Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Savage-Dickey density ratio estimation with normalizing flows for Bayesian model comparison

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The Savage-Dickey Bayes-factor shortcut stays accurate in high dimensions once histograms are replaced by normalizing flows, letting nested cosmological models be compared from one posterior sample set alone.

desk verdict Solid, well-motivated method for nested model comparison, but the field-level 'consistency' claim is several sigma shy of what the abstract advertises. read the letter →

arxiv 2506.04339 v1 pith:SWFO7XYO submitted 2025-06-04 astro-ph.CO astro-ph.IMstat.ML

Kiyam Lin , Alicja Polanska , Davide Piras , Alessio Spurio Mancini , Jason D. McEwen This is my paper
classification astro-ph.COastro-ph.IMstat.ML
keywords Savage-DickeydensityrationormalizingflowsBayesianmodelcomparisonBayesfactorcosmologicalinferencefield-levelweaklensingrationalquadraticsplines
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 argues that the Savage-Dickey density ratio, a standard shortcut for comparing nested Bayesian models, can be made scalable by replacing histogram density estimates with normalizing flows. The claim is that a flow trained on posterior samples of the super model supplies an accurate, normalized value of the marginal posterior at the nesting point, even when there are many extra parameters and histograms become computationally intractable. If this holds, Bayes factors between nested cosmological models can be obtained from a single MCMC run on the super model, roughly halving the cost of model comparison. The authors demonstrate the approach on toy Gaussians and on dark-energy and field-level weak-lensing examples, validating against nested sampling and the learned harmonic mean where possible.

What carries the argument

The load-bearing object is the Savage-Dickey identity $z_1/z_2 = p(\eta_1 \mid d, M_2)/p(\eta_1 \mid M_2)$, which expresses the Bayes factor between nested models as the ratio of the normalized marginal posterior to the prior of the extra parameters, evaluated at the nesting value $\eta_1$. The paper's contribution is to compute the numerator with a normalizing flow: a rational quadratic spline with alternating affine layers, trained by KL-divergence minimization against posterior samples, yields a density that is normalized by construction and can be evaluated at a single point. Bootstrap resampling of the posterior samples, with an independently trained flow per resample, supplies the uncertainty estimate.

What would settle it

Compare flow-based SDDR against an exact Bayes factor (from nested sampling or the learned harmonic mean) for a nested model whose nesting value lies in a low-density tail of a known multi-dimensional Gaussian posterior; if the flow estimate disagrees by more than its bootstrap error, the method is not unbiased at the evaluation point. A cheaper test: in a four-extra-parameter toy with a known density, evaluate the flow at $\eta_1$ against a finely binned histogram or analytic value and measure the pointwise bias across many bootstrapped training sets.

Watch

Extended reading notes

Core claim

The central discovery is a neural estimator for the SDDR: train a rational quadratic spline normalizing flow to minimize the KL divergence to the marginal posterior samples of the extra parameters, then evaluate the learned density at the point $\eta_1$ where the super model reduces to the nested model. Because the flow is normalized by construction, it bypasses the numerical integration that makes histogram-based SDDR impractical beyond a few extra parameters. The paper reports accurate Bayes factors in a 4-extra-parameter toy case where the histogram method could not be applied, and consistent results on DES Y1 wCDM versus $\Lambda$CDM, a Stage-IV $w_0w_a$CDM example with 39 parameters, and a field-level weak-lensing BHM/SBI comparison. In the field-level setting, the paper computes Bayes factors for the BHM wCDM/$\Lambda$CDM comparison for the first time and finds them consistent with the SBI analysis, which it reads as confirmation that SBI extracts the same cosmological information from the field.

Load-bearing premise

The load-bearing premise is that the normalizing flow returns an unbiased value of the marginal posterior at the single nesting point, together with the paper's own caveat that the posterior must have sufficient sample density around that point—a condition the cosmological demonstrations do not directly verify.

Editorial extensions

If this is right

  • Bayes factors for all nested sub-models of a fitted super model become available from a single posterior sample save, eliminating the need to sample the nested model and roughly halving model-comparison cost.
  • Histogram-based SDDR is limited to about one- or two-dimensional extra parameter spaces; the flow version extends accurate SDDR to at least four extra parameters, and the paper argues to substantially more.
  • Field-level cosmological analyses, where evidence calculation is otherwise intractable, become amenable to model comparison when the number of extra parameters is small.
  • Consistent BHM and SBI Bayes factors in the field-level example strengthen the case that the SBI pipeline retains the cosmological information of the full hierarchical model.
  • The method is implemented in open-source software and runs in minutes on a CPU given the MCMC chains.

Reading between the lines

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

  • A natural extension is to report a local diagnostic for the SDDR validity condition, such as the effective sample density of the posterior near $\eta_1$, since the flow's point evaluation cannot detect whether that condition is met.
  • The bootstrap uncertainty quoted in the paper covers sampling noise but not flow approximation bias; an ensemble of flows with different architectures or a calibration check against the learned harmonic mean in low dimensions would separate those error sources.
  • The same point-evaluation strategy could be applied to other one-point summaries, such as marginal likelihoods at fixed parameter values in profile or integrated nested Laplace approximations, wherever a normalized marginal density is needed.
  • If the method generalizes to dozens of extra parameters, it could make routine the comparison of cosmological models that differ in nuisance or systematics parametrizations, where nested sampling is currently prohibitive.
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 / 5 minor

Summary. The paper proposes a Savage-Dickey density ratio (SDDR) estimator that uses normalizing flows (specifically rational quadratic spline flows) to evaluate the marginal posterior density at the nesting point, replacing histogram-based density estimates. The method is implemented in the open-source harmonic package. The authors validate it on two Gaussian toy models (1 and 4 extra parameters), a DES Y1-like wCDM versus ΛCDM analysis, a Stage-IV weak lensing ΛCDM versus w0waCDM analysis, and a field-level weak lensing BHM versus SBI comparison. The central claims are that the flow-based SDDR scales to high-dimensional extra-parameter spaces, that it agrees with nested sampling and learned harmonic mean in low-dimensional cases, and that BHM and SBI Bayes factors are consistent in the field-level setting.

Significance. If the method is sound, it offers a practical way to compute Bayes factors for nested models using only super-model posterior samples, potentially halving the computational cost of model comparison in high-dimensional cosmological analyses. The open-source implementation in harmonic is a useful contribution. The toy and DES examples provide supporting evidence that the method works in low-dimensional settings. However, the load-bearing validation claims in the cosmological examples are not all supported by the reported numbers: the flow-based SDDR differs from the independent learned harmonic mean estimate in the Stage-IV case at the ~2.4 sigma level, and in the field-level SBI case at the 6–8 sigma level. These discrepancies directly affect the paper's headline conclusion that BHM and SBI Bayes factors are consistent, and they warrant substantive revision rather than minor polishing.

major comments (3)
  1. [Section 3.4, Table 5] The claim that BHM and SBI Bayes factors are consistent is not supported by Table 5. The BHM SDDR (flows) value of 0.947 ± 0.011 differs from the SBI learned harmonic mean value of 1.093 ± 0.014 by about 8 standard deviations, and the SBI SDDR (flows) value of 0.934 ± 0.020 differs from the SBI LHM value by about 6.5 standard deviations. The text states these results are 'consistent' and 'in close agreement' based on the Jeffreys scale, but a 0.15 difference in log Bayes factor is not statistical consistency and can change model-selection conclusions in other contexts. This discrepancy undermines the conclusion that SBI extracts as much cosmological information as the BHM, since the only independent reference point (SBI LHM) disagrees with the flow-based SDDR values. The authors should either explain the offset, soften the consistency claim, or present the discrepancy as a limitation.
  2. [Section 2.4/2.5 and Section 3.3] The SDDR estimate depends on the flow density evaluated at the single point η1, so any flow bias or variance at that point propagates directly into the Bayes factor. The bootstrap procedure in Section 2.5 accounts only for resampling variability of the posterior samples, not for flow approximation error (e.g., limited flexibility, local mismatch near η1, or sensitivity to training hyperparameters). The paper acknowledges in Section 2.2 that the SDDR requires sufficient sample density around η1, but no diagnostic is provided for the cosmological examples. The need for such a diagnostic is highlighted by Table 4, where the flow-based SDDR 1.75 ± 0.06 disagrees with the learned harmonic mean 1.53 ± 0.07 at roughly 2.4 sigma, and the nested sampling value 0.78 ± 0.71 is too uncertain to arbitrate. The authors should demonstrate that the flow density at η1 is accurate and stable, for instance by comparing with a histogram in low-dimensional marginals, or by reporting flow estimates from multiple random seeds and hyperparameter settings.
  3. [Section 3.4] The agreement between BHM SDDR (flows) and SBI SDDR (flows) is used as evidence that the SBI approach extracts as much cosmological information as the BHM. However, both values are obtained with the same flow-based SDDR method and may share a common bias, so their mutual agreement does not independently validate either the flow method or the SBI information content. The only independent estimate in Table 5 (SBI learned harmonic mean) is not consistent with either flow-based value. Consequently, the conclusion that 'SBI extracts as much cosmological information from the field as the BHM approach' is not established by the presented results. The authors should obtain an independent evidence estimate for the BHM case or explicitly reframe the BHM/SBI comparison as preliminary and in need of external validation.
minor comments (5)
  1. [Section 2 header] The word 'uncertatinties' should be 'uncertainties'.
  2. [Section 2.5] The bootstrap procedure would benefit from stating the number of bootstrap resamples and the number of independently trained flows used in the reported results, as well as a note on the computational cost of retraining.
  3. [Listing 1] The value log_prior_η_1 = -2 appears without context; a few lines of explanation on how to compute the prior log-density at η1 from the model priors would improve usability.
  4. [Section 3.3] In Table 4, the nested sampling uncertainty (±0.71) is much larger than the other estimates; the text should state this explicitly when describing the agreement, rather than saying only that the results are 'broadly consistent'.
  5. [Section 3.4] The phrase 'consistent results and correct preference' in the text describing Table 5 is misleading given the large discrepancy with the learned harmonic mean; consider rewording to report the numerical values without the consistency language unless the offset is resolved.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the SDDR-with-flows derivation is self-contained and validated on independent external benchmarks; the main concerns are validation-consistency issues, not circularity.

full rationale

The paper's derivation chain is not circular. The SDDR formula in Eq. (6) and Appendix A is a standard Bayes-theorem identity; the numerator p(eta_1|d,M_2) is estimated by a normalizing flow trained to minimize KL divergence against posterior samples, and the Bayes factor itself is not used as a training target. Thus the flow-based SDDR is a genuine density estimate, not a fitted parameter renamed as a prediction. The method is checked against independent estimators: nested sampling with nautilus and the learned harmonic mean, on toy Gaussians and the DES wCDM example, with agreement reported in Tables 1-3. Self-citations to Polanska et al. (2024a), Piras et al. (2024), and Spurio Mancini et al. (2024) supply posterior samples and reference values, but those are reproducible external computations and are not invoked as an unverified uniqueness or ansatz argument. Section 2.2 explicitly states the SDDR validity condition and recommends learned harmonic mean when support at eta_1 is insufficient, and Section 2.5 acknowledges that bootstrap errors capture sampling variability, not flow approximation error. The most serious concern in the manuscript is not circularity but internal consistency of the headline validation claim: in Table 5 the independent SBI learned-harmonic-mean value (1.093 +/- 0.014) differs from both BHM SDDR (0.947 +/- 0.011) and SBI SDDR (0.934 +/- 0.020) by roughly 6-8 sigma, and the BHM/SBI SDDR agreement could share a common flow-estimator bias. This is a correctness/validation weakness, not a definitional reduction of the derived quantity to its input, so it does not raise the circularity score.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The method rests on the standard SDDR nesting conditions, the reliability of the flow density at the nesting point, and the validity of reused posterior samples. No new physical entities are introduced; the only hand-chosen inputs are flow training hyperparameters, which are not reported.

free parameters (1)
  • normalizing flow hyperparameters (number of spline bins, number of layers, learning rate, training epochs, bootstrap… = not reported
    The paper specifies RQSpline flows (Durkan et al. 2019) and alternating affine layers in Section 2.4 but does not report architecture or training hyperparameters used in the applications. These choices affect the density estimate and hence the Bayes factor; they are selected by the authors rather than derived.
assumptions (5)
  • domain assumption Prior separability and identical common-parameter priors between nested and super model: p(theta, eta) = p(theta) p(eta) and p(theta | M1) = p(theta | M2).
    Required for the SDDR reduction in Section 2.2 (conditions 3 and 4) and Appendix A. In cosmological applications this holds by construction of the priors, but it is an assumption the user must enforce.
  • domain assumption Sufficient posterior support and sample density at the nesting point eta_1.
    Stated in Section 2.2: 'it is only valid when the posterior has sufficient support with high enough sample density around the parameter values where the super model reduces to the nested model'. No quantitative test is applied in the cosmological examples.
  • ad hoc to paper The normalizing flow is flexible enough to represent the marginal posterior and KL training converges to the true density.
    Section 2.4 assumes the flow learns the marginal posterior; in practice RQSpline flows have finite capacity and optimization is stochastic. The paper provides no convergence diagnostics or comparison of flow density to alternative estimates in high dimensions.
  • domain assumption MCMC chains from previous works (Polanska et al. 2024a; Piras et al. 2024; Lanzieri et al. 2024; Zeghal et al. 2024; Spurio Mancini et al. 2024) are converged and representative.
    The DES, Stage-IV, and field-level analyses reuse posterior samples from other papers (Sections 3.2-3.4); all downstream SDDR estimates inherit any MCMC non-convergence or thinning issues.
  • ad hoc to paper For the SBI validation, the neural likelihood estimation posterior from Spurio Mancini et al. (2024) is an accurate approximation of the true posterior.
    The consistency claim in Section 3.4 treats SBI as a reliable comparator, but SBI itself is approximate; its accuracy is not independently established in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Savage-Dickey density ratio estimation with normalizing flows for Bayesian model comparison." pith.science (2026). https://pith.science/paper/SWFO7XYO

@misc{pith2026250604339,
  author       = {Pith},
  title        = {Pith review of: Savage-Dickey density ratio estimation with normalizing flows for Bayesian model comparison},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SWFO7XYO}},
  note         = {Machine review of arXiv:2506.04339}
}
read the original abstract

A core motivation of science is to evaluate which scientific model best explains observed data. Bayesian model comparison provides a principled statistical approach to comparing scientific models and has found widespread application within cosmology and astrophysics. Calculating the Bayesian evidence is computationally challenging, especially as we continue to explore increasingly more complex models. The Savage-Dickey density ratio (SDDR) provides a method to calculate the Bayes factor (evidence ratio) between two nested models using only posterior samples from the super model. The SDDR requires the calculation of a normalised marginal distribution over the extra parameters of the super model, which has typically been performed using classical density estimators, such as histograms. Classical density estimators, however, can struggle to scale to high-dimensional settings. We introduce a neural SDDR approach using normalizing flows that can scale to settings where the super model contains a large number of extra parameters. We demonstrate the effectiveness of this neural SDDR methodology applied to both toy and realistic cosmological examples. For a field-level inference setting, we show that Bayes factors computed for a Bayesian hierarchical model (BHM) and simulation-based inference (SBI) approach are consistent, providing further validation that SBI extracts as much cosmological information from the field as the BHM approach. The SDDR estimator with normalizing flows is implemented in the open-source harmonic Python package.

Figures

Figures reproduced from arXiv: 2506.04339 by the authors.

Figure 1
Figure 1. Posterior contours of the toy Gaussian model with 4 extra parameters including both the super model (red) and the nested model (blue). The dashed lines depict the ground truth. a robust and precise estimate of the Bayes factor. For the field-level setting, we consider the wCDM ana￾lysis presented in Lanzieri et al. (2024); Zeghal et al. (2024). This scenario considers wCDM with a field-level cosmological analysis co… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 18 canonical work pages

  1. [1]

    Ashton G., et al., 2022, @doi [Nature Reviews Methods Primers] 10.1038/s43586-022-00121-x , https://ui.adsabs.harvard.edu/abs/2022NRvMP...2...39A 2, 39

    doi:10.1038/s43586-022-00121-x 2022

  2. [2]

    Buchner J., 2021, @doi [The Journal of Open Source Software] 10.21105/joss.03001 , https://ui.adsabs.harvard.edu/abs/2021JOSS....6.3001B 6, 3001

    doi:10.21105/joss.03001 2021

  3. [3]

    Buchner J., 2023, @doi [Statistics Surveys] 10.1214/23-SS144 , https://ui.adsabs.harvard.edu/abs/2023StSur..17..169B 17, 169

    doi:10.1214/23-ss144 2023

  4. [4]

    D., Pereyra M., 2021, @doi [arXiv e-prints] 10.48550/arXiv.2106.03646 , https://ui.adsabs.harvard.edu/abs/2021arXiv210603646C p

    Cai X., McEwen J. D., Pereyra M., 2021, @doi [arXiv e-prints] 10.48550/arXiv.2106.03646 , https://ui.adsabs.harvard.edu/abs/2021arXiv210603646C p. arXiv:2106.03646

    arXiv doi:10.48550/arxiv.2106.03646 2021

  5. [5]

    Campagne J.-E., et al., 2023, @doi [The Open Journal of Astrophysics] 10.21105/astro.2302.05163 , https://ui.adsabs.harvard.edu/abs/2023OJAp....6E..15C 6, 15

    arXiv 2023

  6. [6]

    C., 2025, @doi [ ] 10.1093/mnras/staf663 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.539.3220C 539, 3220

    Carrion K., Spurio Mancini A., Piras D., Hidalgo J. C., 2025, @doi [ ] 10.1093/mnras/staf663 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.539.3220C 539, 3220

    doi:10.1093/mnras/staf663 2025

  7. [7]

    Di Valentino E., Melchiorri A., Mena O., 2017, @doi [ ] 10.1103/PhysRevD.96.043503 , https://ui.adsabs.harvard.edu/abs/2017PhRvD..96d3503D 96, 043503

    doi:10.1103/physrevd.96.043503 2017

  8. [8]

    Di Valentino E., Melchiorri A., Silk J., 2020, @doi [Nature Astronomy] 10.1038/s41550-019-0906-9 , https://ui.adsabs.harvard.edu/abs/2020NatAs...4..196D 4, 196

    doi:10.1038/s41550-019-0906-9 2020

Show all 53 references
  1. [9]

    J., Kass R

    DiCiccio T. J., Kass R. E., Raftery A., Wasserman L., 1997, Journal of the American Statistical Association, 92, 903

    1997

  2. [10]

    M., 1971, @doi [The Annals of Mathematical Statistics] 10.1214/aoms/1177693507 , 42, 204

    Dickey J. M., 1971, @doi [The Annals of Mathematical Statistics] 10.1214/aoms/1177693507 , 42, 204

    1971

  3. [11]

    Du G.-H., Wu P.-J., Li T.-N., Zhang X., 2025, @doi [European Physical Journal C] 10.1140/epjc/s10052-025-14094-0 , https://ui.adsabs.harvard.edu/abs/2025EPJC...85..392D 85, 392

    2025 doi

  4. [12]

    J., Roweth D., 1987, @doi [Physics Letters B] https://doi.org/10.1016/0370-2693(87)91197-X , 195, 216

    Duane S., Kennedy A., Pendleton B. J., Roweth D., 1987, @doi [Physics Letters B] https://doi.org/10.1016/0370-2693(87)91197-X , 195, 216

    1987 doi

  5. [13]

    Neural Information Processing Systems Foundation, Inc, pp 7511--7522, https://neurips.cc/

    Durkan C., Bekasovs A., Murray I., Papamakarios G., 2019, in Advances in Neural Information Processing Systems 32 (NeurIPS 2019). Neural Information Processing Systems Foundation, Inc, pp 7511--7522, https://neurips.cc/

    2019

  6. [15]

    P., Bridges M., 2009, @doi [Monthly Notices of the Royal Astronomical Society] 10.1111/j.1365-2966.2009.14548.x , 398, 1601

    Feroz F., Hobson M. P., Bridges M., 2009, @doi [Monthly Notices of the Royal Astronomical Society] 10.1111/j.1365-2966.2009.14548.x , 398, 1601

    2009

  7. [16]

    P., Cameron E., Pettitt A

    Feroz F., Hobson M. P., Cameron E., Pettitt A. N., 2019, @doi [The Open Journal of Astrophysics] 10.21105/astro.1306.2144 , 2

    2019 arXiv

  8. [17]

    W., Lang D., Goodman J., 2013, @doi [ ] 10.1086/670067 , https://ui.adsabs.harvard.edu/abs/2013PASP..125..306F 125, 306

    Foreman-Mackey D., Hogg D. W., Lang D., Goodman J., 2013, @doi [ ] 10.1086/670067 , https://ui.adsabs.harvard.edu/abs/2013PASP..125..306F 125, 306

    2013 doi

  9. [18]

    J., Hobson M

    Handley W. J., Hobson M. P., Lasenby A. N., 2015a, @doi [Monthly Notices of the Royal Astronomical Society: Letters] 10.1093/mnrasl/slv047 , 450, L61

    doi

  10. [19]

    J., Hobson M

    Handley W. J., Hobson M. P., Lasenby A. N., 2015b, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stv1911 , 453, 4384

    doi

  11. [20]

    Heavens A., Fantaye Y., Mootoovaloo A., Eggers H., Hosenie Z., Kroon S., Sellentin E., 2017, Marginal Likelihoods from Monte Carlo Markov Chains ( @eprint arXiv 1704.03472 ), @doi 10.48550/arXiv.1704.03472

    2017 doi arXiv

  12. [21]

    Higson E., Handley W., Hobson M., Lasenby A., 2019, @doi [Statistics and Computing] 10.1007/s11222-018-9844-0 , https://ui.adsabs.harvard.edu/abs/2019S&C....29..891H 29, 891

    2019 doi

  13. [22]

    D., Gelman A., 2014, J

    Hoffman M. D., Gelman A., 2014, J. Mach. Learn. Res., 15, 1593–1623

    2014

  14. [23]

    Jeffreys H., 1939, Theory of Probability

    1939

  15. [24]

    B., Liang D., eds, Proceedings of Machine Learning Research Vol

    Jia H., Seljak U., 2020, in Zhang C., Ruiz F., Bui T., Dieng A. B., Liang D., eds, Proceedings of Machine Learning Research Vol. 118, Proceedings of The 2nd Symposium on Advances in Approximate Bayesian Inference. PMLR, pp 1--14, https://proceedings.mlr.press/v118/jia20a.html

    2020

  16. [25]

    D., Cyr-Racine F.-Y., Dor \'e O., 2020, @doi [ ] 10.1103/PhysRevD.101.123505 , https://ui.adsabs.harvard.edu/abs/2020PhRvD.101l3505K 101, 123505

    Kreisch C. D., Cyr-Racine F.-Y., Dor \'e O., 2020, @doi [ ] 10.1103/PhysRevD.101.123505 , https://ui.adsabs.harvard.edu/abs/2020PhRvD.101l3505K 101, 123505

    2020 doi

  17. [26]

    U., 2023, @doi [ ] 10.1093/mnras/stad2441 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.525.3181L 525, 3181

    Lange J. U., 2023, @doi [ ] 10.1093/mnras/stad2441 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.525.3181L 525, 3181

    2023 doi

  18. [27]

    L., Boucaud A., Starck J.-L., Lanusse F., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2407.10877 , https://ui.adsabs.harvard.edu/abs/2024arXiv240710877L p

    Lanzieri D., Zeghal J., Makinen T. L., Boucaud A., Starck J.-L., Lanusse F., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2407.10877 , https://ui.adsabs.harvard.edu/abs/2024arXiv240710877L p. arXiv:2407.10877

    2024 doi arXiv

  19. [28]

    V., Verde L., 2014, @doi [ ] 10.1103/PhysRevLett.113.041301 , https://ui.adsabs.harvard.edu/abs/2014PhRvL.113d1301L 113, 041301

    Leistedt B., Peiris H. V., Verde L., 2014, @doi [ ] 10.1103/PhysRevLett.113.041301 , https://ui.adsabs.harvard.edu/abs/2014PhRvL.113d1301L 113, 041301

    2014 doi

  20. [29]

    arXiv:0910.1452

    Marin J.-M., Robert C., 2009, @doi [arXiv e-prints] 10.48550/arXiv.0910.1452 , https://ui.adsabs.harvard.edu/abs/2009arXiv0910.1452M p. arXiv:0910.1452

    2009 doi arXiv

  21. [30]

    D., Wallis C

    McEwen J. D., Wallis C. G. R., Price M. A., Spurio Mancini A., 2021, @doi [arXiv e-prints] 10.48550/arXiv.2111.12720 , https://ui.adsabs.harvard.edu/abs/2021arXiv211112720M p. arXiv:2111.12720

    2021 doi arXiv

  22. [31]

    D., Liaudat T

    McEwen J. D., Liaudat T. I., Price M. A., Cai X., Pereyra M., 2023. p. arXiv:2307.00056 ( @eprint arXiv 2307.00056 ), @doi 10.48550/arXiv.2307.00056

    2023 doi arXiv

  23. [32]

    M., 1996, Bayesian Learning for Neural Networks, Vol

    Neal R. M., 1996, Bayesian Learning for Neural Networks, Vol. 118 of Lecture Notes in Statistics. Springer-Verlag

    1996

  24. [33]

    Nesseris S., Garc \' a-Bellido J., 2013, @doi [ ] 10.1088/1475-7516/2013/08/036 , https://ui.adsabs.harvard.edu/abs/2013JCAP...08..036N 2013, 036

    2013 doi

  25. [34]

    Vol.2B: Bayesian inference, 2nd ed.] 10.2307/2291686 , 2B

    O'Hagan A., Forster J., 2004, @doi [Kendall's advanced theory of statistics. Vol.2B: Bayesian inference, 2nd ed.] 10.2307/2291686 , 2B

    2004 doi

  26. [35]

    J., Mohamed S., Lakshminarayanan B., 2021, The Journal of Machine Learning Research, 22, 2617

    Papamakarios G., Nalisnick E., Rezende D. J., Mohamed S., Lakshminarayanan B., 2021, The Journal of Machine Learning Research, 22, 2617

    2021

  27. [36]

    S., Price M

    Piras D., Polanska A., Mancini A. S., Price M. A., McEwen J. D., 2024, @doi [The Open Journal of Astrophysics] 10.33232/001c.123368 , https://ui.adsabs.harvard.edu/abs/2024OJAp....7E..73P 7, 73

    2024 doi

  28. [37]

    A., Spurio Mancini A., McEwen J

    Polanska A., Price M. A., Spurio Mancini A., McEwen J. D., 2023, @doi [Physical Sciences Forum] 10.3390/psf2023009010 , 9

    2023 doi

  29. [38]

    A., Piras D., Spurio Mancini A., McEwen J

    Polanska A., Price M. A., Piras D., Spurio Mancini A., McEwen J. D., 2024a, @doi [arXiv e-prints] 10.48550/arXiv.2405.05969 , https://ui.adsabs.harvard.edu/abs/2024arXiv240505969P p. arXiv:2405.05969

    doi

  30. [39]

    Polanska A., Wouters T., Pang P. T. H., Wong K. W. K., McEwen J. D., 2024b, in Proceedings of the Machine Learning and Physical Sciences Workshop as part of the 38th International Conference on Neural Information Processing Systems. p. arXiv:2410.21076 ( @eprint arXiv 2410.21076 )

    arXiv

  31. [40]

    A., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2405.07504 , https://ui.adsabs.harvard.edu/abs/2024arXiv240507504R p

    Rinaldi S., Demasi G., Del Pozzo W., Hannuksela O. A., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2405.07504 , https://ui.adsabs.harvard.edu/abs/2024arXiv240507504R p. arXiv:2405.07504

    2024 doi arXiv

  32. [41]

    Salvatelli V., Said N., Bruni M., Melchiorri A., Wands D., 2014, @doi [ ] 10.1103/PhysRevLett.113.181301 , https://ui.adsabs.harvard.edu/abs/2014PhRvL.113r1301S 113, 181301

    2014 doi

  33. [42]

    Skilling J., 2006, @doi [Bayesian Analysis] 10.1214/06-BA127 , 1, 833

    2006 doi

  34. [43]

    S., 2020, @doi [ ] 10.1093/mnras/staa278 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.493.3132S 493, 3132

    Speagle J. S., 2020, @doi [ ] 10.1093/mnras/staa278 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.493.3132S 493, 3132

    2020 doi

  35. [44]

    M., Price M

    Spurio Mancini A., Docherty M. M., Price M. A., McEwen J. D., 2023, @doi [RAS Techniques and Instruments] 10.1093/rasti/rzad051 , https://ui.adsabs.harvard.edu/abs/2023RASTI...2..710S 2, 710

    2023 doi

  36. [45]

    D., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2410.10616 , https://ui.adsabs.harvard.edu/abs/2024arXiv241010616S p

    Spurio Mancini A., Lin K., McEwen J. D., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2410.10616 , https://ui.adsabs.harvard.edu/abs/2024arXiv241010616S p. arXiv:2410.10616

    2024 doi arXiv

  37. [46]

    arXiv:2404.12294

    Srinivasan R., Crisostomi M., Trotta R., Barausse E., Breschi M., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2404.12294 , https://ui.adsabs.harvard.edu/abs/2024arXiv240412294S p. arXiv:2404.12294

    2024 doi arXiv

  38. [47]

    Stiskalek R., Desmond H., 2024, @doi [Research Notes of the American Astronomical Society] 10.3847/2515-5172/ad8fb1 , https://ui.adsabs.harvard.edu/abs/2024RNAAS...8..281S 8, 281

    2024 doi

  39. [48]

    J., Bartlett D

    Stiskalek R., Desmond H., Devriendt J., Slyz A., Lavaux G., Hudson M. J., Bartlett D. J., Courtois H. M., 2025, @doi [arXiv e-prints] 10.48550/arXiv.2502.00121 , https://ui.adsabs.harvard.edu/abs/2025arXiv250200121S p. arXiv:2502.00121

    2025 doi

  40. [49]

    The Dark Energy Survey Collaboration 2005, @doi [arXiv e-prints] 10.48550/arXiv.astro-ph/0510346 , https://ui.adsabs.harvard.edu/abs/2005astro.ph.10346T pp astro--ph/0510346

    2005 doi arXiv

  41. [50]

    Trotta R., 2007, @doi [Monthly Notices of the Royal Astronomical Society] 10.1111/j.1365-2966.2007.11738.x , 378, 72

    2007

  42. [51]

    M., Mortlock D

    Verde L., Feeney S. M., Mortlock D. J., Peiris H. V., 2013, @doi [ ] 10.1088/1475-7516/2013/09/013 , https://ui.adsabs.harvard.edu/abs/2013JCAP...09..013V 2013, 013

    2013 doi

  43. [52]

    Verdinelli I., Wasserman L., 1995, Journal of the American Statistical Association, 90, 614

    1995

  44. [53]

    E., The LSST Dark Energy Science Collaboration 2024, @doi [arXiv e-prints] 10.48550/arXiv.2409.17975 , https://ui.adsabs.harvard.edu/abs/2024arXiv240917975Z p

    Zeghal J., Lanzieri D., Lanusse F., Boucaud A., Louppe G., Aubourg E., Bayer A. E., The LSST Dark Energy Science Collaboration 2024, @doi [arXiv e-prints] 10.48550/arXiv.2409.17975 , https://ui.adsabs.harvard.edu/abs/2024arXiv240917975Z p. arXiv:2409.17975

    2024 doi arXiv

  45. [54]

    H., 2025, @doi [ ] 10.1051/0004-6361/202450487 , https://ui.adsabs.harvard.edu/abs/2025A&A...694A.223V 694, A223

    von Wietersheim-Kramsta M., Lin K., Tessore N., Joachimi B., Loureiro A., Reischke R., Wright A. H., 2025, @doi [ ] 10.1051/0004-6361/202450487 , https://ui.adsabs.harvard.edu/abs/2025A&A...694A.223V 694, A223

    2025 doi

Pith tools

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