astro [astro-ph/9812167]

  2. [10]

    Giavalisco,Lyman-Break Galaxies, ARA&A40(2002) 579

    M. Giavalisco,Lyman-Break Galaxies, ARA&A40(2002) 579

  3. [11]

    Shapley, C.C

    A.E. Shapley, C.C. Steidel, M. Pettini and K.L. Adelberger,Rest-frame ultraviolet spectra of z 3 lyman break galaxies,The Astrophysical Journal588(2003) 65–89. – 32 –

  4. [12]

    C. Ly, M.A. Malkan, T. Treu, J.-H. Woo, T. Currie, M. Hayashi et al.,Lyman break galaxies at z∼1.8-2.8:galex/nuv imaging of the subaru deep field,The Astrophysical Journal697(2009) 1410–1432

  5. [13]

    Shapley,Physical properties of galaxies fromz= 2–4,Annual Review of Astronomy and Astrophysics49(2011) 525–580

    A.E. Shapley,Physical properties of galaxies fromz= 2–4,Annual Review of Astronomy and Astrophysics49(2011) 525–580

  6. [14]

    Förster Schreiber and S

    N.M. Förster Schreiber and S. Wuyts,Star-forming galaxies at cosmic noon,Annual Review of Astronomy and Astrophysics58(2020) 661–725

  7. [15]

    Payerne, W

    C. Payerne, W. d’Assignies, C. Yèche, H. Hildebrandt, D. Lang, T. de Boer et al.,Forecasting local primordial non-Gaussianities from UNIONS Lyman-break galaxies and Planck CMB lensing, A&A709(2026) A70 [2511.22243]

  8. [16]

    Chaussidon, C

    E. Chaussidon, C. Yèche, N. Palanque-Delabrouille, D.M. Alexander, J. Yang, S. Ahlen et al., Target Selection and Validation of DESI Quasars, ApJ944(2023) 107 [2208.08511]

  9. [17]

    Chaussidon, C

    E. Chaussidon, C. Yèche, A. de Mattia, C. Payerne, P. McDonald, A.J. Ross et al., Constraining primordial non-Gaussianity with DESI 2024 LRG and QSO samples,arXiv e-prints(2024) arXiv:2411.17623 [2411.17623]

  10. [18]

    Herrera-Alcantar, E

    H.K. Herrera-Alcantar, E. Armengaud, C. Yèche, C. Gordon, L. Casas, A. Font-Ribera et al., The lyman-αforest from lbgs: First 3d correlation measurement with desi and prospects for cosmology,Journal of Cosmology and Astroparticle Physics2025(2025) 053

  11. [19]

    Contarini, G

    S. Contarini, G. Verza and A. Pisani,The era of precision cosmology with voids, A&A Rev.34 (2026) 1 [2601.14362]

  12. [20]

    Chung, M

    A.S. Chung, M. Dijkstra, B. Ciardi, K. Kakiichi and T. Naab,The circumgalactic medium in lymanα: a new constraint on galactic outflow models,Monthly Notices of the Royal Astronomical Society484(2019) 2420–2432

  13. [21]

    Steidel, M

    C.C. Steidel, M. Bogosavljević, A.E. Shapley, N.A. Reddy, G.C. Rudie, M. Pettini et al.,The keck lyman continuum spectroscopic survey (klcs): The emergent ionizing spectrum of galaxies at z∼3⋆,The Astrophysical Journal869(2018) 123

  14. [22]

    Abareshi, J

    DESI Collaboration, B. Abareshi, J. Aguilar, S. Ahlen, S. Alam, D.M. Alexander et al., Overview of the Instrumentation for the Dark Energy Spectroscopic Instrument, AJ164(2022) 207 [2205.10939]

  15. [23]

    Aghamousa, J

    DESI Collaboration, A. Aghamousa, J. Aguilar, S. Ahlen, S. Alam, L.E. Allen et al.,The DESI Experiment Part II: Instrument Design,arXiv e-prints(2016) arXiv:1611.00037 [1611.00037]

  16. [24]

    Schlafly, D

    E.F. Schlafly, D. Kirkby, D.J. Schlegel, A.D. Myers, A. Raichoor, K. Dawson et al.,Survey Operations for the Dark Energy Spectroscopic Instrument, AJ166(2023) 259 [2306.06309]

  17. [25]

    Miller, P

    T.N. Miller, P. Doel, G. Gutierrez, R. Besuner, D. Brooks, G. Gallo et al.,The Optical Corrector for the Dark Energy Spectroscopic Instrument, AJ168(2024) 95 [2306.06310]

  18. [26]

    Poppett, L

    C. Poppett, L. Tyas, J. Aguilar, C. Bebek, D. Bramall, T. Claybaugh et al.,Overview of the Fiber System for the Dark Energy Spectroscopic Instrument, AJ168(2024) 245

  19. [27]

    J. Guy, S. Bailey, A. Kremin, S. Alam, D.M. Alexander, C. Allende Prieto et al.,The Spectroscopic Data Processing Pipeline for the Dark Energy Spectroscopic Instrument, AJ165 (2023) 144 [2209.14482]

  20. [28]

    Abdul-Karim, A.G

    DESI Collaboration, M. Abdul-Karim, A.G. Adame, D. Aguado, J. Aguilar, S. Ahlen et al., Data Release 1 of the Dark Energy Spectroscopic Instrument,arXiv e-prints(2025) arXiv:2503.14745 [2503.14745]

  21. [29]

    Abdul-Karim, J

    DESI Collaboration, M. Abdul-Karim, J. Aguilar, S. Ahlen, S. Alam, L. Allen et al.,DESI DR2 Results II: Measurements of Baryon Acoustic Oscillations and Cosmological Constraints, arXiv e-prints(2025) arXiv:2503.14738 [2503.14738]. – 33 –

  22. [30]

    Adame, J

    DESI Collaboration, A.G. Adame, J. Aguilar, S. Ahlen, S. Alam, D.M. Alexander et al.,DESI 2024 VII: cosmological constraints from the full-shape modeling of clustering measurements, J. Cosmology Astropart. Phys.2025(2025) 028 [2411.12022]

  23. [31]

    Ivezić, S.M

    Ž. Ivezić, S.M. Kahn, J.A. Tyson, B. Abel, E. Acosta, R. Allsman et al.,LSST: From Science Drivers to Reference Design and Anticipated Data Products, ApJ873(2019) 111 [0805.2366]

  24. [32]

    Ruhlmann-Kleider, C

    V. Ruhlmann-Kleider, C. Yèche, C. Magneville, H. Coquinot, E. Armengaud, N. Palanque-Delabrouille et al.,High redshift lbgs from deep broadband imaging for future spectroscopic surveys,Journal of Cosmology and Astroparticle Physics2024(2024) 059

  25. [33]

    Payerne, W

    C. Payerne, W. d’Assignies Doumerg, C. Yèche, V. Ruhlmann-Kleider, A. Raichoor, D. Lang et al.,Selection of high-redshift Lyman-Break Galaxies from broadband and wide photometric surveys, J. Cosmology Astropart. Phys.2025(2025) 031 [2410.08062]

  26. [34]

    Busca and C

    N. Busca and C. Balland,Quasarnet: Human-level spectral classification and redshifting with deep neural networks, 2018

  27. [35]

    Bailey et al., in preparation(2025)

  28. [36]

    Raichoor, J

    A. Raichoor, J. Moustakas, J.A. Newman, T. Karim, S. Ahlen, S. Alam et al.,Target Selection and Validation of DESI Emission Line Galaxies, AJ165(2023) 126 [2208.08513]

  29. [37]

    Green, D

    D. Green, D. Kirkby, J. Aguilar, S. Ahlen, D.M. Alexander, E. Armengaud et al.,Using active learning to improve quasar identification for the DESI spectra processing pipeline, J. Cosmology Astropart. Phys.2025(2025) 087 [2505.01596]

  30. [38]

    T. Chen, S. Kornblith, M. Norouzi and G. Hinton,A simple framework for contrastive learning of visual representations, inProceedings of the 37th International Conference on Machine Learning, ICML’20, JMLR.org, 2020

  31. [39]

    Khosla, P

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y. Tian, P. Isola et al.,Supervised contrastive learning,2004.11362

  32. [40]

    Martínez-Solaeche, R

    G. Martínez-Solaeche, R. García-Benito, R.M. González Delgado, L. Díaz-García, S.F. Sánchez, A.M. Conrado et al.,Exploring galaxy properties of ecalifa with contrastive learning,Astronomy & Astrophysics688(2024) A160

  33. [41]

    Cheng, N

    T.-Y. Cheng, N. Li, C.J. Conselice, A. Aragón-Salamanca, S. Dye and R.B. Metcalf,Identifying strong lenses with unsupervised machine learning using convolutional autoencoder,Monthly Notices of the Royal Astronomical Society494(2020) 3750–3765

  34. [42]

    Hayat, G

    M.A. Hayat, G. Stein, P. Harrington, Z. Lukić and M. Mustafa,Self-supervised representation learning for astronomical images,The Astrophysical Journal Letters911(2021) L33

  35. [43]

    Stein, J

    G. Stein, J. Blaum, P. Harrington, T. Medan and Z. Lukić,Mining for strong gravitational lenses with self-supervised learning,The Astrophysical Journal932(2022) 107

  36. [44]

    A.-B. Wang, Y. Yuan, H. Cai and X.-L. Fan,Classifying Core-Collapse Supernova Gravitational Waves using Supervised Contrastive Learning,arXiv e-prints(2026) arXiv:2601.01376 [2601.01376]

  37. [45]

    Parker, F

    L. Parker, F. Lanusse, S. Golkar, L. Sarra, M. Cranmer, A. Bietti et al.,Astroclip: a cross-modal foundation model for galaxies,Monthly Notices of the Royal Astronomical Society 531(2024) 4990–5011

  38. [46]

    Parker, F

    L. Parker, F. Lanusse, J. Shen, O. Liu, T. Hehir, L. Sarra et al.,AION-1: Omnimodal Foundation Model for Astronomical Sciences,arXiv e-prints(2025) arXiv:2510.17960 [2510.17960]

  39. [47]

    Huertas-Company, R

    M. Huertas-Company, R. Sarmiento and J. Knapen,A brief review of contrastive learning applied to astrophysics, 2023. – 34 –

  40. [48]

    Ebina, M

    H. Ebina, M. White, A. Raichoor, A. Dey, D. Schlegel, D. Lang et al.,Clustering analysis of medium-band selected high-redshift galaxies,Journal of Cosmology and Astroparticle Physics 2026(2026) 019

  41. [49]

    Ebina, M.J

    H. Ebina, M.J. White, R. Zhou, A. Dey, D. Schlegel, J.N. Aguilar et al.,The 3d clustering of lyman alpha emitters measured with desi, 2026

  42. [50]

    Gil-Marín, J.E

    H. Gil-Marín, J.E. Bautista, R. Paviot, M. Vargas-Magaña, S. de la Torre, S. Fromenteau et al.,The completed sdss-iv extended baryon oscillation spectroscopic survey: measurement of the bao and growth rate of structure of the luminous red galaxy sample from the anisotropic pow...

  43. [51]

    Alexander, T.M

    D.M. Alexander, T.M. Davis, E. Chaussidon, V.A. Fawcett, A. X. Gonzalez-Morales, T.-W. Lan et al.,The DESI Survey Validation: Results from Visual Inspection of the Quasar Survey Spectra, AJ165(2023) 124 [2208.08517]

  44. [52]

    T.-W. Lan, R. Tojeiro, E. Armengaud, J.X. Prochaska, T.M. Davis, D.M. Alexander et al.,The DESI Survey Validation: Results from Visual Inspection of Bright Galaxies, Luminous Red Galaxies, and Emission-line Galaxies, ApJ943(2023) 68 [2208.08516]

  45. [53]

    McInnes, J

    L. McInnes, J. Healy and J. Melville,Umap: Uniform manifold approximation and projection for dimension reduction,1802.03426

  46. [54]

    van der Maaten and G

    L. van der Maaten and G. Hinton,Visualizing data using t-sne,Journal of Machine Learning Research9(2008) 2579

  47. [55]

    Srinivasa, J

    R.S. Srinivasa, J. Cho, C. Yang, Y.M. Saidutta, C.-H. Lee, Y. Shen et al.,Cwcl: Cross-modal transfer with continuously weighted contrastive loss, 2023

  48. [56]

    Hendrycks and K

    D. Hendrycks and K. Gimpel,Gaussian error linear units (gelus), 2023

  49. [57]

    Chen and C

    T. Chen and C. Guestrin,Xgboost: A scalable tree boosting system, inProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’16, p. 785–794, ACM, Aug., 2016, DOI

  50. [58]

    Harris, K.J

    C.R. Harris, K.J. Millman, S.J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau et al., Array programming with NumPy,Nature585(2020) 357

  51. [59]

    Virtanen, R

    P. Virtanen, R. Gommers, T.E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau et al., SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python,Nature Methods17 (2020) 261

  52. [60]

    Hunter,Matplotlib: A 2d graphics environment,Computing in Science & Engineering9 (2007) 90

    J.D. Hunter,Matplotlib: A 2d graphics environment,Computing in Science & Engineering9 (2007) 90

  53. [61]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel et al.,Scikit-learn: Machine learning in python,Journal of Machine Learning Research12(2011) 2825

  54. [62]

    Robitaille, E.J

    Astropy Collaboration, T.P. Robitaille, E.J. Tollerud, P. Greenfield, M. Droettboom, E. Bray et al.,Astropy: A community Python package for astronomy, A&A558(2013) A33 [1307.6212]

  55. [63]

    Price-Whelan, B.M

    Astropy Collaboration, A.M. Price-Whelan, B.M. Sipőcz, H.M. Günther, P.L. Lim, S.M. Crawford et al.,The Astropy Project: Building an Open-science Project and Status of the v2.0 Core Package, AJ156(2018) 123 [1801.02634]

  56. [64]

    Price-Whelan, P.L

    Astropy Collaboration, A.M. Price-Whelan, P.L. Lim, N. Earl, N. Starkman, L. Bradley et al., The Astropy Project: Sustaining and Growing a Community-oriented Open-source Project and the Latest Major Release (v5.0) of the Core Package, ApJ935(2022) 167 [2206.14220]

  57. [65]

    56 – 61, 2010, DOI

    Wes McKinney,Data Structures for Statistical Computing in Python, inProceedings of the 9th – 35 – Python in Science Conference, Stéfan van der Walt and Jarrod Millman, eds., pp. 56 – 61, 2010, DOI

  58. [66]

    pandas development team,pandas-dev/pandas: Pandas, Feb., 2020

    T. pandas development team,pandas-dev/pandas: Pandas, Feb., 2020. 10.5281/zenodo.3509134

  59. [67]

    Abadi, A

    M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro et al.,TensorFlow: Large-scale machine learning on heterogeneous systems, 2015

  60. [68]

    Chollet et al., “Keras.”https://keras.io, 2015

    F. Chollet et al., “Keras.”https://keras.io, 2015. – 36 –

Pith tools

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