Pith. sign in

REVIEW 4 major objections 6 minor 85 references

Supervised Machine Learning Methods with Uncertainty Quantification for Exoplanet Atmospheric Retrievals from Transmission Spectroscopy

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper shows that XGBoost and SVM regressors, trained on per-spectrum normalized spectra with logarithmic abundance targets, can reproduce exoplanet atmospheric parameters from transmission spectra and recover WASP-39b abundances consis

desk verdict A useful, systematic ML-retrieval benchmark whose headline ranking is plausible but whose test error may be inflated by an ungrouped train/test split; worth a referee, but needs a grouped split and a data release first. read the letter →

arxiv 2508.04982 v1 pith:IZ3OYGWM submitted 2025-08-07 astro-ph.EP astro-ph.IMcs.LGphysics.data-an

Roy T. Forestano , Konstantin T. Matchev , Katia Matcheva , Eyup B. Unlu This is my paper
classification astro-ph.EPastro-ph.IMcs.LGphysics.data-an
keywords exoplanetatmospherestransmissionspectroscopyatmosphericretrievalmachinelearningregressionXGBoostsupportvectormachinesuncertaintyquantificationWASP-39b
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that ordinary, interpretable regression algorithms—especially extreme gradient boosting (XGB) and support vector machines (SVM)—can retrieve an exoplanet's temperature and five trace-gas abundances from a transmission spectrum, as long as the spectra are normalized per sample and the abundance targets are logarithmic. The motivation is speed: standard Bayesian retrievals take hours to days per planet, while a trained regressor predicts almost instantly. In a benchmark on 105,887 synthetic spectra built from the Ariel Big Challenge database, XGB and SVM outperform partial least squares, k-nearest neighbors, decision trees, random forests, voting, and stacking. The paper also finds that the preprocessing of the spectra can affect performance as much as the choice of algorithm. The best model, applied to the JWST spectrum of WASP-39b, predicts abundances inside the ranges reported by earlier Bayesian retrievals.

What carries the argument

The key mechanism is per-spectrum normalization: each 52-bin transit spectrum is centered by subtracting its own mean and scaled by its own standard deviation, so the model learns spectral shape rather than overall flux level. This is paired with logarithmic targets for chemical abundances, which spreads out the low-abundance end of the training distribution. The third piece is the XGB regressor, a sequential ensemble of shallow decision trees that concentrates training weight on the hardest examples. The paper's information-content analysis, using feature correlations and a transverse decomposition, explains why normalization exposes feature-target correlations that standardization leaves h

What would settle it

Train the same XGB/NL regressor on the ABC-like database but generate test spectra with the forward model including clouds or hazes, non-isothermal temperature profiles, and realistic JWST noise; if retrieval errors grow sharply for the same abundances, or if the WASP-39b predictions leave the published ranges when noise realizations are varied, the central transfer claim fails. A cheaper check: feed the trained model a chemically self-consistent atmosphere (with CO/CH4 equilibrium) and see whether the predicted abundances remain accurate.

Watch

Extended reading notes

Core claim

The central claim is that an XGB regressor trained on normalized transmission spectra with logarithmic abundance targets reliably reproduces the atmospheric parameters used to generate the spectra, and that this transfers to a real JWST spectrum. On the test set, the best models show low bias and variance across most of the parameter space, with accuracy degrading at very low abundances (a 'knee' near $10^{-6}$--$10^{-7}$) and at high temperatures, where the training set is sparse. XGB and SVM outperform PLS, KNN, DT, RF, VOTE, and STACK; per-spectrum normalization is consistently better than feature-wise standardization, while adding the spectral mean and standard deviation as extra feature

Load-bearing premise

The paper assumes that a model trained on noiseless, isothermal, cloud-free, hydrogen-helium synthetic spectra with trace abundances sampled independently and uniformly can be applied to a real, noisy, cloudy, non-isothermal JWST spectrum after simple rebinning, and that point predictions outside the training range (e.g., CO and NH3 for WASP-39b) are still physically meaningful.

Editorial extensions

If this is right

  • Trained regressors can turn a JWST/Ariel transmission spectrum into temperature and abundance estimates in milliseconds after a one-time training investment, making population-scale exoplanet surveys affordable.
  • Per-spectrum normalization plus log-abundance targets is a broadly useful preprocessing recipe for transmission-spectrum retrieval, not a quirk of one algorithm.
  • The abundance 'knee' near $10^{-6}$--$10^{-7}$ defines a practical detection floor for trace gases in this approach; predictions below it should be reported as upper limits.
  • Retrieval errors at high temperatures trace directly to sparse training coverage, so a uniformly sampled training database is needed before these models can be trusted for hot planets.
  • The binned bias-and-variance curves versus predicted values give a practical way to attach uncertainty estimates to individual real-spectrum predictions, where the true parameters are unknown.

Reading between the lines

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

  • The benchmark uses noiseless synthetic spectra; injecting realistic JWST or Ariel noise into training or test spectra is a direct extension the paper does not perform, and it would test whether the ranking of methods survives.
  • Because the database samples abundances independently, it ignores chemical networks; testing on chemically self-consistent forward models could change which correlations the regressors exploit and where the low-abundance knee sits.
  • The paper's correlation-matrix diagnostic—predictions should stay uncorrelated when true targets are uncorrelated—could be turned into an automatic model-selection screen for future retrieval surrogates.
  • Applying the same normalization-plus-log-target recipe to emission spectra or to retrievals with cloud and haze parameters is untested; those settings have stronger degeneracies, and per-spectrum normalization may behave differently.
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 benchmarks eight classical supervised regression algorithms (PLS, SVM, KNN, DT, RF, VOTE, STACK, XGB) for retrieving equilibrium temperature and five trace-gas abundances from synthetic exoplanet transmission spectra. It uses 105,887 noiseless spectra generated from 5,900 ABC/TauREx planets at 52 wavelengths and six preprocessing configurations (standardization vs per-spectrum normalization, linear vs log abundance targets, with/without spectral mean and standard deviation as extra features). Performance is assessed on 12,000 held-out spectra through scatter plots and quantile-binned error plots, along with training-time comparisons. The paper concludes that XGB and SVM trained on normalized spectra with log-abundance targets perform best, and it applies the XGB/NL model to the JWST WASP-39b spectrum, reporting abundances within ranges previously reported in the literature.

Significance. If the ranking survives a properly controlled validation, the paper is a useful, interpretable benchmark of classical ML methods against a realistic synthetic database, with concrete guidance on preprocessing and on accuracy/precision behavior across parameter space. Strengths include the large ground-truth test evaluation, explicit discussion of the low-abundance sensitivity 'knee' and of nonuniform sampling, realistic hedges on WASP-39b upper limits, and comparison with literature retrieval values. The main ranking, however, rests on an instance-level train/test split, unspecified hyperparameter selection, and visual rather than quantitative metrics, so the findings are not yet established at the level claimed. The work is potentially publishable after a corrected, more quantitative validation.

major comments (4)
  1. [§2, §3.4] The train/test split is performed at the level of individual spectra, not grouped by unique planet. §2 says 105,887 spectra come from 5,900 planets with fixed stellar/planetary parameters and equilibrium temperature, only trace abundances being resampled (~18 spectra per planet). An 80/20 split of 60,000 spectra therefore places spectra from the same planet in both training and test sets. Since T is fixed per planet and the spectral baseline (scale height, Rayleigh/CIA slope) depends on planet parameters, the regressors can memorize planet-specific baselines and inflate test accuracy, particularly for T but possibly also for abundances through feature isolation. The reported benchmark then measures interpolation over repeated configurations of 5,900 planets, not generalization to new planets. The WASP-39b case is one unseen target and cannot compensate. Please redo the evaluation with a
  2. [§4.8] Hyperparameters are listed (C=500, gamma=0.028, epsilon=0.1 for SVM; k=6 for KNN; 115 trees for RF; max_depth=6 and learning rate 0.1 for XGB, etc.) but no tuning procedure is described. The statement 'we optimized the respective hyperparameters to achieve best performance' does not say what data were used for selection. If the same test set was used to choose hyperparameters, the comparison is biased and the reported ranking could be circular. Provide the search space, selection criterion, and use a validation split or nested cross-validation independent of the test set.
  3. [§5.2–5.3] The central ranking (XGB/SVM best, RF underperforming) is supported by visual inspection of Figures 9–26 without a table of quantitative global metrics. The paper has no MAE, RMSE, R², or similar values per target and per configuration; the claim that XGB and SVM have 'low bias and variance' is not quantified. For a systematic benchmark, a summary table of metrics with uncertainties is essential to substantiate 'best performing' and to allow comparison with future work. Please add such a table, at least for the six preprocessing configurations and the top methods, ideally for all 48 experiments.
  4. [§2, §6] The training database is noiseless, cloud-free, isothermal, H/He-dominated with fixed mean molecular weight, while the WASP-39b application uses a real, noisy, cloudy, non-isothermal spectrum. Test-set accuracy is therefore an upper bound for real data, and the single-planet agreement may be fortuitous; moreover two predicted abundances (CO, NH3) lie outside the sampled training range. The paper does hedge these points, but Section 7 states the models are 'capable of reliably reproducing the planetary parameters' and 'we were able to reproduce' literature values. Please add a noise-injected and/or cloudy test set, or explicitly limit the claim to the noiseless database and describe the WASP-39b results as illustrative rather than confirmatory.
minor comments (6)
  1. [Abstract] Typo: 'validated on a the case study' should be 'validated on the case study'; also 'W ASP-39b' has an anomalous space.
  2. [§2] Typo: 'as paert of a NeurIPS 2022 competition' should be 'as part of'. Also '105,887 planet instances' and later 's=105,887 planets' in §3.1 conflict with the 5,900 unique planets; use 'spectra' or 'samples' for clarity.
  3. [§4.5] Typo: 'esemble method' should be 'ensemble method'; also 'random subset of of the features' has a duplicated 'of'.
  4. [§5.3] Typo: 'preformance' should be 'performance'. The cross-reference 'Figure 5.3 provides yet another way' is not a valid figure number; it should refer to the correlation-matrix figure (currently labeled Figure 6).
  5. [§6] Figure cross-references are inconsistent: 'Figure 6 depicts the original (blue dots) and the rebinned (red diamonds) spectra' should be Figure 7, and 'we reproduce in Figure 6' referring to the last row of Figure 12 should be Figure 8. Also 'the same exact 52 spectral wavelength values' is redundant.
  6. [§3.4 / Data Availability] The N M notation uses M_i and σ_i(M) for spectral mean and standard deviation, which can be confused with the spectrum M(λ). Consider renaming these features (e.g., μ_i, s_i). In addition, the manuscript provides no code or preprocessing scripts; only a link to the public ABC database is given, which is not sufficient to reproduce the exact 105,887-spectrum realization.

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity: the ML benchmark is evaluated against held-out simulation labels and the WASP-39b case study is an external application; only minor non-load-bearing self-citations are present.

  1. other [Section 3.3, after Eq. (13)]
    "It has been shown to be beneficial in practice during the Ariel Machine Learning Data Challenge (Yip et al. 2023; Unlu et al. 2023) and was theoretically motivated in Matchev et al. (2022b)."

    This is a self-citation: the normalized-feature preprocessing that feeds the best-performing models is motivated by prior work by the same authors (Unlu et al. 2023 includes Forestano, Matchev, and Matcheva; Matchev et al. 2022b is by Matchev and Matcheva). However, the citation is not load-bearing: the paper independently benchmarks normalized versus standardized features on held-out TauREx spectra with known labels, so the ranking does not reduce to the cited prior work. Scored as a minor self-citation, not as a circular derivation.

full rationale

The central claim is a supervised regression benchmark: 52-bin TauREx spectra are the features and the generating atmospheric parameters are the targets, with 48,000 training and 12,000 test samples. Since test predictions are compared with known simulation truths, the reported XGB/SVM ranking is measured, not constructed by definition. The normalization and log-target preprocessing schemes are defined in the paper and compared across 48 configurations; the favorable result for normalization is empirically demonstrated, even though the choice is also motivated by self-citations. The WASP-39b case study is an out-of-sample application to real JWST data, and the literature values in Table 2 are not used as training labels, so the agreement is external validation rather than circular. The principal correctness caveat is that Section 3.4 describes a 'random subset of n=60,000 samples' split without grouping by the 5,900 unique planets from which the 105,887 spectra were generated; an instance-level split could permit the same planet's spectra in both train and test, inflating generalization estimates. That is a data-leakage/generalization risk, not a definitional equivalence or fitted-input-called-prediction circularity. Overall, the derivation is self-contained against external benchmarks, with only minor non-load-bearing self-citations, hence score 2.

Assumptions & free parameters 1 free parameters · 7 assumptions · 0 invented entities

The paper contributes an empirical benchmark, not a derivation. Its central claims rest on the domain assumptions inherited from TauREx and the ABC database (noiseless, isothermal, cloud-free, independently-sampled abundances) plus hand-tuned hyperparameters. No new physical entities are introduced.

free parameters (1)
  • Model hyperparameters (e.g., SVM C=500, gamma=0.028, epsilon=0.1; KNN k=6; RF n_estimators=115; XGB max_depth=6, learnin = Values listed in Section 4.8; stated as 'optimized to achieve best performance'
    The benchmark ranking (XGB/SVM best) depends on these choices; the tuning procedure is not described.
assumptions (7)
  • domain assumption TauREx3 forward model correctly simulates transmission spectra of hot Jupiters
    Invoked in Section 2 to generate all training/testing spectra.
  • domain assumption Isothermal, 100-layer atmosphere with 1 Pa to 1e6 Pa pressure grid, He/H2=0.17, no clouds or hazes
    Section 2; the synthetic database restricts the retrieval problem to this simple physics.
  • domain assumption Trace-gas abundances are independently sampled uniformly in log-space over fixed ranges (H2O 1e-9..1e-3, CO2 1e-9..1e-4, etc.)
    Section 2; no chemical model; results may not transfer to chemically correlated atmospheres.
  • domain assumption Training on noiseless spectra is sufficient for models to perform on noisy real observations
    Section 2 states spectra have no instrument noise; Section 6 applies the trained model to noisy JWST data.
  • domain assumption The test distribution is representative; non-uniform planet/star sampling in the ABC database does not invalidate accuracy claims over the dynamical range
    Sections 2 and 7 acknowledge non-uniform distributions and training biases.
  • domain assumption The rebinned WASP-39b spectrum is compatible with the 52-bin synthetic training features after normalization
    Section 6; relies on TauREx FluxBinner and the same normalization transform.
  • standard math scikit-learn and XGBoost implementations are correct
    Section 4; standard library functions trusted for all regressors.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Supervised Machine Learning Methods with Uncertainty Quantification for Exoplanet Atmospheric Retrievals from Transmission Spectroscopy." pith.science (2026). https://pith.science/paper/IZ3OYGWM

@misc{pith2026250804982,
  author       = {Pith},
  title        = {Pith review of: Supervised Machine Learning Methods with Uncertainty Quantification for Exoplanet Atmospheric Retrievals from Transmission Spectroscopy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IZ3OYGWM}},
  note         = {Machine review of arXiv:2508.04982}
}
read the original abstract

Standard Bayesian retrievals for exoplanet atmospheric parameters from transmission spectroscopy, while well understood and widely used, are generally computationally expensive. In the era of the JWST and other upcoming observatories, machine learning approaches have emerged as viable alternatives that are both efficient and robust. In this paper we present a systematic study of several existing machine learning regression techniques and compare their performance for retrieving exoplanet atmospheric parameters from transmission spectra. We benchmark the performance of the different algorithms on the accuracy, precision, and speed. The regression methods tested here include partial least squares (PLS), support vector machines (SVM), k nearest neighbors (KNN), decision trees (DT), random forests (RF), voting (VOTE), stacking (STACK), and extreme gradient boosting (XGB). We also investigate the impact of different preprocessing methods of the training data on the model performance. We quantify the model uncertainties across the entire dynamical range of planetary parameters. The best performing combination of ML model and preprocessing scheme is validated on a the case study of JWST observation of WASP-39b.

Figures

Figures reproduced from arXiv: 2508.04982 by the authors.

Figure 1
Figure 1. Sample transmission spectrum from the database used in this work. M is the observed modulation of the stellar flux as defined in eq. (1). preliminary Target List selected from the official TESS candidate list (Edwards et al. 2019a; Edwards & Tinetti 2022). While each planet retains its unique stellar and planetary parameters, e.g. radius, mass, distance, etc., the planet atmospheric chemical composition was randomly… view at source ↗
Figure 2
Figure 2. Distributions of selected stellar and planetary physical parameters over the ABC dataset. As labeled, these include the star distance, star mass, star radius, star temperature, planet mass, planet orbital period, planet distance, planet radius, and planet surface gravity [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Distributions of the target parameters, including the temperature T and log-mixing ratios, log(Xi), for i ∈ {H2O, CO2, CH4, CO, NH3} [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (23 more)
Figure 4
Figure 4. Figure 4: Standardization and normalization effects on the transit spectra. Top left: five sample spectra from the database. Top right: spectra standardized according to eq. (7). Bottom left: subtracting the mean of spectrum has the effect of centering all of the original spectr…
Figure 5
Figure 5. Figure 5: The correlation matrices among the original (top left), standardized (top right), centered by the sample means (bottom left), and normalized (bottom right) features. the variations among constituent variables. In the normalization case, the cross-correlations among the…
Figure 6
Figure 6. Figure 6: Correlation matrices of the predicted and true target values for a few selected pairings of regressors and preprocessing techniques. For reference, the baseline calculation of the true target correlations is provided in the top left plot. discussed, the predicted value…
Figure 7
Figure 7. Figure 7: The original observed JWST and rebinned WASP-39b spectra. N L dataset of normalized spectra and log chemical abundances, on WASP-39b spectral data from the James Webb Space Telescope (JWST) (Rustamkulov et al. 2023; Powell et al. 2024). The two original spectra were co…
Figure 8
Figure 8. Figure 8: XGB (N L) model prediction for WASP 39b put in the context of the model performance on the test data set (last row in [PITH_FULL_IMAGE:figures/full_fig_p026_8.png]
Figure 9
Figure 9. Figure 9: (S) Scatter plots of the model predictions for the 6 target variables from eq. (4) (y-axis) versus the true values yt of the target variables (x-axis). The model is trained and tested on the standardized spectral data {S[M], S[y]} (see Section 3.4). The colorbars indic…
Figure 10
Figure 10. Figure 10: (SL) The same as [PITH_FULL_IMAGE:figures/full_fig_p035_10.png]
Figure 11
Figure 11. Figure 11: (N ) The same as [PITH_FULL_IMAGE:figures/full_fig_p036_11.png]
Figure 12
Figure 12. Figure 12: (N L) The same as [PITH_FULL_IMAGE:figures/full_fig_p037_12.png]
Figure 13
Figure 13. Figure 13: (NM) The same as [PITH_FULL_IMAGE:figures/full_fig_p038_13.png]
Figure 14
Figure 14. Figure 14: (NML) The same as [PITH_FULL_IMAGE:figures/full_fig_p039_14.png]
Figure 15
Figure 15. Figure 15: (S) Plots of the average absolute error |yt − yp|, versus the average true value ¯yt per yt quantile bin. The model is trained and tested on the standardized spectral data {S[M], S[y]} (see Section 3.4). The error bars indicate the corresponding standard deviations of…
Figure 16
Figure 16. Figure 16: (SL) The same as [PITH_FULL_IMAGE:figures/full_fig_p041_16.png]
Figure 17
Figure 17. Figure 17: (N ) The same as [PITH_FULL_IMAGE:figures/full_fig_p042_17.png]
Figure 18
Figure 18. Figure 18: (N L) The same as [PITH_FULL_IMAGE:figures/full_fig_p043_18.png]
Figure 19
Figure 19. Figure 19: (NM) The same as [PITH_FULL_IMAGE:figures/full_fig_p044_19.png]
Figure 20
Figure 20. Figure 20: (NML) The same as [PITH_FULL_IMAGE:figures/full_fig_p045_20.png]
Figure 21
Figure 21. Figure 21: (S) Plots of the average predicted value ¯yp versus the average absolute deviation |yt −yp| of the target prediction yp from the true value yt . The model is trained and tested on the standardized spectral data {S[M], S[y]} (see Section 3.4). The quantile bins were fo…
Figure 22
Figure 22. Figure 22: (SL) The same as [PITH_FULL_IMAGE:figures/full_fig_p047_22.png]
Figure 23
Figure 23. Figure 23: (N ) The same as [PITH_FULL_IMAGE:figures/full_fig_p048_23.png]
Figure 24
Figure 24. Figure 24: (N L) The same as [PITH_FULL_IMAGE:figures/full_fig_p049_24.png]
Figure 25
Figure 25. Figure 25: (NM) The same as [PITH_FULL_IMAGE:figures/full_fig_p050_25.png]
Figure 26
Figure 26. Figure 26: (NML) The same as [PITH_FULL_IMAGE:figures/full_fig_p051_26.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

85 extracted references · 24 canonical work pages

  1. [1]

    B., Mansfield, M., et al

    Ahrer, E.-M., Stevenson, K. B., Mansfield, M., et al. 2023, Nature, 614, 653, 10.1038/s41586-022-05590-4

    doi:10.1038/s41586-022-05590-4 2023

  2. [2]

    F., Changeat , Q., Waldmann , I

    Al-Refaie , A. F., Changeat , Q., Waldmann , I. P., & Tinetti , G. 2021, , 917, 37, 10.3847/1538-4357/ac0252

    doi:10.3847/1538-4357/ac0252 2021

  3. [3]

    R., Alam, M

    Alderson, L., Wakeford, H. R., Alam, M. K., et al. 2023, Nature, 614, 664, 10.1038/s41586-022-05591-3

    doi:10.1038/s41586-022-05591-3 2023

  4. [4]

    Ard \'e vol Mart \' nez , F., Min , M., Huppenkothen , D., Kamp , I., & Palmer , P. I. 2024, , 681, L14, 10.1051/0004-6361/202348367

    work page doi:10.1051/0004-6361/202348367 2024

  5. [5]

    Ardevol Martinez , F., Min , M., Kamp , I., & Palmer , P. I. 2022, arXiv e-prints, arXiv:2203.01236. 2203.01236

    work page Pith review arXiv 2022

  6. [6]

    2023, arXiv e-prints, arXiv:2309.09337, 10.48550/arXiv.2309.09337

    Aubin , M., Cuesta-Lazaro , C., Tregidga , E., et al. 2023, arXiv e-prints, arXiv:2309.09337, 10.48550/arXiv.2309.09337

    arXiv doi:10.48550/arxiv.2309.09337 2023

  7. [7]

    2022, Nature Communications, 13, 2453, 10.1038/s41467-022-29939-5

    Batzner, S., Musaelian, A., Sun, L., et al. 2022, Nature Communications, 13, 2453, 10.1038/s41467-022-29939-5

    doi:10.1038/s41467-022-29939-5 2022

  8. [8]

    2012, The Astrophysical Journal, 753, 100, 10.1088/0004-637X/753/2/100

    Benneke, B., & Seager, S. 2012, The Astrophysical Journal, 753, 100, 10.1088/0004-637X/753/2/100

    doi:10.1088/0004-637x/753/2/100 2012

Show all 85 references
  1. [9]

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

    2006

  2. [10]

    E., et al

    Blecic , J., Harrington , J., Cubillos , P. E., et al. 2021, arXiv e-prints, arXiv:2104.12525. 2104.12525

    2021 arXiv

  3. [11]

    A., & J., S

    Breiman, L., Friedman, J., Olshen, R. A., & J., S. C. 1984, Classification and Regression Trees (1st ed.). (Chapman and Hall/CRC.), https://doi.org/10.1201/9781315139470

    1984 doi

  4. [12]

    V., et al

    Changeat , Q., Al-Refaie , A., Mugnai , L. V., et al. 2020, , 160, 80, 10.3847/1538-3881/ab9a53

    2020 doi

  5. [13]

    Changeat , Q., & Yip , K. H. 2023, RAS Techniques and Instruments, 2, 45, 10.1093/rasti/rzad001

    2023 doi

  6. [14]

    M., Latham , D

    Charbonneau , D., Brown , T. M., Latham , D. W., & Mayor , M. 2000, , 529, L45, 10.1086/312457

    2000 doi

  7. [15]

    2006, IEEE Transactions on Neural Networks, 17, 893, 10.1109/TNN.2006.875973

    Chen, P.-H., Fan, R.-E., & Lin, C.-J. 2006, IEEE Transactions on Neural Networks, 17, 893, 10.1109/TNN.2006.875973

    2006

  8. [16]

    Chen, T., & Guestrin, C. 2016, in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD '16 (New York, NY, USA: Association for Computing Machinery), 785–794, 10.1145/2939672.2939785

    2016

  9. [17]

    D., Himes , M

    Cobb , A. D., Himes , M. D., Soboczenski , F., et al. 2019, , 158, 33, 10.3847/1538-3881/ab2390

    2019 doi

  10. [18]

    2023, The Astrophysical Journal Letters, 943, L10, 10.3847/2041-8213/acaead

    Constantinou, S., Madhusudhan, N., & Gandhi, S. 2023, The Astrophysical Journal Letters, 943, L10, 10.3847/2041-8213/acaead

    2023 doi

  11. [19]

    2013, The Astrophysical Journal, 768, 42, 10.1088/0004-637X/768/1/42

    Cubillos, P., Harrington, J., Madhusudhan, N., et al. 2013, The Astrophysical Journal, 768, 42, 10.1088/0004-637X/768/1/42

    2013 doi

  12. [20]

    E., Harrington , J., Blecic , J., et al

    Cubillos , P. E., Harrington , J., Blecic , J., et al. 2021, arXiv e-prints, arXiv:2104.12524. 2104.12524

    2021 arXiv

  13. [21]

    Dietterich, T. G. 1998, Neural Computation, 10, 1895, 10.1162/089976698300017197

    1998 doi

  14. [22]

    2019 a , , 157, 242, 10.3847/1538-3881/ab1cb9

    Edwards , B., Mugnai , L., Tinetti , G., Pascale , E., & Sarkar , S. 2019 a , , 157, 242, 10.3847/1538-3881/ab1cb9

    2019 doi

  15. [23]

    2022, , 164, 15, 10.3847/1538-3881/ac6bf9

    Edwards , B., & Tinetti , G. 2022, , 164, 15, 10.3847/1538-3881/ac6bf9

    2022 doi

  16. [24]

    2019 b , Experimental Astronomy, 47, 29, 10.1007/s10686-018-9611-4

    Edwards , B., Rice , M., Zingales , T., et al. 2019 b , Experimental Astronomy, 47, 29, 10.1007/s10686-018-9611-4

    2019 doi

  17. [25]

    2022, , 658, A41, 10.1051/0004-6361/202141940

    Falco , A., Zingales , T., Pluriel , W., & Leconte , J. 2022, , 658, A41, 10.1051/0004-6361/202141940

    2022 doi

  18. [26]

    Fan, R.-E., Chen, P.-H., & Lin, C.-J. 2005, J. Mach. Learn. Res., 6, 1889–1918

    2005

  19. [27]

    J., Kitzmann , D., et al

    Fisher , C., Hoeijmakers , H. J., Kitzmann , D., et al. 2020, , 159, 192, 10.3847/1538-3881/ab7a92

    2020 doi

  20. [28]

    T., Matchev, K

    Forestano, R. T., Matchev, K. T., Matcheva, K., et al. 2023 a , Machine Learning: Science and Technology, 4, 025027, 10.1088/2632-2153/acd989

    2023 doi

  21. [29]

    2023 b , Physics Letters B, 844, 10.1016/j.physletb.2023.138086

    ---. 2023 b , Physics Letters B, 844, 10.1016/j.physletb.2023.138086

    2023

  22. [30]

    2023 c , Physics Letters B, 847, 138266, https://doi.org/10.1016/j.physletb.2023.138266

    ---. 2023 c , Physics Letters B, 847, 138266, https://doi.org/10.1016/j.physletb.2023.138266

    2023

  23. [31]

    T., Matchev, K

    Forestano, R. T., Matchev, K. T., Matcheva, K., & Unlu, E. B. 2023 d , The Astrophysical Journal, 958, 106, 10.3847/1538-4357/ad0047

    2023 doi

  24. [32]

    2000, The Annals of Statistics, 28, 337 , 10.1214/aos/1016218223

    Friedman, J., Hastie, T., & Tibshirani, R. 2000, The Annals of Statistics, 28, 337 , 10.1214/aos/1016218223

    2000

  25. [33]

    2018, Monthly Notices of the Royal Astronomical Society, 474, 271–, https://doi.org/10.1093/mnras/stx2748

    Gandhi, S., & Madhusudhan, N. 2018, Monthly Notices of the Royal Astronomical Society, 474, 271–, https://doi.org/10.1093/mnras/stx2748

    2018 doi

  26. [34]

    E., Aronsson, J., Carlsson, O., et al

    Gerken, J. E., Aronsson, J., Carlsson, O., et al. 2023, Artificial Intelligence Review, 56, 14605, 10.1007/s10462-023-10502-7

    2023 doi

  27. [35]

    P., Line , M

    Greene , T. P., Line , M. R., Montero , C., et al. 2016, , 817, 17, 10.3847/0004-637X/817/1/17

    2016 doi

  28. [36]

    Griffith , C. A. 2014, Philosophical Transactions of the Royal Society of London Series A, 372, 20130086, 10.1098/rsta.2013.0086

    2014

  29. [37]

    2020, , 160, 15, 10.3847/1538-3881/ab9176

    Guzm \'a n-Mesa , A., Kitzmann , D., Fisher , C., et al. 2020, , 160, 15, 10.3847/1538-3881/ab9176

    2020 doi

  30. [38]

    2023, , 672, A180, 10.1051/0004-6361/202243230

    Haldemann , J., Ksoll , V., Walter , D., et al. 2023, , 672, A180, 10.1051/0004-6361/202243230

    2023 doi

  31. [39]

    D., Cubillos , P

    Harrington , J., Himes , M. D., Cubillos , P. E., et al. 2021, arXiv e-prints, arXiv:2104.12522. 2104.12522

    2021

  32. [40]

    2001, The Elements of Statistical Learning, Springer Series in Statistics (New York, NY, USA: Springer New York Inc.)

    Hastie, T., Tibshirani, R., & Friedman, J. 2001, The Elements of Statistical Learning, Springer Series in Statistics (New York, NY, USA: Springer New York Inc.)

    2001

  33. [41]

    Hayes , J. J. C., Kerins , E., Awiphan , S., et al. 2020, , 494, 4492, 10.1093/mnras/staa978

    2020 doi

  34. [42]

    2017, , 470, 2972, 10.1093/mnras/stx1453

    Heng , K., & Kitzmann , D. 2017, , 470, 2972, 10.1093/mnras/stx1453

    2017 doi

  35. [43]

    Hunter, J. D. 2007, Comput. Sci. Eng., 9, 90, 10.1109/MCSE.2007.55

    2007 doi

  36. [44]

    Inc., P. T. 2015, Collaborative data science, Montreal, QC: Plotly Technologies Inc. https://plot.ly

    2015

  37. [45]

    J., et al

    Kirk , J., L \'o pez-Morales , M., Wheatley , P. J., et al. 2019, , 158, 144, 10.3847/1538-3881/ab397d

    2019 doi

  38. [46]

    2020, , 890, 174, 10.3847/1538-4357/ab6d71

    Kitzmann , D., Heng , K., Oreshenko , M., et al. 2020, , 890, 174, 10.3847/1538-4357/ab6d71

    2020 doi

  39. [47]

    2016, in Positioning and Power in Academic Publishing: Players, Agents and Agendas, ed

    Kluyver, T., Ragan-Kelley, B., P \'e rez, F., et al. 2016, in Positioning and Power in Academic Publishing: Players, Agents and Agendas, ed. F. Loizides & B. Scmidt (IOS Press), 87--90. https://eprints.soton.ac.uk/403913/

    2016

  40. [48]

    Lim, L.-H., & Nelson, B. J. 2022, What is an equivariant neural network? 2205.07362

    2022 arXiv

  41. [49]

    R., Wolf, A

    Line, M. R., Wolf, A. S., Zhang, X., et al. 2013, The Astrophysical Journal, 775, 137, 10.1088/0004-637X/775/2/137

    2013 doi

  42. [50]

    2025, , 984, L32, 10.3847/2041-8213/adc7aa

    Lueber , A., Karchev , K., Fisher , C., et al. 2025, , 984, L32, 10.3847/2041-8213/adc7aa

    2025 doi

  43. [51]

    J., & Lewis , N

    MacDonald , R. J., & Lewis , N. K. 2022, , 929, 20, 10.3847/1538-4357/ac47fe

    2022 doi

  44. [52]

    2018, Atmospheric Retrieval of Exoplanets, ed

    Madhusudhan, N. 2018, Atmospheric Retrieval of Exoplanets, ed. H. J. Deeg & J. A. Belmonte (Cham: Springer International Publishing), 1--30, 10.1007/978-3-319-30648-3_104-1

    2018 doi

  45. [53]

    2009, The Astrophysical Journal, 707, 24, 10.1088/0004-637X/707/1/24

    Madhusudhan, N., & Seager, S. 2009, The Astrophysical Journal, 707, 24, 10.1088/0004-637X/707/1/24

    2009 doi

  46. [54]

    2018, Nature Astronomy, 2, 719, 10.1038/s41550-018-0504-2

    M \'a rquez-Neila , P., Fisher , C., Sznitman , R., & Heng , K. 2018, Nature Astronomy, 2, 719, 10.1038/s41550-018-0504-2

    2018 doi

  47. [55]

    T., Matcheva , K., & Roman , A

    Matchev , K. T., Matcheva , K., & Roman , A. 2022 a , The Planetary Sciences Journal, 3, 205, 10.3847/PSJ/ac880b

    2022 doi

  48. [56]

    2022 b , , 939, 95, 10.3847/1538-4357/ac82f3

    ---. 2022 b , , 939, 95, 10.3847/1538-4357/ac82f3

    2022 doi

  49. [57]

    2022 c , , 930, 33, 10.3847/1538-4357/ac610c

    ---. 2022 c , , 930, 33, 10.3847/1538-4357/ac610c

    2022 doi

  50. [58]

    V., Pascale, E., Edwards, B., Papageorgiou, A., & Sarkar, S

    Mugnai, L. V., Pascale, E., Edwards, B., Papageorgiou, A., & Sarkar, S. 2020, Experimental Astronomy, 50, 303–328, 10.1007/s10686-020-09676-7

    2020 doi

  51. [59]

    C., & Madhusudhan , N

    Nixon , M. C., & Madhusudhan , N. 2020, , 496, 269, 10.1093/mnras/staa1150

    2020 doi

  52. [60]

    L., et al

    Oreshenko, M., Lavie, B., Grimm, S. L., et al. 2017, The Astrophysical Journal Letters, 847, L3, 10.3847/2041-8213/aa8acf

    2017 doi

  53. [61]

    2020, , 159, 6, 10.3847/1538-3881/ab5955

    Oreshenko , M., Kitzmann , D., M \'a rquez-Neila , P., et al. 2020, , 159, 6, 10.3847/1538-3881/ab5955

    2020 doi

  54. [62]

    2011, Journal of Machine Learning Research, 12, 2825

    Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825

    2011

  55. [63]

    2023, Remote Sensing, 15, 635, 10.3390/rs15030635

    Pluriel , W. 2023, Remote Sensing, 15, 635, 10.3390/rs15030635

    2023 doi

  56. [64]

    2022, , 658, A42, 10.1051/0004-6361/202141943

    Pluriel , W., Leconte , J., Parmentier , V., et al. 2022, , 658, A42, 10.1051/0004-6361/202141943

    2022 doi

  57. [65]

    D., Lee, E

    Powell, D., Feinstein, A. D., Lee, E. K. H., et al. 2024, Nature, 626, 979, 10.1038/s41586-024-07040-9

    2024 doi

  58. [66]

    T., Matchev, K

    Roman, A., Forestano, R. T., Matchev, K. T., Matcheva, K., & Unlu, E. B. 2023, Symmetry, 15, 10.3390/sym15071352

    2023 doi

  59. [67]

    K., Mukherjee, S., et al

    Rustamkulov, Z., Sing, D. K., Mukherjee, S., et al. 2023, Nature, 614, 659, 10.1038/s41586-022-05677-y

    2023 doi

  60. [68]

    1994, , 212, 321, 10.1007/BF00984535

    Schneider , J. 1994, , 212, 321, 10.1007/BF00984535

    1994 doi

  61. [69]

    D., O'Beirne , M

    Soboczenski , F., Himes , M. D., O'Beirne , M. D., et al. 2018, arXiv e-prints, arXiv:1811.03390. 1811.03390

    2018 arXiv

  62. [70]

    2021, arXiv e-prints, arXiv:2104.04824, 10.48550/arXiv.2104.04824

    Tinetti , G., Eccleston , P., Haswell , C., et al. 2021, arXiv e-prints, arXiv:2104.04824, 10.48550/arXiv.2104.04824

    2021 doi

  63. [71]

    Trevor Hastie, Robert Tibshirani, J. F. 2009, The Elements of Statistical Learning (New York, New York: Springer)

    2009

  64. [72]

    P., Zingales , T., et al

    Tsiaras , A., Waldmann , I. P., Zingales , T., et al. 2018, The Astronomical Journal, 155, 156, 10.3847/1538-3881/aaaf75

    2018 doi

  65. [73]

    B., Forestano, R

    Unlu, E. B., Forestano, R. T., Matchev, K. T., & Matcheva, K. 2023, Reproducing Bayesian Posterior Distributions for Exoplanet Atmospheric Parameter Retrievals with a Machine Learning Surrogate Model. 2310.10521

    2023 arXiv

  66. [74]

    C., & Varoquaux, G

    van der Walt, S., Colbert, S. C., & Varoquaux, G. 2011, Computing in Science Engineering, 13, 22, 10.1109/MCSE.2011.37

    2011 doi

  67. [75]

    E., et al

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261, 10.1038/s41592-019-0686-2

    2020 doi

  68. [76]

    Waldmann , I. P. 2016, , 820, 107, 10.3847/0004-637X/820/2/107

    2016 doi

  69. [77]

    P., Tinetti, G., Rocchetto, M., et al

    Waldmann, I. P., Tinetti, G., Rocchetto, M., et al. 2015, The Astrophysical Journal, 802, 107, 10.1088/0004-637X/802/2/107

    2015 doi

  70. [78]

    2000, Technical report

    Wegelin, J. 2000, Technical report

    2000

  71. [79]

    2021, , 913, 114, 10.3847/1538-4357/abee94

    Welbanks , L., & Madhusudhan , N. 2021, , 913, 114, 10.3847/1538-4357/abee94

    2021 doi

  72. [80]

    2022, , 933, 79, 10.3847/1538-4357/ac6df1

    ---. 2022, , 933, 79, 10.3847/1538-4357/ac6df1

    2022 doi

  73. [81]

    H., Changeat , Q., Al-Refaie , A., & Waldmann , I

    Yip , K. H., Changeat , Q., Al-Refaie , A., & Waldmann , I. P. 2024, , 961, 30, 10.3847/1538-4357/ad063f

    2024 doi

  74. [82]

    H., Changeat, Q., Nikolaou, N., et al

    Yip, K. H., Changeat, Q., Nikolaou, N., et al. 2021, The Astronomical Journal, 162, 195, 10.3847/1538-3881/ac1744

    2021 doi

  75. [83]

    H., Waldmann , I

    Yip , K. H., Waldmann , I. P., Changeat , Q., et al. 2022, arXiv e-prints, arXiv:2206.14642. 2206.14642

    2022 arXiv

  76. [84]

    H., Changeat, Q., Waldmann, I., et al

    Yip, K. H., Changeat, Q., Waldmann, I., et al. 2023, Proceedings of Machine Learning Research, 220, 1. https://proceedings.mlr.press/v220/yip23a.html

    2023

  77. [85]

    Zingales , T., & Waldmann , I. P. 2018, , 156, 268, 10.3847/1538-3881/aae77c

    2018 doi

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.