Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

PatchNet shows that a hybrid hierarchical summary can extract nearly all cosmological information from a (1 Gpc/h)^3 dark matter field at 7.8 Mpc/h resolution.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Combining patch-level neural summaries with power spectrum and bispectrum extracts roughly as much cosmological information from dark matter simulations as wavelet statistics, apparently nearing the information limit at 7.8 Mpc/h resolution.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A sensible hybrid that adds patch-level CNN summaries to P(k)+B(k), shows real gains, and roughly matches WST; the main caveat is the unvalidated Gaussian-likelihood Fisher formula behind the 'near-optimal' claim. the 3 major comments →

arxiv 2509.03165 v1 pith:LQLFONGG submitted 2025-09-03 astro-ph.CO cs.ITmath.IT

PatchNet: A hierarchical approach for neural field-level inference from Quijote Simulations

Anirban Bairagi , Benjamin Wandelt This is my paper
classification astro-ph.CO cs.ITmath.IT
keywords simulation-based inferencefield-level inferenceFisher informationdark matter density fieldpower spectrum and bispectrumwavelet scattering transformPatchNetQuijote simulations
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 asks how much cosmological information sits in a cubic gigaparsec of dark matter at z=0 on a 7.8 Mpc/h grid. It argues that no single standard summary captures it: the power spectrum and bispectrum are sufficient at large scales but miss nonlinear small-scale structure, while a neural network trained on whole large fields is data-hungry and underperforms. The paper's proposal, PatchNet, is a hierarchy: keep P(k) and B(k) for the full volume, and add a compact 3D CNN summary trained on many overlapping 125 Mpc/h subvolumes, which need only small simulations and a small network. The paper shows this combined summary contains more Fisher information than P(k)+B(k) alone and matches wavelet scattering transform information. The paper reads the convergence of two very different summaries as evidence that the combination approaches the full information content of the field at that resolution.

Core claim

The central claim is that a hybrid summary consisting of the full-volume power spectrum, the full-volume bispectrum, and a mean-aggregated neural summary of small subvolumes (patches) extracts nearly all of the cosmological information available in a (1 Gpc/h)^3 dark matter field sampled at 128^3 voxels. PatchNet, the patch-based 3D CNN, is trained on 125 Mpc/h patches drawn from Quijote dark matter fields, and its outputs are combined with P(k) and B(k). The paper's Fisher information analysis shows that this combined summary substantially outperforms P(k)+B(k) alone and matches the information content of an overcomplete wavelet scattering transform. Because two very different summaries giv

What carries the argument

The load-bearing object is the hierarchical summary statistic: concatenate (i) the power spectrum and bispectrum of the full simulation volume, which are known to be sufficient in the linear and quasi-linear regimes, with (ii) the mean over many patches of a 3D CNN's predicted parameters from 16^3 subvolumes of side 125 Mpc/h. To avoid losing clustering signal that straddles patch boundaries, the field is covered eight times by shifting the patch grid by half a patch width along axes and diagonals. Information is measured with the Gaussian Fisher matrix F = ∇θμ^T C^{-1} ∇θμ, using 5,000 fiducial simulations for the covariance and 500 finite-difference simulation pairs for the derivatives.

Load-bearing premise

The argument assumes the Gaussian Fisher formula with a fixed covariance correctly measures information for every summary, including neural patch outputs and wavelet coefficients, even though the paper never tests that assumption for these non-Gaussian summaries; the 'full information' reading also assumes the two converging methods do not share the same information loss.

What would settle it

Compute the actual posterior for P(k)+B(k)+patches on the Quijote Latin Hypercube fields with a likelihood-free estimator and compare 68% credible intervals against the Fisher forecast; if coverage departs strongly from the forecast, the Gaussian-covariance assumption is violated and the Fisher-based 'full information' claim needs qualification. A second check: correlate PatchNet and wavelet estimator residuals against true parameters—if they share the same failure modes, their agreement is not independent evidence of saturation.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • For a fixed voxel size, scaling PatchNet to survey-size volumes is limited by computing P(k) and B(k), not by GPU memory for the neural network.
  • Patch-based training multiplies the effective training set: one simulation yields many 125 Mpc/h patches, so fewer full-volume simulations are needed than for full-field CNNs.
  • Because each patch yields its own parameter estimate, the same framework can be applied to redshift slices or light-cone data, and further to making spatial maps of cosmological parameters.
  • The information in P(k)+B(k)+patches equals that of wavelet scattering coefficients; if both saturate the field's information, then the hybrid summary is a practical near-optimal compression for nonlinear dark matter fields.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A natural reading is that information saturation is real, but the inference depends on PatchNet and wavelet coefficients not sharing the same information loss; the paper does not test this directly.
  • A testable extension: train a likelihood-free posterior on the hybrid summary and compare its credible intervals with the Fisher forecast; disagreement would signal that the Gaussian-likelihood assumption in Eq. 3.4 is doing work.
  • The same hierarchy could be applied to halo/galaxy fields or to redshift-dependent patches; if the information bound holds for biased tracers, it would give a route to field-level inference on real survey data.
  • One could vary the patch size to map information as a function of scale; the current 125 Mpc/h choice is motivated by perturbation theory, but the optimal patch size may depend on the parameter and cosmology.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper introduces PatchNet, a hierarchical simulation-based inference approach that combines large-scale analytical summaries (power spectrum P(k) and bispectrum B(k)) with neural-network summaries of small-scale subvolumes (patches) of the dark matter density field. Using Quijote simulations, the authors compute Fisher information matrices via the Gaussian-likelihood formula (Eq. 3.4), with covariance estimated from 5,000 fiducial simulations and derivatives from 500 finite-difference pairs. They compare P(k), P(k)+B(k), a full-field CNN, PatchNet, and wavelet scattering transform (WST) summaries. The central empirical findings are that PatchNet alone improves on P(k)+B(k), and that P(k)+B(k)+patches matches or exceeds the Fisher information of WST. The paper interprets this cross-method agreement as evidence that the hybrid summary approaches the full information content of the (1 Gpc/h)^3 dark matter field at 7.8 Mpc/h resolution. The work is explicitly conceptual, using idealized dark matter fields, and frames the result as a lower bound on the extractable information.

Significance. If the Fisher estimates are reliable, the paper offers a practical and computationally efficient route to field-level cosmological inference at survey scale, circumventing GPU memory limits and increasing effective training-data volume. The comparison against WST as an independent benchmark is a valuable check, and the use of the large Quijote simulation suite for covariance and derivative estimation is a strength. However, the quantitative claims rest on two load-bearing assumptions that are not validated: the applicability of the Gaussian-likelihood Fisher formula to highly non-Gaussian summaries, and the interpretation of agreement between two methods as evidence of information saturation. The paper is candid in Section 7 that the result is a lower bound, but the abstract's 'evidence that we are estimating the full information content' is stronger than the tests currently support. With additional validation and error estimates, the manuscript could make a solid contribution.

major comments (3)
  1. [§3, Eq. (3.4)] The Gaussian-likelihood Fisher formula is applied to every summary, including the neural patch outputs and WST coefficients. Section 3 states 'As long as the conditions leading to Eq. 3.4 are satisfied,' but this condition is never tested. For non-Gaussian summaries, Eq. 3.4 is the Fisher information of the best Gaussian approximation, not of the actual summary, and can either over- or under-estimate the true information. In addition, Eq. 3.3 contains a log|C| term, so if the covariance is parameter-dependent, Eq. 3.4 omits a contribution; the paper estimates C only at the fiducial cosmology. This directly affects all Fisher numbers in Figs. 5 and 6 and the 'matches WST' claim. Please add validation, e.g., normality diagnostics on the summaries, a simulation-based likelihood estimate of the FIM (as in the cited Coulton & Wandelt method), or at least an estimate of the covariance-derivati
  2. [§6, Fig. 6] The paper reports point estimates of Fisher information and marginalized contours without uncertainties. The covariance uses 5,000 fiducial simulations and the derivatives use 500 finite-difference pairs, so the FIM estimates have sampling noise; the claim that PatchNet and WST 'match' is a comparison of two noisy point estimates. Please provide error bars on the Fisher information or the marginalized uncertainties, for example via jackknife over simulations or analytic estimates of the FIM covariance. Also report the finite-difference step sizes used and, if the inverse covariance is not debiased, state whether a Hartlap-type correction is applied.
  3. [§6 and Abstract] Interpreting the agreement between PatchNet and WST as evidence of approaching the full information content is an assumption, not a derived consequence. The paper itself acknowledges in Section 6 that 'we cannot prove that we have reached the information limit' and in Section 7 that the result is a lower bound, yet the abstract states that the agreement provides 'evidence that we are estimating the full information content.' Two different summaries can share the same information loss (e.g., both may be insensitive to certain phase correlations at the grid resolution). Please soften the claim to 'consistent with' saturation, or provide a concrete test that the two methods are not losing the same information, such as adding a third independent summary or comparing against a known information bound in a simplified setting.
minor comments (5)
  1. [§5.2.3 and Fig. 5] '83 discrete patches' should be 8^3 = 512 patches. It would also help to explicitly state that 8^3 patches of 16^3 voxels tile the 128^3 volume.
  2. [§5.2.3] The sentence 'Once trained for 5 parameters we use the mean value of the target parameters (θ : {Ωm, σ8}) from patches as our predicted output' is ambiguous. Clarify whether the PatchNet summary vector is 2-dimensional or 5-dimensional, and whether the other parameter outputs are discarded or included in the Fisher analysis.
  3. [§7] Typo: 'transfrom' should be 'transform'. The phrase 'at least no excessively loose' is awkward and should be rephrased.
  4. [Appendix A] The WST redundancy criterion removes one coefficient from each pair with |r| > 0.99. Specify the rule for choosing which member is removed and whether the 0.90/0.99 test fully covers the sensitivity to this choice.
  5. [General] For reproducibility, please provide a data/code availability statement and specify exactly which Quijote subsamples and finite-difference step sizes are used for the derivative estimates.

Circularity Check

0 steps flagged

No significant circularity: PatchNet's Fisher information is measured from network output statistics and benchmarked against independent WST; self-citations are ancillary.

full rationale

The paper's central claim—that P(k)+B(k)+patches approaches the information content of the dark matter field—is supported by externally measured Fisher matrices, not by construction. Section 3 computes F from 5,000 fiducial simulations for the covariance and 500 finite-difference pairs for the derivatives (Eq. 3.4), so the PatchNet summary's information is an empirical statistic of the network output. The comparison to WST is an independent benchmark (Section 4.2, Appendix A), and the authors explicitly qualify the saturation interpretation with 'While we cannot prove that we have reached the information limit' and 'This will be a lower bound since we cannot exclude that our inference approach is still somewhat suboptimal' (Sections 6-7). The only author-self citations ([71] on required training-set size and [78] FishNet aggregation) are not load-bearing: [71] merely contextualizes the full-field CNN's underperformance, and [78] was tried and abandoned in favor of mean aggregation. The main caveat—Eq. (3.4)'s Gaussian-likelihood assumption is stated but not validated for non-Gaussian neural and WST summaries ('As long as the conditions leading to Eq. 3.4 are satisfied')—is a statistical robustness concern, not a circularity: it does not make the claimed information content equal to an input by definition, nor does it rename a fitted parameter as a prediction.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

PatchNet introduces a method, not a new physical entity. The central claim rests on hand-chosen scale parameters, unverified statistical assumptions, and an interpretive convergence argument, rather than on new particles, forces, or dimensions.

free parameters (4)
  • Patch size = 125 Mpc/h (k_patch = 0.05 h/Mpc)
    Chosen by hand in Sec 5.2.3 to balance computational feasibility against reaching perturbative scales; controls what small-scale information the network sees.
  • Field and patch resolution = 128^3 full field, 16^3 patches, 7.8 Mpc/h voxels
    Data resolution limits all summaries and defines the resolution at which 'full information' is claimed.
  • WST correlation cutoff = |r| <= 0.99 (robustness check at 0.90)
    Appendix A uses this threshold to reduce 985 WST coefficients to 246; the choice affects the WST benchmark Fisher information.
  • PatchNet architecture and hyperparameters = 3 conv blocks with 3x3x3 kernels and pooling, 4 FC layers, batch 32, 64 patches per realization, 500 epochs, lr 0.001
    Selected after roughly 400-500 architecture and training combinations in Appendix B; the neural summary quality depends on these choices.
axioms (4)
  • domain assumption Fisher information via Gaussian likelihood with parameter-independent covariance (Eq. 3.4) applies to P(k), B(k), WST, and neural summaries.
    The paper states this condition in Sec 3 but does not verify it for non-Gaussian neural or wavelet summaries.
  • domain assumption Quijote dark matter simulations faithfully represent the non-linear density field at 128^3 resolution.
    All claims are relative to these idealized simulations, as stated in Sec 2 and Sec 7.
  • domain assumption The neural network trained on Latin Hypercube simulations generalizes to the fiducial finite-difference simulations used for Fisher derivatives.
    No explicit transfer or calibration check is reported in Sec 5.2 or Sec 6.
  • ad hoc to paper Agreement between PatchNet and WST Fisher information indicates that the full information content has been approached.
    This interpretive step in Sec 6 is the bridge from measured summaries to the 'full information' conclusion; both methods could share the same information loss.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of PatchNet: A hierarchical approach for neural field-level inference from Quijote Simulations." pith.science (2026). https://pith.science/paper/LQLFONGG

@misc{pith2026250903165,
  author       = {Pith},
  title        = {Pith review of: PatchNet: A hierarchical approach for neural field-level inference from Quijote Simulations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LQLFONGG}},
  note         = {Machine review of arXiv:2509.03165}
}
Share X Bluesky LinkedIn Reddit HN
abstract

\textit{What is the cosmological information content of a cubic Gigaparsec of dark matter? } Extracting cosmological information from the non-linear matter distribution has high potential to tighten parameter constraints in the era of next-generation surveys such as Euclid, DESI, and the Vera Rubin Observatory. Traditional approaches relying on summary statistics like the power spectrum and bispectrum, though analytically tractable, fail to capture the full non-Gaussian and non-linear structure of the density field. Simulation-Based Inference (SBI) provides a powerful alternative by learning directly from forward-modeled simulations. In this work, we apply SBI to the \textit{Quijote} dark matter simulations and introduce a hierarchical method that integrates small-scale information from field sub-volumes or \textit{patches} with large-scale statistics such as power spectrum and bispectrum. This hybrid strategy is efficient both computationally and in terms of the amount of training data required. It overcomes the memory limitations associated with full-field training. We show that our approach enhances Fisher information relative to analytical summaries and matches that of a very different approach (wavelet-based statistics), providing evidence that we are estimating the full information content of the dark matter density field at the resolution of $\sim 7.8~\mathrm{Mpc}/h$.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

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

  1. Learning Cosmology from Nearest Neighbour Statistics

    astro-ph.CO 2025-11 conditional novelty 6.0

    Nearest-neighbour distance maps, combined with kNN-CDFs in a hybrid neural network, constrain Ωm and σ8 from Quijote halos with R2=0.80 and 0.93, matching or beating point-cloud methods at a fraction of the compute.

  2. Hierarchical summaries for primordial non-Gaussianities

    astro-ph.CO 2025-10 conditional novelty 6.0

    Combining a patch-based neural summary with the power spectrum and bispectrum improves simulated f_NL constraints by 30-45% at k_max ≈ 0.1 h/Mpc and captures information beyond the bispectrum.

Reference graph

Works this paper leans on

78 extracted references · 52 canonical work pages · cited by 2 Pith papers · 5 internal anchors

  1. [1]

    Collaboration et al.,The desi experiment part i: Science,targeting, and survey design, arXiv e-prints (2016) [ 1611.00036]

    D. Collaboration et al.,The desi experiment part i: Science,targeting, and survey design, arXiv e-prints (2016) [ 1611.00036]

    Pith/arXiv arXiv 2016

  2. [2]

    Doré et al.,Cosmology with the spherex all-sky spectral survey, arXiv e-prints (2014) [1412.4872]

    O. Doré et al.,Cosmology with the spherex all-sky spectral survey, arXiv e-prints (2014) [1412.4872]

    Pith/arXiv arXiv 2014

  3. [3]

    Laureijs et al.,Euclid definition study report, arXiv e-prints (2011) [ 1110.3193]

    R. Laureijs et al.,Euclid definition study report, arXiv e-prints (2011) [ 1110.3193]

    Pith/arXiv arXiv 2011

  4. [4]

    Collaboration, R.L

    V.C.R.O.L.S.S.S. Collaboration, R.L. Jones, M.T. Bannister, B.T. Bolin, C.O. Chandler, S.R. Chesley et al.,The scientific impact of the vera c. rubin observatory’s legacy survey of space and time (lsst) for solar system science, arXiv e-prints (2020) [ 2009.07653]

    Pith/arXiv arXiv 2020

  5. [5]

    Weinberg, M.J

    D.H. Weinberg, M.J. Mortonson, D.J. Eisenstein, C. Hirata, A.G. Riess and E. Rozo, Observational probes of cosmic acceleration, Physics Reports 530 (2013) 87

    work page 2013

  6. [6]

    Peebles,Principles of Physical Cosmology, Princeton University Press (1993)

    P. Peebles,Principles of Physical Cosmology, Princeton University Press (1993)

    work page 1993

  7. [7]

    Nonlinear evolution of cosmological power spectra

    J.A. Peacock and S.J. Dodds,Nonlinear evolution of cosmological power spectra, Mon. Not. Roy. Astron. Soc.280 (1996) L19 [astro-ph/9603031]

    work page internal anchor Pith review Pith/arXiv arXiv 1996

  8. [8]

    SDSS collaboration, The 3-D power spectrum of galaxies from the SDSS, Astrophys. J. 606 (2004) 702 [astro-ph/0310725]

    Pith/arXiv arXiv 2004

  9. [9]

    Sánchez and S

    A.G. Sánchez and S. Cole,The galaxy power spectrum: precision cosmology from large-scale structure?, Monthly Notices of the Royal Astronomical Society385 (2008) 830

    work page 2008

  10. [10]

    Bernardeau, S

    F. Bernardeau, S. Colombi, E. Gaztañaga and R. Scoccimarro,Large-scale structure of the universe and cosmological perturbation theory, Physics Reports 367 (2002) 1. – 16 –

    work page 2002

  11. [11]

    Gil-Marín, C

    H. Gil-Marín, C. Wagner, L. Verde, C. Porciani and R. Jimenez,Perturbation theory approach for the power spectrum: from dark matter in real space to massive haloes in redshift space, Journal of Cosmology and Astroparticle Physics2012 (2012) 029–029

    work page 2012

  12. [12]

    Z. Vlah, U. Seljak, M.Y. Chu and Y. Feng,Perturbation theory, effective field theory, and oscillations in the power spectrum, Journal of Cosmology and Astroparticle Physics2016 (2016) 057–057

    work page 2016

  13. [13]

    Carron and M.C

    J. Carron and M.C. Neyrinck,Information content of the power spectrum, Astrophys. J. 750 (2012) 28

    work page 2012

  14. [14]

    Neyrinck, I

    M.C. Neyrinck, I. Szapudi and A.S. Szalay,Rejuvenating the matter power spectrum: Restoring information with a logarithmic density mapping, Astrophys. J. Lett.698 (2009) L90

    work page 2009

  15. [15]

    Dawson et al.,The baryon oscillation spectroscopic survey of sdss-iii, Astronomical Journal 145 (2013) 10

    K.S. Dawson et al.,The baryon oscillation spectroscopic survey of sdss-iii, Astronomical Journal 145 (2013) 10

    work page 2013

  16. [16]

    S. Alam, M. Ata, S. Bailey, F. Beutler, D. Bizyaev, J.A. Blazek et al.,The clustering of galaxies in the completed sdss-iii baryon oscillation spectroscopic survey: cosmological analysis of the dr12 galaxy sample, Monthly Notices of the Royal Astronomical Society470 (2017) 2617–2652

    work page 2017

  17. [17]

    Scoccimarro,The bispectrum: from theory to observations, Astrophys

    R. Scoccimarro,The bispectrum: from theory to observations, Astrophys. J. 544 (2000) 597 [astro-ph/0004086]

    Pith/arXiv arXiv 2000

  18. [18]

    Yankelevich and C

    V. Yankelevich and C. Porciani,Cosmological information in the redshift-space bispectrum, Mon. Not. Roy. Astron. Soc.483 (2019) 2078 [1807.07076]

    Pith/arXiv arXiv 2019

  19. [19]

    Fry,The galaxy three-point correlation function and the clustering of galaxies, Astrophys

    J.N. Fry,The galaxy three-point correlation function and the clustering of galaxies, Astrophys. J. 279 (1984) 499

    work page 1984

  20. [20]

    Sefusatti and E

    E. Sefusatti and E. Komatsu,The bispectrum of galaxies from high-redshift galaxy surveys: Primordial non-gaussianity and non-linear galaxy bias, Phys. Rev. D76 (2007) 083004

    work page 2007

  21. [21]

    Verde and A.F

    L. Verde and A.F. Heavens,On the trispectrum as a gaussian test for cosmology, The Astrophysical Journal 553 (2001) 14–24

    work page 2001

  22. [22]

    Verde,A practical guide to basic statistical techniques for data analysis in cosmology, 2008

    L. Verde,A practical guide to basic statistical techniques for data analysis in cosmology, 2008

    work page 2008

  23. [23]

    Cosmology: from theory to data, from data to theory

    F. Leclercq, A. Pisani and B.D. Wandelt,Cosmology: from theory to data, from data to theory, Proc. Int. Sch. Phys. Fermi186 (2014) 189 [1403.1260]

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  24. [24]

    Carron and I

    J. Carron and I. Szapudi,Optimal non-linear transformations for large-scale structure statistics, Mon. Not. R. Astron. Soc.434 (2013) 2961 [1306.1230]

    Pith/arXiv arXiv 2013

  25. [25]

    Massara, F

    E. Massara, F. Villaescusa-Navarro, C. Hahn, M.M. Abidi, M. Eickenberg, S. Ho et al., Cosmological Information in the Marked Power Spectrum of the Galaxy Field, Astrophys. J. 951 (2023) 70 [2206.01709]

    Pith/arXiv arXiv 2023

  26. [26]

    Hitting the mark: Optimising Marked Power Spectra for Cosmology

    J.A. Cowell, D. Alonso and J. Liu,Optimizing marked power spectra for cosmology, Mon. Not. R. Astron. Soc.535 (2024) 3129 [2409.05695]

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  27. [27]

    Neyrinck, I

    M.C. Neyrinck, I. Szapudi and A.S. Szalay,Rejuvenating power spectra. ii. the gaussianized galaxy density field, The Astrophysical Journal731 (2011) 116

    work page 2011

  28. [28]

    Jasche and B.D

    J. Jasche and B.D. Wandelt,Bayesian physical reconstruction of initial conditions from large-scale structure surveys, Monthly Notices of the Royal Astronomical Society432 (2013) 894–913

    work page 2013

  29. [29]

    Lavaux, J

    G. Lavaux, J. Jasche and F. Leclercq,Systematic-free inference of the cosmic matter density field from sdss3-boss data, 2019

    work page 2019

  30. [30]

    Neal et al.,Mcmc using hamiltonian dynamics, Handbook of markov chain monte carlo2 (2011) 2

    R.M. Neal et al.,Mcmc using hamiltonian dynamics, Handbook of markov chain monte carlo2 (2011) 2. – 17 –

    work page 2011

  31. [31]

    Kaiser,Clustering in real space and in redshift space, Mon

    N. Kaiser,Clustering in real space and in redshift space, Mon. Not. Roy. Astron. Soc.227 (1987) 1

    work page 1987

  32. [32]

    Porto, L

    R.A. Porto, L. Senatore and M. Zaldarriaga,The lagrangian-space effective field theory of large scale structures, Journal of Cosmology and Astroparticle Physics2014 (2014) 022–022

    work page 2014

  33. [33]

    Stadler, F

    J. Stadler, F. Schmidt and M. Reinecke,Cosmology inference at the field level from biased tracers in redshift-space, Journal of Cosmology and Astroparticle Physics2023 (2023) 069

    work page 2023

  34. [34]

    Tucci and F

    B. Tucci and F. Schmidt,EFTofLSS meets simulation-based inference:σ 8 from biased tracers, JCAP 05 (2024) 063 [2310.03741]

    Pith/arXiv arXiv 2024

  35. [35]

    SimBIG Collaboration collaboration, Field-level simulation-based inference of galaxy clustering with convolutional neural networks, Phys. Rev. D109 (2024) 083536

    work page 2024

  36. [36]

    Papamakarios, E

    G. Papamakarios, E. Nalisnick, D.J. Rezende, S. Mohamed and B. Lakshminarayanan, Normalizing flows for probabilistic modeling and inference, Journal of Machine Learning Research 22 (2021) 1

    work page 2021

  37. [37]

    Cranmer, J

    K. Cranmer, J. Brehmer and G. Louppe,The frontier of simulation-based inference, Proc. Natl. Acad. Sci. USA117 (2020) 30055

    work page 2020

  38. [38]

    Alsing, T

    J. Alsing, T. Charnock, S. Feeney and B. Wandelt,Fast likelihood-free cosmology with neural density estimators and active learning, Mon. Not. R. Astron. Soc.488 (2019) 4440

    work page 2019

  39. [39]

    Ravanbakhsh, J.B

    S. Ravanbakhsh, J.B. Oliva, S. Fromenteau, L.C. Price, J. Schneider, B. Póczos et al., Estimating cosmological parameters from the dark matter distribution, Proceedings of the 34th International Conference on Machine Learning(2017) 2407

    work page 2017

  40. [40]

    Modi and O.H.E

    C. Modi and O.H.E. Philcox,Hybrid SBI or How I Learned to Stop Worrying and Learn the Likelihood, arXiv e-prints (2023) arXiv:2309.10270 [2309.10270]

    Pith/arXiv arXiv 2023

  41. [41]

    Desjacques, D

    V. Desjacques, D. Jeong and F. Schmidt,Large-scale galaxy bias, Phys. Rep. 733 (2018) 1

    work page 2018

  42. [42]

    Eisenstein and W

    D.J. Eisenstein and W. Hu,Baryonic features in the matter transfer function, The Astrophysical Journal 496 (1998) 605–614

    work page 1998

  43. [43]

    Villaescusa-Navarro et al.,The Quijote simulations, Astrophys

    F. Villaescusa-Navarro et al.,The Quijote simulations, Astrophys. J. Suppl.250 (2020) 2 [1909.05273]

    Pith/arXiv arXiv 2020

  44. [44]

    Springel,The cosmological simulation code gadget-2, Monthly Notices of the Royal Astronomical Society 364 (2005) 1105–1134

    V. Springel,The cosmological simulation code gadget-2, Monthly Notices of the Royal Astronomical Society 364 (2005) 1105–1134

    2005

  45. [45]

    Planck collaboration, Planck 2018 results. VI. Cosmological parameters, Astron. Astrophys. 641 (2020) A6 [1807.06209]

    Pith/arXiv arXiv 2018

  46. [46]

    Alsing and B

    J. Alsing and B. Wandelt,Generalized massive optimal data compression, Mon. Not. Roy. Astron. Soc.476 (2018) L60 [1712.00012]

    Pith/arXiv arXiv 2018

  47. [47]

    G. Jung, A. Ravenni, M. Liguori, M. Baldi, W.R. Coulton, F. Villaescusa-Navarro et al., Quijote-PNG: Optimizing the Summary Statistics to Measure Primordial Non-Gaussianity, Astrophys. J. 976 (2024) 109 [2403.00490]

    Pith/arXiv arXiv 2024

  48. [48]

    Tegmark, A.N

    M. Tegmark, A.N. Taylor and A.F. Heavens,Karhunen-Loève Eigenvalue Problems in Cosmology: How Should We Tackle Large Data Sets?, Astrophys. J. 480 (1997) 22 [astro-ph/9603021]

    Pith/arXiv arXiv 1997

  49. [49]

    Tegmark,How to measure cmb power spectra without losing information, Physical Review D 55 (1997) 5895–5907

    M. Tegmark,How to measure cmb power spectra without losing information, Physical Review D 55 (1997) 5895–5907

    work page 1997

  50. [50]

    Heavens, A.N

    A.F. Heavens, A.N. Taylor and W.E. Ballinger,Maximum-likelihood estimation of power spectra from galaxy redshift surveys, Mon. Not. R. Astron. Soc.317 (2000) 965

    work page 2000

  51. [51]

    Rao,Information and the accuracy attainable in the estimation of statistical parameters, Bulletin of the Calcutta Mathematical Society37 (1945) 81

    C.R. Rao,Information and the accuracy attainable in the estimation of statistical parameters, Bulletin of the Calcutta Mathematical Society37 (1945) 81. – 18 –

    work page 1945

  52. [52]

    Cramér,Mathematical Methods of Statistics, Princeton University Press, Princeton, NJ (1946)

    H. Cramér,Mathematical Methods of Statistics, Princeton University Press, Princeton, NJ (1946)

    work page 1946

  53. [53]

    Coulton and B.D

    W.R. Coulton and B.D. Wandelt,How to estimate fisher information matrices from simulations, 2023

    work page 2023

  54. [54]

    Sefusatti, M

    E. Sefusatti, M. Crocce, S. Pueblas and R. Scoccimarro,Cosmology and the bispectrum, Phys. Rev. D 74 (2006) 023522

    work page 2006

  55. [55]

    D’Amico, Y

    G. D’Amico, Y. Donath, M. Lewandowski, L. Senatore and P. Zhang,The BOSS bispectrum analysis at one loop from the Effective Field Theory of Large-Scale Structure, JCAP 05 (2024) 059 [2206.08327]

    Pith/arXiv arXiv 2024

  56. [56]

    SDSS collaboration, Cosmological parameters from SDSS and WMAP, Phys. Rev. D69 (2004) 103501 [astro-ph/0310723]

    Pith/arXiv arXiv 2004

  57. [57]

    Mandelbaum, A

    R. Mandelbaum, A. Slosar, T. Baldauf, U. Seljak, C.M. Hirata, R. Nakajima et al., Cosmological parameter constraints from galaxy–galaxy lensing and galaxy clustering with the sdss dr7, Monthly Notices of the Royal Astronomical Society432 (2013) 1544–1575

    work page 2013

  58. [58]

    Heymans et al.,KiDS-1000 Cosmology: Multi-probe weak gravitational lensing and spectroscopic galaxy clustering constraints, Astron

    C. Heymans et al.,KiDS-1000 Cosmology: Multi-probe weak gravitational lensing and spectroscopic galaxy clustering constraints, Astron. Astrophys.646 (2021) A140 [2007.15632]

    Pith/arXiv arXiv 2021

  59. [59]

    SDSS collaboration, Cosmological Constraints from the SDSS Luminous Red Galaxies, Phys. Rev. D 74 (2006) 123507 [astro-ph/0608632]

    Pith/arXiv arXiv 2006

  60. [60]

    DES collaboration, Dark Energy Survey Year 3 results: Constraints on extensions toΛCDM with weak lensing and galaxy clustering, Phys. Rev. D107 (2023) 083504 [2207.05766]

    Pith/arXiv arXiv 2023

  61. [61]

    Ivanov, O.H

    M.M. Ivanov, O.H. Philcox, G. Cabass, T. Nishimichi, M. Simonović and M. Zaldarriaga, Cosmology with the galaxy bispectrum multipoles: Optimal estimation and application to boss data, Physical Review D107 (2023)

    work page 2023

  62. [62]

    Hinton,ChainConsumer, The Journal of Open Source Software1 (2016) 00045

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

    work page 2016

  63. [63]

    Carron,On the assumption of gaussianity for cosmological two-point statistics and parameter dependent covariance matrices, Astronomy & Astrophysics551 (2013) A88

    J. Carron,On the assumption of gaussianity for cosmological two-point statistics and parameter dependent covariance matrices, Astronomy & Astrophysics551 (2013) A88

    work page 2013

  64. [64]

    Hahn and F

    C. Hahn and F. Villaescusa-Navarro,Constraining mν with the bispectrum. part ii. the information content of the galaxy bispectrum monopole, Journal of Cosmology and Astroparticle Physics 2021 (2021) 029

    work page 2021

  65. [65]

    Mallat,Group invariant scattering, Communications on Pure and Applied Mathematics65 (2012) 1331

    S. Mallat,Group invariant scattering, Communications on Pure and Applied Mathematics65 (2012) 1331

    work page 2012

  66. [66]

    Bruna and S

    J. Bruna and S. Mallat,Invariant scattering convolution networks, 2012

    work page 2012

  67. [67]

    Cheng, Y.-S

    S. Cheng, Y.-S. Ting, B. Ménard and J. Bruna,A new approach to observational cosmology using the scattering transform, Monthly Notices of the Royal Astronomical Society499 (2020) 5902

    work page 2020

  68. [68]

    Valogiannis and C

    G. Valogiannis and C. Dvorkin,Towards an optimal estimation of cosmological parameters with the wavelet scattering transform, Physical Review D105 (2022)

    work page 2022

  69. [69]

    Eickenberg, E

    M. Eickenberg, E. Allys, A.M. Dizgah, P. Lemos, E. Massara, M. Abidi et al.,Wavelet moments for cosmological parameter estimation, 2022

    2022

  70. [70]

    Kingma and J

    D.P. Kingma and J. Ba,Adam: A method for stochastic optimization, 2017

    work page 2017

  71. [71]

    Bairagi, B

    A. Bairagi, B. Wandelt and F. Villaescusa-Navarro,How many simulations do we need for simulation-based inference in cosmology?, arXiv e-prints(2025) arXiv:2503.13755 [2503.13755]

    arXiv 2025

  72. [72]

    Making maps of cosmological parameters

    S. Mukherjee and B.D. Wandelt,Making maps of cosmological parameters, JCAP 2018 (2018) 042 [1712.01986]. – 19 –

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  73. [73]

    SimBIG collaboration, Galaxy clustering analysis with SimBIG and the wavelet scattering transform, Phys. Rev. D109 (2024) 083535 [2310.15250]

    Pith/arXiv arXiv 2024

  74. [74]

    Validation on simulations, Phys

    DES collaboration, Dark Energy Survey Year 3 results: Simulation-based cosmological inference with wavelet harmonics, scattering transforms, and moments of weak lensing mass maps. Validation on simulations, Phys. Rev. D109 (2024) 063534 [2310.17557]

    Pith/arXiv arXiv 2024

  75. [75]

    X. Zhao, Y. Mao, S. Zuo and B.D. Wandelt,Simulation-based inference of reionization parameters from 3d tomographic 21 cm light-cone images – ii: Application of solid harmonic wavelet scattering transform, 2024

    work page 2024

  76. [76]

    Andreux, T

    M. Andreux, T. Angles, G. Exarchakis, R. Leonarduzzi, G. Rochette, L. Thiry et al.,Kymatio: Scattering transforms in python, 2022

    work page 2022

  77. [77]

    Kendall and A

    M. Kendall and A. Stuart,The Advanced Theory of Statistics: Distribution Theory, Distribution Theory, Macmillan (1977)

    work page 1977

  78. [78]

    Fishnets: Information-Optimal, Scalable Aggregation for Sets and Graphs

    T.L. Makinen, J. Alsing and B.D. Wandelt,Fishnets: Information-Optimal, Scalable Aggregation for Sets and Graphs, arXiv e-prints (2023) arXiv:2310.03812 [2310.03812]. – 20 –

    work page internal anchor Pith review Pith/arXiv arXiv 2023

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.