Pith. sign in

REVIEW 3 major objections 6 minor 40 references

Reconstructing Cosmic History with Machine Learning: A Study Using CART, MLPR, and SVR

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Support vector regression reconstructs the cosmic age–redshift relation from simulated galaxy ages more accurately than two competing regressors, recovering standard cosmological parameters.

desk verdict A transparent ML benchmark on simulated galaxy ages, undermined by a circular cosmological claim that recovers the input model as a 'prediction.' read the letter →

arxiv 2505.17205 v1 pith:A5L3RQZN submitted 2025-05-22 astro-ph.CO gr-qc

classification astro-ph.COgr-qc
keywords galaxyagescosmicchronometersmachinelearningsupportvectorregressiondecisiontreemulti-layerperceptrondarkenergyequationofstateflatLambdaCDM
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

Can machine learning reconstruct the expansion history of the Universe from the ages of old, passively evolving galaxies? This paper argues yes, with one method doing so more accurately than two others. The authors generate simulated galaxy-age catalogs of five sizes by scattering 32 measured ages around a flat $\Lambda$CDM model, train three supervised regressors on them, and find that support vector regression (SVR) gives the lowest reconstruction error and the most stable bias–variance balance. The best SVR reconstruction recovers $\Omega_m = 0.329 \pm 0.010$ and a dark-energy equation-of-state parameter $w = -1.054 \pm 0.087$, consistent with standard cosmological values, and the 600-point predicted sample matches the parameter constraints of the full 2004-point sample. If the claim holds, it means future galaxy-age surveys of a few thousand objects could constrain dark energy without committing to a specific expansion history at the reconstruction step.

What carries the argument

The key machinery is a supervised regression pipeline on the age–redshift relation $t(z)$. Measured ages of 32 passively evolving galaxies are scattered around a flat $\Lambda$CDM fiducial model with Gaussian 10% errors via Monte Carlo sampling, producing training sets of 100 to 6680 points. Three regressors — Classification and Regression Trees, a multilayer perceptron regressor, and support vector regression — are trained on 70% of each sample and tested on the remaining 30%, and the reconstructed ages are fed to a $\chi^2$ likelihood with Markov chain Monte Carlo sampling to estimate $\Omega_m$ and $w$ in a flat wCDM model. The bias–variance decomposition of the prediction error is the diagnostic that identifies SVR as the best-balanced reconstruction.

What would settle it

Generate simulated galaxy ages from a non-$\Lambda$CDM fiducial model, for instance one with $w = -0.8$, run the same SVR pipeline, and check whether the recovered $\Omega_m$ and $w$ match that fiducial within $1\sigma$. If the recovered parameters instead drift toward flat $\Lambda$CDM values, the reconstruction is not model-independent and the paper's accuracy claim would not carry over to unknown cosmologies. Alternatively, apply the trained SVR to the original 32 observed ages and compare the recovered parameters with independent cosmic-microwave-background constraints; disagreement beyond $2\sigma$ would falsify the claim that this method recovers true cosmic history.

Watch

Extended reading notes

Core claim

The paper's central claim is that SVR is the most accurate of the three supervised methods for reconstructing the cosmic age–redshift relation from simulated galaxy ages, and that the resulting reconstruction yields cosmological parameters consistent with the fiducial model. On the 2000-point simulated sample (600 test points), SVR returns $\Omega_m = 0.329 \pm 0.010$ and $w = -1.054 \pm 0.087$; CART and MLPR show mean squared errors roughly ten times larger, and SVR has the lowest bias–variance decomposition at every sample size. The authors also find that the 600-point predicted sample reproduces the best-fit parameters obtained from the full 2004-point simulated sample, and that the reconstructed age of the Universe is around 13.7–13.8 Gyr across all techniques. This leads them to conclude that ML predictions, especially SVR, can serve as a computationally cheaper proxy for the full dataset when constraining cosmological parameters.

Load-bearing premise

The entire demonstration assumes the simulated ages are a faithful stand-in for real galaxy ages, with a flat $\Lambda$CDM cosmology as the true expansion history and Gaussian 10% errors.

Editorial extensions

If this is right

  • If SVR reconstruction is as accurate as claimed, a few thousand galaxy ages could yield competitive constraints on the dark-energy equation of state without presupposing a parametric form of the expansion history.
  • Because the 600-point predicted sample matches the 2004-point full sample, the pipeline offers a computationally cheaper route to parameter constraints from large future surveys.
  • SVR's factor-of-ten smaller mean squared error relative to CART and MLPR suggests kernel-based regression is the safer default for smooth cosmologically relevant relations.
  • Recovered universe ages around 13.7–13.8 Gyr imply the reconstruction preserves the global integral of the expansion rate, not just the shape of $t(z)$.

Reading between the lines

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

  • A test the paper leaves implicit: the performance ranking might change if real age errors are non-Gaussian or if a delay-factor prior is included, since the simulations omit the incubation time.
  • The same training-and-reconstruction recipe could be transferred to other smooth cosmological probes, such as the Hubble parameter $H(z)$ or supernova distance moduli, where SVR's bias–variance behavior would likely be similar.
  • The 30-point results in the paper's Table II suggest a minimum sample size below which reconstruction-based parameter estimation loses reliability; quantifying that threshold would be a practical guide for survey design.
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 / 6 minor

Summary. The paper applies three supervised regression methods (CART, MLPR, SVR) to reconstruct the galaxy age-redshift relation from Monte Carlo simulated samples of 100, 1000, 2000, 3334, and 6680 galaxies. The simulated ages are Gaussian draws centered on a flat ΛCDM model with Planck-like parameters (Ωm=0.315±0.007, H0=67.4±0.5, w=-1). The authors compare the three regressors using reconstruction curves, MSE/BVT, and then fit Ωm and the dark-energy equation-of-state parameter w to the reconstructed ages. They report that SVR performs best, and that the 600-point SVR sample yields Ωm=0.329±0.010 and w=-1.054±0.087, which they claim are consistent with values from the literature.

Significance. The machine-learning benchmark part has some value as a controlled comparison: the paper transparently specifies the simulated data generation, uses three standard regressors, and presents a bias-variance decomposition. However, the central cosmological-parameter claim is not supported because the data are generated from the same flat ΛCDM model that is later fitted; recovering the input parameters is a self-consistency check, not an empirical validation. The paper does not provide code or machine-checked proofs, and the surviving contribution—SVR outperforming CART and MLPR on synthetic data—is modest. If the cosmological-parameter claims are removed, the paper becomes a limited methods study rather than a cosmological constraint.

major comments (3)
  1. [§IV A and §V, Table II] The cosmological parameter estimates are circular. The simulated galaxy ages are generated as Gaussian draws centered on a flat ΛCDM model with Ωm=0.315±0.007, H0=67.4±0.5, and w=-1 (Section IV A). Fitting the same model to those ages (Section V) and recovering Ωm≈0.329, w≈-1.054 for the SVR-600 sample is a self-consistency check, not an independent measurement. The paper's own 'without ML' fit of the 2004-point sample (Ωm=0.323, w=-1.01) confirms that the generator already encodes the Planck-like values. The claim of consistency with the literature is therefore guaranteed by construction and cannot validate the reconstruction pipeline or constrain real cosmic history.
  2. [§V, Table II] The selection of the 600-point SVR result is post hoc. Of the fifteen technique/sample combinations, the manuscript highlights the one that happens to match Planck, while other entries deviate strongly (e.g., SVR-30 gives Ωm=0.107, w=-0.265; CART-30 gives Ωm=0.388, w=-0.330). Because the choice is made after examining the fits, the quoted agreement is subject to selection effects and the reported uncertainties do not account for this multiplicity.
  3. [§II A, Eq. (1)] Equation (1) as printed, t(z_i,p)=∫_0^∞ dz'/[(1+z')H(z',p)], has no lower redshift limit and is independent of z_i. The correct lookback age integral is ∫_{z_i}^∞ dz'/[(1+z')H(z',p)]. Unless this is a typographical omission in the manuscript, the theoretical relation cannot produce the redshift-dependent ages used in the simulations and reconstructions; the equation must be corrected.
minor comments (6)
  1. [§II A, Eq. (2)] The notation in Eq. (2) is inconsistent with Eq. (1): H(z',p) is written with factors (1+z) instead of (1+z'), so the integration variable z' does not appear in the Hubble parameter.
  2. [§V, Figure 4 caption] The caption states that the x-axis corresponds to predicted age and the y-axis to redshift, but the plotted panels have redshift on the horizontal axis and age on the vertical axis.
  3. [§V and Table II] The asymmetric error notation (e.g., Ωm=0.329±0.010/0.010) is not defined; the text should state whether asymmetries are from the 16th/84th percentiles or from a different convention.
  4. [§V] The text asserts that SVR's MSE is approximately ten times smaller than that of the other techniques, but no numerical MSE table is provided; the claim should be supported by reported values.
  5. [§V, bullet list] The manuscript states 'We cannot fully explain why this latter behavior occurs' regarding the dip in the reconstructed ages; this unexplained feature is a limitation that should be resolved or explicitly discussed as a caveat before claiming accurate reconstruction.
  6. [References] Reference [34] is formatted as 'N. Planck Collaboration'; the author list should be corrected to the Planck Collaboration.

Circularity Check

2 steps flagged · score 8.0 of 10

Headline cosmological constraints are circular: the simulated galaxy ages are Monte-Carlo draws centered on the same flat ΛCDM fiducial model that is later fitted, so the recovered Ωm and w merely re-express the input parameters.

  1. fitted input called prediction [Section IV A, 'Simulated Data']
    "we used the Monte Carlo Method (MCM) [26] and adopted a Gaussian distribution centered on a flat fiducial ΛCDM model with Ωm = 0.315±0.007 and H0 = 67.4±0.5 km s−1 Mpc−1, values consistent with the best fits obtained by [34]"

    The simulated ages that feed all three ML regressors are generated as Gaussian draws around a flat ΛCDM model with Ωm=0.315 and w=−1. The later parameter fit (Eq. 6 and the emcee analysis of Section V) estimates Ωm and w from these reconstructed ages. For data generated from a known fiducial model, a correctly specified maximum-likelihood fit is centered on the generator's inputs; any departure is only finite-sample scatter. The headline SVR-600 result, Ωm=0.329±0.010 and w=−1.054±0.087, is therefore a self-consistency recovery of the fiducial values, not an independent empirical constraint.

  2. fitted input called prediction [Section V, 'Results']
    "In order to corroborate the methodology adopted in our work, we also calculated the best fit for the simulated samples (without using ML) and found, for example, for the samples with 2004 points, values of the order of Ωm = 0.323±0.007, ω=−1.01±0.052."

    This 'without ML' fit demonstrates that the input flat ΛCDM model is recoverable from the simulated ages even before any machine-learning reconstruction. The ML-predicted parameters are thus inheriting the fiducial cosmology baked into the Monte-Carlo generator. The paper presents the no-ML fit as corroboration that the 600-point predicted sample 'represents the base dataset very well,' but this only confirms that the whole pipeline—simulation, regression, and fit—is internally consistent with its own inputs. It does not provide external evidence about the real cosmic history. The legitimate, non-circular part of the paper is the comparative MSE/BVT ranking of CART, MLPR, and SVR on a fixed synthetic benchmark; the cosmological parameter 'predictions' are not independent.

full rationale

The central cosmological-parameter claim reduces to its own inputs by construction. Section IV A creates every simulated age as a Gaussian draw centered on a flat ΛCDM fiducial with Ωm=0.315 and H0=67.4, which is equivalent to w=−1. Section V then fits the same class of flat ωCDM model (Eq. 6 and the emcee posteriors) to ML-reconstructed versions of those simulated ages, recovering Ωm≈0.329 and w≈−1.054 for the preferred SVR-600 sample. Because the data are generated from the fitted model, consistency with the fiducial values is not evidence of empirical validity; it is a closed-loop check. The paper's own no-ML fit of the simulated 2004-point sample (Ωm=0.323±0.007, w=−1.01±0.052) makes this explicit: the generator already encodes Planck-like parameters, and both the ML and no-ML fits merely return them. The 600-point SVR result is also highlighted partly because it best matches [34], which is the same source used to fix the simulation's input parameters, so the favorable comparison is partly an artifact of sample selection. What survives as non-circular is the methodological benchmark: SVR has lower MSE and BVT than CART and MLPR on this fixed synthetic data set, and the reconstructed age curves can be compared as a regression exercise. That part is self-contained and does not depend on the fiducial parameters. However, the paper's headline and abstract present the recovered Ωm and w as cosmological constraints 'consistent with the values from the literature,' which overstates their epistemic status. The strongest interpretation warranted by the derivation is that SVR can accurately invert a known fiducial model from synthetic data, not that it independently measures cosmic history. Score 8 reflects that the paper's central quantitative claim is forced by the simulation design rather than by a self-citation chain; there is no reliance on the authors' prior work as authority.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim depends on a closed simulation loop. The only fitted cosmological parameters are Omega_m and w; ML hyperparameters are tuned on the same simulated data, adding flexibility. No new entities are postulated.

free parameters (3)
  • Omega_m = 0.329 (SVR, 600 predicted points)
    Free parameter in the flat wCDM fit; the value is close to the simulation input Omega_m=0.315 from Planck.
  • w = -1.054 (SVR, 600 predicted points)
    Dark energy equation of state parameter in the fit; close to the input w=-1.
  • ML hyperparameters (C, gamma, epsilon for SVR; max_depth, min_samples_leaf for CART; alpha, solver, activation for MLPR) = selected by GridSearchCV on each training sample
    Tuned on the simulated data; this flexibility is not accounted for in the reported parameter errors.
assumptions (4)
  • standard math Flat Friedmann-Lemaitre cosmology with age-redshift relation t(z)=integral dz/[(1+z) H(z)]
    Used in equation (1) and throughout; standard cosmology.
  • domain assumption Simulated galaxy ages follow a Gaussian distribution centered on the fiducial flat LambdaCDM curve with 10% scatter
    Section IV A: the Monte Carlo samples are generated from this assumption, which controls the whole test.
  • ad hoc to paper The delay factor (incubation time) can be ignored in the simulated samples
    Section II A and IV A state theoretical ages used in simulations do not require the delay factor; this avoids a nuisance parameter but may not hold for real data.
  • ad hoc to paper The 70/30 train-test split and the chosen sample sizes are representative of future surveys
    Section III and V; the split is conventional, and the sample sizes are justified by expected ACT/SALT observations, but this is an assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reconstructing Cosmic History with Machine Learning: A Study Using CART, MLPR, and SVR." pith.science (2026). https://pith.science/paper/A5L3RQZN

@misc{pith2026250517205,
  author       = {Pith},
  title        = {Pith review of: Reconstructing Cosmic History with Machine Learning: A Study Using CART, MLPR, and SVR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A5L3RQZN}},
  note         = {Machine review of arXiv:2505.17205}
}
abstract

In this work, we reconstruct cosmic history via supervised learning through three methods: Classification and Regression Trees (CART), Multi-layer Perceptron Regressor (MLPR), and Support Vector Regression (SVR). For this purpose, we use ages of simulated galaxies based on 32 massive, early-time, passively evolving galaxies in the range $0.12 < z < 1.85$, with absolute ages determined. Using this sample, we simulate subsamples of 100, 1000, 2000, 3334, 6680 points, through the Monte Carlo Method and adopting a Gaussian distribution centering on a spatially flat $\Lambda$CDM as a fiducial model. We found that the SVR method demonstrates the best performance during the process. The methods MLPR and CART also present satisfactory performance, but their mean square errors are greater than those found for the SVR. Using the reconstructed ages, we estimate the matter density parameter and equation of state (EoS) and our analysis found the SVR with 600 predict points obtains $\Omega_m=0.329\pm{}^{0.010}_{0.010}$ and the dark energy EoS parameter $\omega= -1.054\pm{}^{0.087}_{0.126}$, which are consistent with the values from the literature. We highlight that we found the most consistent results for the subsample with 2000 points, which returns 600 predicted points and has the best performance, considering its small sample size and high accuracy. We present the reconstructed curves of galaxy ages and the best fits cosmological parameters.

Figures

Figures reproduced from arXiv: 2505.17205 by the authors.

Figure 1
Figure 1. Original sample of 32 galaxies, as described in the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Simulated age sample with 100 points [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. As expected, all methods display a decreasing trend [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 3
Figure 3. Figure 3: Bias-variance tradeoff (BVT) as a function of the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png]
Figure 4
Figure 4. Figure 4: Reconstructions via ML for five samples using three techniques: CART (first column), MLPR (second column), and [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Results for the ωCDMk = 0 model with 30 predicted age points using the CART, MLPR, and SVR techniques, respectively. The 68.3% and 95.4% confidence regions are shown in the Ωm − ω plane. The dashed lines represent the best fits illustrated in the figures with their res…
Figure 6
Figure 6. Figure 6: Same as Figure 5, but with 300 predicted ages. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Same as Figure 5, but with 600 predicted ages. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Same as Figure 5, but with 1001 predicted ages. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Same as Figure 5, but with 2004 predicted ages. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 35 canonical work pages

  1. [1]

    J. P. Ostriker and P. J. Steinhardt, Cosmic concordance (1995), arXiv:astro-ph/9505066 [astro-ph]

  2. [2]

    Jimenez, The age of the universe (1997), arXiv:astro- ph/9701222 [astro-ph]

    R. Jimenez, The age of the universe (1997), arXiv:astro- ph/9701222 [astro-ph]

  3. [3]

    Chaboyer, Physics Reports307, 23–30 (1998)

    B. Chaboyer, Physics Reports307, 23–30 (1998)

  4. [4]

    L. M. Krauss and M. S. Turner, General Relativity and Gravitation27, 1137–1144 (1995)

  5. [5]

    L. M. Krauss and B. Chaboyer, Science299, 65 (2003)

  6. [6]

    Jimenez and A

    R. Jimenez and A. Loeb, The Astrophysical Journal573, 37–42 (2002)

  7. [7]

    Simon, L

    J. Simon, L. Verde, and R. Jimenez, Physical Review D 71, 123001 (2005)

  8. [8]

    L. A. Nolan, J. S. Dunlop, and R. Jimenez, Monthly No- tices of the Royal Astronomical Society323, 385 (2001)

Show all 40 references
  1. [9]

    Vagnozzi, F

    S. Vagnozzi, F. Pacucci, and A. Loeb, Journal of High Energy Astrophysics36, 27–35 (2022)

  2. [10]

    Wei and F

    J.-J. Wei and F. Melia, The Astrophysical Journal928, 165 (2022)

  3. [11]

    Dantas, J

    M. Dantas, J. Alcaniz, D. Jain, and A. Dev, Astronomy & Astrophysics467, 421 (2007)

  4. [12]

    Dantas, J

    M. Dantas, J. Alcaniz, D. Mania, and B. Ratra, Physics Letters B699, 239 (2011)

  5. [13]

    Dantas, J

    M. Dantas, J. Alcaniz, and N. Pires, Physics Letters B 679, 423–427 (2009)

  6. [14]

    Capozziello, V

    S. Capozziello, V. F. Cardone, M. Funaro, and S. An- dreon, Physical Review D70, 123501 (2004)

  7. [15]

    Pires, Z.-H

    N. Pires, Z.-H. Zhu, and J. S. Alcaniz, Physical Review D73, 123530 (2006)

  8. [16]

    S. S. Binici, C. Deliduman, and F. S ¸akir Dilsiz, The ages of the oldest astrophysical objects in an ellipsoidal uni- verse (2024), arXiv:2402.16646 [astro-ph.CO]

  9. [17]

    C.-Y. Gao, M. L´ opez-Corredoira, and J.-J. Wei, The As- trophysical Journal970, 142 (2024)

  10. [18]

    Castro-Rodr ´ ıguez and M

    N. Castro-Rodr ´ ıguez and M. L´ opez-Corredoira, Astron- omy and Astrophysics537, A31 (2012), arXiv:1111.2726 [astro-ph.CO]

  11. [19]

    Chac´ on, J

    J. Chac´ on, J. A. V´ azquez, and E. Almaraz, arXiv preprint arXiv:2106.06587 (2021)

  12. [20]

    Arjona, A

    R. Arjona, A. Melchiorri, and S. Nesseris, Testing the λcdm paradigm with growth rate data and machine learning (2021), arXiv:2107.04343 [astro-ph.CO]

  13. [21]

    von Marttens, L

    R. von Marttens, L. Casarini, N. R. Napolitano, S. Wu, V. Amaro, R. Li, C. Tortora, A. Canabarro, and Y. Wang, arXiv preprint arXiv:2111.01185 (2021)

  14. [22]

    Bengaly, M

    C. Bengaly, M. A. Dantas, L. Casarini, and J. Al- caniz, European Physical Journal C83, 548 (2023), arXiv:2209.09017 [astro-ph.CO]

  15. [23]

    Pal and R

    S. Pal and R. Saha, Physica Scripta99, 115007 (2024), arXiv:2309.15179 [astro-ph.CO]

  16. [24]

    Mukherjee and A

    P. Mukherjee and A. A. Sen, A new∼5σtension at characteristic redshift from desi-dr1 bao and des-sn5yr observations (2025), arXiv:2503.02880 [astro-ph.CO]

  17. [25]

    Borghetto, A

    G. Borghetto, A. Malhotra, G. Tasinato, and I. Zavala, Bounded dark energy (2025), arXiv:2503.11628 [astro- ph.CO]

  18. [26]

    Metropoliset al., Los Alamos Science15, 125 (1987)

    N. Metropoliset al., Los Alamos Science15, 125 (1987)

  19. [27]

    Sandage, Annual Review of Astron and Astrophys26, 561 (1988)

    A. Sandage, Annual Review of Astron and Astrophys26, 561 (1988)

  20. [28]

    P. J. E. Peebles,Principles of physical cosmology, Vol. 27 (Princeton university press, 1993)

  21. [29]

    Savage, N

    C. Savage, N. Sugiyama, and K. Freese, Journal of Cos- mology and Astroparticle Physics2005(10), 007

  22. [30]

    Doran, S

    M. Doran, S. Stern, and E. Thommes, Journal of Cos- mology and Astroparticle Physics2007(04), 015

  23. [31]

    Komatsu, J

    E. Komatsu, J. Dunkley, M. Nolta, C. Bennett, B. Gold, G. Hinshaw, N. Jarosik, D. Larson, M. Limon, L. Page, et al., The Astrophysical Journal Supplement Series180, 330 (2009)

  24. [32]

    R. G. Abraham, K. Glazebrook, P. J. McCarthy, D. Crampton, R. Murowinski, I. Jørgensen, K. Roth, I. M. Hook, S. Savaglio, H.-W. Chen,et al., The As- tronomical Journal127, 2455 (2004)

  25. [33]

    W. J. Percival, B. A. Reid, D. J. Eisenstein, N. A. Bah- call, T. Budavari, J. A. Frieman, M. Fukugita, J. E. Gunn, ˇZ. Ivezi´ c, G. R. Knapp,et al., Monthly Notices of the Royal Astronomical Society401, 2148 (2010)

  26. [34]

    Planck Collaboration, Aghanim, Y

    N. Planck Collaboration, Aghanim, Y. Akrami, M. Ash- down, J. Aumont, C. Baccigalupi, M. Ballardini, A. Ban- day, R. Barreiro, N. Bartolo, S. Basak,et al., Astronomy & Astrophysics641, A6 (2020)

  27. [35]

    Kosowsky, New Astronomy Reviews47, 939 (2003)

    A. Kosowsky, New Astronomy Reviews47, 939 (2003)

  28. [36]

    Stobie, J

    R. Stobie, J. G. Meiring, and D. A. Buckley, inOptical Design, Materials, Fabrication, and Maintenance, Vol. 4003 (International Society for Optics and Photonics,

  29. [37]

    Foreman-Mackey, D

    D. Foreman-Mackey, D. W. Hogg, D. Lang, and J. Good- man, Publications of the Astronomical Society of the Pa- cific125, 306–312 (2013)

  30. [38]

    S. R. Hinton, The Journal of Open Source Software1, 00045 (2016)

  31. [39]

    Valcin, R

    D. Valcin, R. Jimenez, U. Seljak, and L. Verde, The age of the universe with globular clusters iii: Gaia distances and hierarchical modeling (2025), arXiv:2503.19481 [astro- ph.CO]

  32. [6680]

    The results demonstrated that the SVR technique performed the best, accurately capturing the nonlinear behavior of the data while avoiding overfitting

    generated by the Monte Carlo Method (MCM). The results demonstrated that the SVR technique performed the best, accurately capturing the nonlinear behavior of the data while avoiding overfitting. The CART method, despite re- producing nonlinear behavior, yielded higher MSEs and...

Pith tools

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