Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Attention-based Neural Network Emulators for Multi-Probe Data Vectors Part III: Modeling The Next Generation Surveys

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 Attention-based emulators reproduce CAMB's CMB power spectra within cosmic-variance errors out to multipole 5000 with a few hundred thousand training spectra.

desk verdict A solid, honest extension of transformer emulators to CMB power spectra; the headline accuracy claim is slightly overbroad because the hardest part of the parameter volume is cut from the testing set, but the central methodological result holds. read the letter →

arxiv 2505.22574 v2 pith:3IF72MYC submitted 2025-05-28 astro-ph.CO

Yijie Zhu , Evan Saraivanov , Joshua A. Kable , Artemis Sofia Giannakopoulou , Amritpal Nijjar , Vivian Miranda , Marco Bonici , Tim Eifler
show 1 more author
Elisabeth Krause
This is my paper
classification astro-ph.CO MSC 85A4068T07
keywords CMBpowerspectracosmicmicrowavebackgroundneuralnetworkemulatortransformerarchitectureattentionmechanismvarianceCAMBLambda-CDMcosmology
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

The paper sets out to show that an emulator built on self-attention (transformer) layers can reproduce the CMB temperature and polarization power spectra that the Boltzmann code CAMB computes, with errors small enough that no importance-sampling correction is needed when the emulator replaces CAMB in a likelihood analysis. The target is demanding: for a cosmic-variance-limited measurement, the fraction of test cosmologies whose emulated spectra differ from CAMB by more than $\Delta\chi^2 = 0.2$ should stay below 10 percent, out to multipole $\ell = 5000$ and across a wide $\Lambda$CDM parameter volume. The paper finds that dot-product attention does this with roughly $2\times10^5$ to $4\times10^5$ training spectra for each of the four next-generation experimental configurations considered, and that the attention mechanism is the ingredient that suppresses the outlier tail relative to plain multilayer-perceptron architectures. If the claim holds, a costly bottleneck of Markov-chain cosmological inference, namely repeated Boltzmann-code evaluations taking minutes per spectrum, is replaced by a sub-second neural evaluation.

What carries the argument

The central object is the scaled dot-product self-attention block: the multipole vector is split into $N$ channels of length $d$, linearly mapped to query, key, and value matrices $Q = W_Q X$, $K = W_K X$, $V = W_V X$; the block computes $\mathrm{Softmax}(QK^T/\sqrt{d})V$, so each output channel is a weighted mixture of all channels, with weights drawn from the similarity of the query and key vectors. This operation lets the network exploit correlations among different parts of the CMB spectrum, and the paper's comparison to ResMLP-only models attributes the suppression of the outlier tail to this mechanism. Around it, three supporting pieces carry much of the accuracy: dividing the spectra by $A_s e^{-2\tau}$ to remove the dominant amplitude degeneracy; the learned activation $h(x)$ of Eq. (6), which interpolates between linear and sigmoid-gated behavior with trainable parameters; and the $L_4$ loss, the square root of the cosmic-variance $\Delta\chi^2$ computed on the rescaled spectra, which weights outliers linearly instead of quadratically. Tempered Gaussian sampling of the training cosmologies, with the training distribution at temperature $T_{\rm train}=256$ and testing at $T_{\rm test}=128$, provides the wide-but-physical parameter coverage.

What would settle it

Regenerate the validation power spectra with a higher CAMB accuracy setting (AccuracyBoost 2.5 or above) and recompute the outlier fraction for the same trained emulators, restricting to $\log(10^{10}A_s)\leq 3.5$ and $\ell \leq 5000$; if the fraction with $\Delta\chi^2 > 0.2$ rises above 10 percent, the claim as stated is falsified for that configuration.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the scaled dot-product attention mechanism -- the same operation used in language-model transformers, applied here by splitting the power-spectrum vector into 16 channels -- changes the scaling of emulator accuracy with training-set size. For every training-set size tested, the attention-based model yields a lower median $\Delta\chi^2$ and a lower fraction of outlier cosmologies than a ResMLP without attention; the ResMLP would need hundreds of thousands of additional CAMB spectra to reach the same outlier fraction. Combined with three supporting choices -- rescaling the spectra by $A_s e^{-2\tau}$ before training, using the learned activation $h(x)$, and training on the loss $L_4 = \langle\sqrt{\Delta\tilde\chi^2_{XY}}\rangle$ evaluated on the rescaled spectra -- the transformer reaches the target of fewer than 10 percent outliers with $\Delta\chi^2 > 0.2$ for the Planck, Simons Observatory, CMB-S4, and CMB-HD configurations examined. The same conclusion is reached by a 1D convolutional architecture at the largest training sets, and the attention-free transformer performs nearly as well as dot-product attention, while three cheaper approximations to the attention matrix (linear, latent, and locality-sensitive hashing) perform worse.

Load-bearing premise

The accuracy claim rests on treating the reference computations produced by the Boltzmann code CAMB with its numerical-accuracy knob set to 1.5 as the truth; if the emulator learns numerical artifacts that appear at that setting, especially at high $A_s$, its true error against a more accurate computation would be larger than the reported cosmic-variance-level error.

Editorial extensions

If this is right

  • A cosmology pipeline can replace a CAMB evaluation costing about 100 seconds on nine CPU cores with a transformer emulator costing $0.01$--$0.1$ seconds on one core, without an importance-sampling correction step.
  • For Planck-like and Simons Observatory-like analyses, around $2\times10^5$ training spectra suffice; for CMB-S4-like and CMB-HD-like precision, around $4\times10^5$ are needed to hold the outlier fraction below 10 percent.
  • Tempered Gaussian sampling reaches the target with roughly three to five times fewer training spectra than uniform sampling, so the training-data bottleneck can be reduced by using a Fisher-informed, correlated sampling distribution.
  • The outlier suppression is architectural, not just data-driven: dot-product attention and its attention-free variant outperform linear, latent, and locality-sensitive-hashing attention, and attention outperforms a ResMLP of comparable size.
  • The same attention/loss recipe transfers to other probes: a $3\times2$pt weak-lensing and galaxy-clustering emulator reaches the $\Delta\chi^2$-based threshold with $10^5$ training points, and the appendix extends the approach to supernova distances.

Reading between the lines

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

  • Beyond the paper: because the reference spectra are CAMB outputs at AccuracyBoost 1.5, the claimed outlier fractions should be read as relative to that specific numerical target; retraining or retesting against a higher-accuracy setting could shift the required training-set size, especially for $\log(10^{10}A_s)>3.5$ or high multipoles.
  • Beyond the paper: the $\Delta\chi^2=0.2$ threshold is a fixed number taken from prior survey practice; an experiment with much smaller error bars, such as a CMB-HD-class survey, may need a stricter threshold, which would likely push the required training-set size above the quoted $4\times10^5$.
  • Beyond the paper: the demonstrated transfer to weak lensing and clustering suggests that a single attention-based emulator architecture, trained with a rescaled cosmic-variance loss, could serve as a common backend for multi-probe analyses, although the paper does not itself train a combined multi-probe data vector.
  • Beyond the paper: the symbolic-regression damping-tail rescaling in Appendix F gives about a factor of two improvement with only $1.8\times10^4$ training points; adopting such analytic pre-processing as standard could reduce training-data costs further than the paper's headline numbers.
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 manuscript presents an extensive comparison of neural-network emulators for CAMB TT, TE, and EE power spectra over a broad flat-ΛCDM prior, focusing on a dot-product-attention transformer baseline with ResMLP blocks, the H(x) activation, the L4 loss, and an A_s e^{-2τ} rescaling pre-processing step. Accuracy is measured by the median Δχ² and by the outlier fraction with Δχ²>0.2 relative to CAMB, using a cosmic-variance covariance and forecast/real covariances for Planck, Simons Observatory, CMB-S4, and CMB-HD, as well as ACT DR6 chains. The central claim is that with roughly 2×10^5–4×10^5 training points the attention-based emulators achieve outlier fractions below 10% and can therefore replace CAMB in real MCMC analyses without importance-sampling corrections.

Significance. If the claims hold, this is a useful and timely contribution: it provides a systematic architecture comparison, demonstrates that dot-product attention reduces outlier fractions relative to ResMLP-only models at fixed training-set size, evaluates several loss functions and pre-processing choices, and includes validation on Planck and ACT DR6 chains as well as on forecast covariances for next-generation experiments. The interpolation test down to T=1 (Section IV G, Figure 13) and the small outlier fractions on real chains are genuine strengths. The main caveat is that all quantitative validation is referenced to CAMB with AccuracyBoost=1.5, which the paper itself shows has numerical artifacts above cosmic variance in the high-A_s region, and the testing sets exclude part of that region; this tempers the strong claim of a 'maximized volume of applicability' but does not eliminate the practical value of the emulator for posterior volumes near the current fiducial cosmology.

major comments (3)
  1. [Appendix A, Section III A, Table I, Figures 5 and 15] The reference CAMB setting (AccuracyBoost=1.5) is shown in Figure 14 to deviate from AccuracyBoost=1.8 by amounts visible under the cosmic-variance covariance, with the deviations growing as log(10^10 A_s) increases, and Section III A excludes log(10^10 A_s)>3.5 from the testing set for exactly this reason. Since the emulator is trained on AB=1.5 outputs and Appendix A warns that the emulator will attempt to learn these numerical instabilities, the reported outlier fractions are conditional on a reference that is itself not accurate to cosmic variance over the full claimed training volume. The volume-of-applicability claim in the abstract and conclusion should either be re-scoped to the tested region or supported by re-validation with a higher AccuracyBoost setting over the full prior, including the high-A_s region.
  2. [Appendix B, Figure 16] The Planck and ACT DR6 chain validations compare emulator outputs to CAMB outputs generated with the same AB=1.5 setting used for training, so they establish interpolation accuracy against that code version rather than against a more accurate Boltzmann calculation or the true sky. Given Appendix A, the paper should include a direct comparison of the AB=1.5 reference itself against a higher-accuracy CAMB or CLASS setting over the posterior volume of those real-data chains, so that reference error is separated from emulator error and cannot be absorbed into the emulator in a way that biases real-data analyses.
  3. [Section IV B and Section IV G] The headline result that a transformer reaches ê(Δχ²>0.2)<10% with approximately 4×10^5 training points is reported for testing sets drawn at Ttest=128 with the A_s cut discussed above, except that the caption of Figure 5 states Ttest=256. Because the abstract and conclusion quote the 4×10^5 point as the threshold, the testing conditions used for that number should be stated consistently and exactly in the main text, and the corresponding outlier fraction should be quoted under the same conditions.
minor comments (5)
  1. [Figure 5 and Section IV B] The Figure 5 caption says the model was tested with Ttest=256, while the text in Section IV and Figure 6 use Ttest=128; please reconcile this inconsistency.
  2. [Section III A and Table I] The text says a stringent cut of log(10^10 A_s)<3.5 is applied to the testing set, but Table I shows the Gaussian test set extends to log(10^10 A_s)=4.0 and only the uniform test set is cut at 3.5; please clarify which cut applies to each sampling scheme.
  3. [Appendix B, Figure 15] For the CMB-HD-like forecast, the validation is restricted to ℓ≤5000 because the emulator is trained only to ℓ=5000, whereas the experiment is planned to reach ℓ~20000; the abstract and main text should state that the CMB-HD claim applies only to the emulated multipole range.
  4. [Appendix F and Section IV A] The damping-tail pre-processing test is presented as preliminary, using only three free parameters, a ResMLP architecture, and 1.8×10^4 training points; this should be flagged clearly wherever it is cited in the main text so it is not read as part of the baseline transformer validation.
  5. [General] Please add a data and code availability statement for the trained models, training sets, and analysis scripts, as the reproducibility of the numerical comparisons would be greatly enhanced by releasing them.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the emulator is a supervised fit to CAMB and is validated on held-out cosmologies, with a reference-accuracy caveat in Appendix A.

full rationale

The paper is an empirical supervised-learning study: the emulator is trained on CAMB-generated power spectra and validated on held-out CAMB cosmologies. The central accuracy metric, Eq. (7), is defined relative to CAMB, and the L4 training loss is a per-spectrum variant of the same cosmic-variance-weighted Delta-chi-squared. This creates a self-referential benchmark, but it is not circular, because the testing cosmologies are never used in training; the reported <10% outlier fractions measure out-of-sample generalization, which is exactly the claim being made. The architecture comparisons (transformer vs ResMLP, dot-product vs alternative attention mechanisms, H(x) vs Tanh, L2/L3/L4/L5) are controlled experiments on the same held-out distribution, so the headline 'attention reduces outliers' is an empirical finding rather than an assumption imported from prior work. Citations to Parts I and II ([20,21]) supply the attention mechanism, tempered-Gaussian sampling procedure, and Tanh baseline, but the paper re-evaluates these choices and sometimes overturns them (H(x) outperforms Tanh; dot-product attention outperforms linear, latent, and LSH variants), so the self-citations are not load-bearing. The genuine caveat is Appendix A: CAMB with AccuracyBoost=1.5 shows numerical inaccuracies visible under cosmic-variance covariance for log(10^10 A_s)>3.5, and Section III A imposes a test cut at log(10^10 A_s)<3.5 accordingly. This restricts the demonstrated 'volume of applicability' and means the precision claims are conditional on the accuracy of the reference code, but this is a benchmark-accuracy limitation rather than a circular derivation: the paper explicitly defines its task as emulating CAMB, not as deriving the spectra from first principles. The ACT DR6 and Planck-chain tests provide some external anchoring, though the ACT Delta-chi-squared comparison is still CAMB-referenced. Overall, no step reduces to its own input by construction.

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

The paper introduces no new physics entities. Its free parameters are primarily network hyperparameters and preprocessing choices. The most consequential assumptions are the treatment of CAMB with AccuracyBoost=1.5 as truth and the use of a single fiducial cosmology for the covariance matrix that defines emulator accuracy.

free parameters (4)
  • Gaussian sampling temperature T = 256
    Chosen manually to define training set width; affects parameter coverage and concentrates samples near the fiducial cosmology.
  • Number of principal components for ResMLP = 96
    Retained PCA modes for the ResMLP architecture; chosen as sufficient to reach the accuracy threshold, not derived from first principles.
  • Damping tail fit parameters alpha1-alpha15 = Listed in Table IX
    Learned via genetic algorithm on CAMB outputs in Appendix F; used only in a preliminary test, not in the main baseline emulator.
  • Hyperbolic loss function weights (in L3/L5) = N/A
    The functional forms of the loss functions L3 and L5 are chosen heuristically to damp outlier penalties; they are not derived from a statistical principle.
assumptions (3)
  • domain assumption CAMB outputs with AccuracyBoost=1.5 are treated as ground truth
    The emulator is trained to match these outputs; numerical inaccuracies in CAMB at high A_s are noted in Appendix A.
  • domain assumption Cosmic variance covariance is diagonal and Gaussian with fsky=1
    Used to define Delta chi^2 and the loss functions; non-Gaussian at ell<=30 is acknowledged but ignored.
  • ad hoc to paper Tempered Gaussian sampling with T=256 covers the relevant parameter space
    The training distribution is constructed from a Fisher forecast and expanded by temperature; not derived from a physical prior.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Attention-based Neural Network Emulators for Multi-Probe Data Vectors Part III: Modeling The Next Generation Surveys." pith.science (2026). https://pith.science/paper/3IF72MYC

@misc{pith2026250522574,
  author       = {Pith},
  title        = {Pith review of: Attention-based Neural Network Emulators for Multi-Probe Data Vectors Part III: Modeling The Next Generation Surveys},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3IF72MYC}},
  note         = {Machine review of arXiv:2505.22574}
}
abstract

Machine learning can accelerate cosmological inferences that involve many sequential evaluations of computationally expensive data vectors. Previous works in this series have examined how machine learning architectures impact emulator accuracy and training time for optical shear and galaxy clustering 2-point function. In this final manuscript, we explore neural network performance when emulating Cosmic Microwave Background temperature and polarization power spectra. We maximize the volume of applicability in the parameter space of our emulators within the standard $\Lambda$-cold-dark-matter model while ensuring that errors are below cosmic variance. Relative to standard multi-layer perceptron architectures, we find the dot-product-attention mechanism reduces the number of outliers among testing cosmologies, defined as the fraction of testing points with $\Delta \chi^2 > 0.2$ relative to \textsc{CAMB} outputs, for a wide range of training set sizes. Such precision enables attention-based emulators to be directly applied to real data without requiring any additional correction via importance sampling. Combined with pre-processing techniques and optimized activation and loss functions, attention-based models can meet the precision criteria set by current and future CMB and lensing experiments. For each of Planck, Simons Observatory, CMB S4, and CMB HD, we find the fraction of outlier points to be less than $10\%$ with around $2\times10^5$ to $4\times10^5$ training data vectors. We further explore the applications of these methods to supernova distance, weak lensing, and galaxy clustering, as well as alternative architectures and pre-processing techniques.

Figures

Figures reproduced from arXiv: 2505.22574 by the authors.

Figure 1
Figure 1. FIG. 1. The baseline transformer architecture used in this [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Comparison between emulator outputs and [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Comparison between emulator outputs and [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Distributions of the ratio between absolute error of [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. The median ∆ [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: shows the distribution of ∆χ 2 values between the CAMB and an emulator that either includes or ex￾cludes a self-attention layer for all of the points in the testing set. The ∆χ 2 distribution for the transformer￾based architecture is shifted toward smaller ∆χ 2 values …
Figure 8
Figure 8. Figure 8: FIG. 8. A comparison of the fraction of outliers, [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 10
Figure 10. Figure 10: FIG. 10. The difference between the TT power spectrum out [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: FIG. 11 [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 13
Figure 13. Figure 13: FIG. 13. The fraction of outliers, [PITH_FULL_IMAGE:figures/full_fig_p016_13.png]
Figure 14
Figure 14. Figure 14: FIG. 14. Distribution of log (∆ [PITH_FULL_IMAGE:figures/full_fig_p017_14.png]
Figure 15
Figure 15. Figure 15: FIG. 15. Comparison of the fraction of outlier points, [PITH_FULL_IMAGE:figures/full_fig_p018_15.png]
Figure 16
Figure 16. Figure 16: FIG. 16. ∆ [PITH_FULL_IMAGE:figures/full_fig_p018_16.png]
Figure 17
Figure 17. Figure 17: FIG. 17. Histograms displaying the ∆ [PITH_FULL_IMAGE:figures/full_fig_p019_17.png]
Figure 18
Figure 18. Figure 18: FIG. 18. The projected spatial distribution of ∆ [PITH_FULL_IMAGE:figures/full_fig_p020_18.png]
Figure 20
Figure 20. Figure 20: FIG. 20. Heatmap of ∆ [PITH_FULL_IMAGE:figures/full_fig_p021_20.png]
Figure 22
Figure 22. Figure 22: FIG. 22. A comparison of the accuracy of a transformer and [PITH_FULL_IMAGE:figures/full_fig_p022_22.png]
Figure 21
Figure 21. Figure 21: FIG. 21. Architecture of the 1D convolutional neural network [PITH_FULL_IMAGE:figures/full_fig_p022_21.png]
Figure 23
Figure 23. Figure 23: FIG. 23. This plot shows the distributions of the ratio be [PITH_FULL_IMAGE:figures/full_fig_p023_23.png]
Figure 24
Figure 24. Figure 24: FIG. 24. This plot shows the comparison of the lensing tail [PITH_FULL_IMAGE:figures/full_fig_p025_24.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Floquet Dissipative Phase Transitions

    quant-ph 2026-03 unverdicted novelty 6.0 of 10

    Dissipative phase transitions in Floquet open systems are defined through the Floquet propagator spectrum, with applications to driven Kerr resonators and the quantum Rabi model.

Reference graph

Works this paper leans on

55 extracted references · 10 canonical work pages · cited by 1 Pith paper

  1. [1]

    Ade et al

    P. Ade et al. (Simons Observatory), JCAP 02, 056 (2019), arXiv:1808.07445 [astro-ph.CO]

    arXiv 2019

  2. [2]

    Abitbol et al

    M. Abitbol et al. (Simons Observatory), (2025), arXiv:2503.00636 [astro-ph.IM]

    arXiv 2025

  3. [3]

    Abazajian, A

    K. Abazajian, A. Abdulghafour, G. E. Addison, P. Ad- shead, Z. Ahmed, M. Ajello, D. Akerib, S. W. Allen, D. Alonso, M. Alvarez, M. A. Amin, M. Amiri, A. An- derson, B. Ansarinejad, M. Archipley, K. S. Arnold, M. Ashby, H. Aung, C. Baccigalupi, C. Baker, A. Bak- shi, D. Bard, D. Barkats, D. Barron, P. S. Barry, J. G. Bartlett, P. Barton, R. Basu Thakur, N....

    arXiv 2022

  4. [4]

    Sehgal et al., (2019), arXiv:1906.10134 [astro-ph.CO]

    N. Sehgal et al., (2019), arXiv:1906.10134 [astro-ph.CO]

    arXiv 2019

  5. [5]

    Lewis, A

    A. Lewis, A. Challinor, and A. Lasenby, ApJ 538, 473 (2000), arXiv:astro-ph/9911177 [astro-ph]

    arXiv 2000

  6. [6]

    Lesgourgues, arXiv e-prints , arXiv:1104.2932 (2011), arXiv:1104.2932 [astro-ph.IM]

    J. Lesgourgues, arXiv e-prints , arXiv:1104.2932 (2011), arXiv:1104.2932 [astro-ph.IM]

    arXiv 2011

  7. [7]

    D. Blas, J. Lesgourgues, and T. Tram, J. Cosmology As- tropart. Phys. 2011, 034 (2011), arXiv:1104.2933 [astro- ph.CO]

    arXiv 2011

  8. [8]

    J. a. Rebou¸ cas, D. H. F. de Souza, K. Zhong, V. Mi- randa, and R. Rosenfeld, JCAP 02, 024 (2025), arXiv:2408.14628 [astro-ph.CO]

    arXiv 2025

Show all 55 references
  1. [9]

    Hajian, Phys

    A. Hajian, Phys. Rev. D 75, 083525 (2007), arXiv:astro- ph/0608679 26

    2007

  2. [10]

    Bonici, F

    M. Bonici, F. Bianchini, and J. Ruiz-Zapatero, (2023), 10.21105/astro.2307.14339, arXiv:2307.14339 [astro-ph.CO]

    2023 arXiv

  3. [11]

    Ge et al

    F. Ge et al. (SPT-3G), (2024), arXiv:2411.06000 [astro- ph.CO]

    2024 arXiv

  4. [12]

    Piras and A

    D. Piras and A. Spurio Mancini, (2023), 10.21105/as- tro.2305.06347, arXiv:2305.06347 [astro-ph.CO]

    2023 arXiv

  5. [13]

    J. U. Lange, Mon. Not. Roy. Astron. Soc. 525, 3181 (2023), arXiv:2306.16923 [astro-ph.IM]

    2023 arXiv

  6. [14]

    Feroz, M

    F. Feroz, M. P. Hobson, and M. Bridges, Mon. Not. Roy. Astron. Soc. 398, 1601 (2009), arXiv:0809.3437 [astro- ph]

    2009 arXiv

  7. [15]

    Kamionkowski, Phys

    M. Kamionkowski, Phys. Rev. D 104, 063512 (2021), arXiv:2105.02887 [astro-ph.CO]

    2021 arXiv

  8. [16]

    Spurio Mancini, D

    A. Spurio Mancini, D. Piras, J. Alsing, B. Joachimi, and M. P. Hobson, MNRAS 511, 1771 (2022), arXiv:2106.03846 [astro-ph.CO]

    2022 arXiv

  9. [17]

    H. T. Jense, I. Harrison, E. Calabrese, A. Spurio Mancini, B. Bolliet, J. Dunkley, and J. C. Hill, RAS Tech. In- strum. 4, rzaf002 (2025), arXiv:2405.07903 [astro-ph.CO]

    2025 arXiv

  10. [18]

    G.-J. Wang, C. Cheng, Y.-Z. Ma, J.-Q. Xia, A. Abebe, and A. Beesham, Astrophys. J. Suppl. 268, 7 (2023), arXiv:2306.11102 [astro-ph.CO]

    2023 arXiv

  11. [19]

    Bolliet, A

    B. Bolliet, A. Spurio Mancini, J. C. Hill, M. Mad- havacheril, H. T. Jense, E. Calabrese, and J. Dunk- ley, Mon. Not. Roy. Astron. Soc. 531, 1351 (2024), arXiv:2303.01591 [astro-ph.CO]

    2024 arXiv

  12. [20]

    Zhong, E

    K. Zhong, E. Saraivanov, J. Caputi, V. Miranda, S. S. Boruah, T. Eifler, and E. Krause, (2024), arXiv:2402.17716 [astro-ph.CO]

    2024 arXiv

  13. [21]

    Saraivanov, K

    E. Saraivanov, K. Zhong, V. Miranda, S. S. Boruah, T. Eifler, and E. Krause, (2024), arXiv:2403.12337 [astro-ph.CO]

    2024 arXiv

  14. [22]

    Krause, X

    E. Krause, X. Fang, S. Pandey, L. F. Secco, O. Alves, H. Huang, J. Blazek, J. Prat, J. Zuntz, T. F. Eifler, N. MacCrann, J. DeRose, M. Crocce, A. Porredon, B. Jain, M. A. Troxel, S. Dodelson, D. Huterer, A. R. Lid- dle, C. D. Leonard, A. Amon, A. Chen, J. Elvin-Poole, A. Fert´...

    2021 arXiv

  15. [23]

    Bengio, P

    Y. Bengio, P. Simard, and P. Frasconi, IEEE Transac- tions on Neural Networks 5, 157 (1994)

    1994

  16. [24]

    Pascanu, T

    R. Pascanu, T. Mikolov, and Y. Bengio, CoRR abs/1211.5063 (2012), 1211.5063

    2012 arXiv

  17. [25]

    K. He, X. Zhang, S. Ren, and J. Sun, arXiv e-prints , arXiv:1512.03385 (2015), arXiv:1512.03385 [cs.CV]

    2015 arXiv

  18. [26]

    Attention Is All You Need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention Is All You Need,” (2017), arXiv:1706.03762 [cs]

    2017 arXiv

  19. [27]

    Katharopoulos, A

    A. Katharopoulos, A. Vyas, N. Pappas, and F. Fleuret, in Proceedings of the International Conference on Ma- chine Learning (ICML)(2020)

    2020

  20. [28]

    A. Vyas, A. Katharopoulos, and F. Fleuret, Proceedings of the International Conference on Neural Information Processing Systems (NeurIPS), (2020)

    2020

  21. [29]

    R. Li, J. Su, C. Duan, and S. Zheng, arXiv e-prints , arXiv:2007.14902 (2020), arXiv:2007.14902 [cs.CV]

    2020 arXiv

  22. [30]

    Aghanim, Y

    Planck Collaboration, N. Aghanim, Y. Akrami, M. Ash- down, J. Aumont, C. Baccigalupi, M. Ballardini, A. J. Banday, R. B. Barreiro, N. Bartolo, S. Basak, R. Battye, K. Benabed, J. P. Bernard, M. Bersanelli, P. Bielewicz, J. J. Bock, J. R. Bond, J. Borrill, F. R. Bouchet, F. Bou...

    2020 arXiv

  23. [31]

    Dolga, L

    R. Dolga, L. Maystre, M. Cobzarenco, and D. Barber, arXiv e-prints , arXiv:2402.17512 (2024), arXiv:2402.17512 [cs.CL]

    2024

  24. [32]

    Kitaev, L

    N. Kitaev, L. Kaiser, and A. Levskaya, arXiv e-prints , arXiv:2001.04451 (2020), arXiv:2001.04451 [cs.LG]

    2020 arXiv

  25. [33]

    S. Zhai, W. Talbott, N. Srivastava, C. Huang, H. Goh, R. Zhang, and J. Susskind, arXiv e-prints , arXiv:2105.14103 (2021), arXiv:2105.14103 [cs.LG]

    2021 arXiv

  26. [34]

    Alsing, H

    J. Alsing, H. Peiris, J. Leja, C. Hahn, R. Tojeiro, D. Mortlock, B. Leistedt, B. D. Johnson, and C. Conroy, ApJS 249, 5 (2020), arXiv:1911.11778 [astro-ph.IM]

    2020 arXiv

  27. [35]

    Nygaard, E

    A. Nygaard, E. B. Holm, S. Hannestad, and T. Tram, J. Cosmology Astropart. Phys. 2023, 025 (2023), arXiv:2205.15726 [astro-ph.IM]

    2023 arXiv

  28. [36]

    J. R. Bond, A. H. Jaffe, and L. E. Knox, Astrophys. J. 533, 19 (2000), arXiv:astro-ph/9808264

    2000 arXiv

  29. [37]

    H. K. Eriksen, J. B. Jewell, C. Dickinson, A. J. Banday, K. M. G´ orski, and C. R. Lawrence, ApJ676, 10 (2008), arXiv:0709.1058 [astro-ph]

    2008 arXiv

  30. [38]

    Aghanim et al

    N. Aghanim et al. (Planck), Astron. Astrophys. 641, A5 (2020), arXiv:1907.12875 [astro-ph.CO]

    2020 arXiv

  31. [39]

    Prince and J

    H. Prince and J. Dunkley, Phys. Rev. D 105, 023518 (2022), arXiv:2104.05715 [astro-ph.CO]

    2022 arXiv

  32. [40]

    Devon Lin and B

    C. Devon Lin and B. Tang, arXiv e-prints , arXiv:2203.06334 (2022), arXiv:2203.06334 [stat.ME]

    2022 arXiv

  33. [41]

    Hu and M

    W. Hu and M. J. White, Astrophys. J. 479, 568 (1997), arXiv:astro-ph/9609079

    1997 arXiv

  34. [42]

    Adamo, H.-J

    J. Adamo, H.-J. Huang, and T. Eifler, Phys. Rev. D 110, 123517 (2024), arXiv:2405.00125 [astro-ph.CO]

    2024 arXiv

  35. [43]

    J. Su, Y. Lu, S. Pan, A. Murtadha, B. Wen, and Y. Liu, arXiv e-prints , arXiv:2104.09864 (2021), arXiv:2104.09864 [cs.CL]

    2021 arXiv

  36. [44]

    Calabrese et al

    E. Calabrese et al. (ACT), (2025), arXiv:2503.14454 [astro-ph.CO]

    2025 arXiv

  37. [45]

    MacInnis, N

    A. MacInnis, N. Sehgal, and M. Rothermel, arXiv e-prints , arXiv:2309.03021 (2023), arXiv:2309.03021 [astro-ph.CO]

    2023 arXiv

  38. [46]

    Lesgourgues, arXiv e-prints , arXiv:1104.2934 (2011), arXiv:1104.2934 [astro-ph.CO]

    J. Lesgourgues, arXiv e-prints , arXiv:1104.2934 (2011), arXiv:1104.2934 [astro-ph.CO]

    2011 arXiv

  39. [47]

    Aghanim, Y

    Planck Collaboration, N. Aghanim, Y. Akrami, F. Ar- roja, M. Ashdown, J. Aumont, C. Baccigalupi, M. Ballar- dini, A. J. Banday, R. B. Barreiro, N. Bartolo, S. Basak, R. Battye, K. Benabed, J. P. Bernard, M. Bersanelli, P. Bielewicz, J. J. Bock, J. R. Bond, J. Borrill, F. R. Bo...

    2020 arXiv

  40. [48]

    M. J. Mortonson, W. Hu, and D. Huterer, Phys. Rev. D 79, 023004 (2009), arXiv:0810.1744 [astro-ph]

    2009 arXiv

  41. [49]

    Kiranyaz, O

    S. Kiranyaz, O. Avci, O. Abdeljaber, T. Ince, M. Gab- bouj, and D. J. Inman, arXiv e-prints , arXiv:1905.03554 (2019), arXiv:1905.03554 [eess.SP]

    2019 arXiv

  42. [50]

    Pandya, Y

    S. Pandya, Y. Yang, N. Van Alfen, J. Blazek, and R. Walters, (2025), arXiv:2504.05235 [astro-ph.CO]

    2025

  43. [51]

    J. R. Koza, Genetic programming: On the programming of computers by means of natural selection (The MIT Press, 1992)

    1992

  44. [52]

    Analytical emulator for the linear matter power spec- trum from physics-informed machine learning,

    J. B. Orjuela-Quintana, D. Sapone, and S. Nesseris, “Analytical emulator for the linear matter power spec- trum from physics-informed machine learning,” (2024), arXiv:2407.16640 [astro-ph.CO]

    2024

  45. [53]

    D. J. Bartlett, B. D. Wandelt, M. Zennaro, P. G. Ferreira, and H. Desmond, Astron. Astrophys. 686, A150 (2024), arXiv:2402.17492 [astro-ph.CO]

    2024 arXiv

  46. [54]

    J. B. Orjuela-Quintana, S. Nesseris, and W. Car- dona, Physical Review D 107 (2023), 10.1103/phys- revd.107.083520

    2023 doi

  47. [55]

    Interpretable machine learning for sci- ence with pysr and symbolicregression.jl,

    M. Cranmer, “Interpretable machine learning for sci- ence with pysr and symbolicregression.jl,” (2023), arXiv:2305.01582 [astro-ph.IM]

    2023 arXiv

Pith tools

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