Pith. sign in

REVIEW 2 major objections 4 minor 1 cited by

Simulation-based inference for Precision Neutrino Physics through Neural Monte Carlo tuning

T0 review · 2 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that two neural likelihood estimators, a transformer-based binned model and a normalizing-flow model, can substitute for JUNO's slow simulation and recover the three energy-response parameters with near-zero bias and…

desk verdict Solid SBI closure study for JUNO MC tuning; the near-zero bias claim is simulator-scope only, but the validation design is genuinely careful. read the letter →

arxiv 2507.23297 v1 pith:XSH3HMHH submitted 2025-07-31 physics.data-an cs.LGhep-exhep-phphysics.ins-det

A. Gavrikov , A. Serafini , D. Dolzhikov , A. Garfagnini , M. Gonchar , M. Grassi , R. Brugnera , V. Cerrone
show 58 more authors
L. V. D'Auria R. M. Guizzetti L. Lastrucci G. Andronico V. Antonelli A. Barresi D. Basilico M. Beretta A. Bergnoli M. Borghesi A. Brigatti R. Bruno A. Budano B. Caccianiga A. Cammi R. Caruso D. Chiesa C. Clementi C. Coletta S. Dusini A. Fabbri G. Felici G. Ferrante M.G. Giammarchi N. Giudice N. Guardone F. Houria C. Landini I. Lippi L. Loi P. Lombardi F. Mantovani S.M. Mari A. Martini L. Miramonti M. Montuschi M. Nastasi D. Orestano F. Ortica A. Paoloni L. Pelicci E. Percalli F. Petrucci E. Previtali G. Ranucci A.C. Re B. Ricci A. Romani C. Sirignano M. Sisti L. Stanco E. Stanescu Farilla V. Strati M.D.C Torri C. Tuvè C. Venettacci G. Verde L. Votano
This is my paper · ORCID
classification physics.data-ancs.LGhep-exhep-phphysics.ins-det
keywords simulation-basedinferenceneurallikelihoodestimationnormalizingflowstransformerdensityestimatorMonteCarlotuningJUNOdetectorenergyresponsecalibrationBayesiannestedsampling
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

Monte Carlo tuning of a large liquid-scintillator neutrino detector normally means rerunning the full simulation at every candidate parameter value, which becomes impractical when the likelihood has no closed form. This paper claims that two neural density estimators—a transformer that predicts binned spectra and a normalizing flow that gives continuous densities—can stand in for JUNO's simulation and, combined with Bayesian nested sampling, recover the three energy-response parameters ($k_B$, $f_C$, $Y$) from five calibration sources. In closure tests built from roughly one billion simulated events, both estimators show median biases below 0.25% for $k_B$, 0.55% for $f_C$, and 0.02% for $Y$, and the fitted uncertainties match the actual spread of best-fit values across 1,000 fits. Because the surrogate evaluates a spectrum in milliseconds rather than hours, systematic uncertainty studies become feasible. The claim matters because JUNO's physics goals require sub-percent control of the energy scale and resolution.

What carries the argument

The load-bearing object is the learned conditional density $p(x|\psi)$, where $x$ is the total number of photo-electrons and $\psi = (k_B, f_C, Y, S)$ combines the three physics parameters with the calibration-source type. TEDE maps $\psi$ to a temperature-scaled softmax over 800 bins, so the inference uses a binned Poisson likelihood; NFDE composes 120 planar flows conditioned on $\psi$ to map $x$ into a standard normal, so the inference uses an extended unbinned likelihood with an exact Jacobian. Nested sampling over the three physics parameters plus five source normalizations produces the posterior. These surrogates compress roughly $10^6$-fold the cost of evaluating the full simulation at a parameter point.

What would settle it

Fit genuine JUNO calibration-source data with the trained surrogates and compare the recovered ($k_B$, $f_C$, $Y$) against independent determinations, or check the residuals of the predicted photoelectron spectra against observed spectra; a systematic shift or poor goodness-of-fit would show that the uncertainties are not purely statistical.

Watch

Extended reading notes

Core claim

The paper establishes that neural likelihood estimation can replace the expensive simulation chain for detector energy-response tuning without losing statistical accuracy. The TEDE model conditions on the parameters and source type to output an 800-bin histogram of the photoelectron spectrum, while the NFDE model conditions on the same inputs to output an exact continuous density through a change of variables; both are trained on the official JUNO simulation output. Integrated with Bayesian nested sampling, they produce posteriors for ($k_B$, $f_C$, $Y$) whose medians sit on the true values within the quoted uncertainties and whose credible intervals track the true scatter of fits. The authors present this as a general template: if a simulation is too slow for iterative tuning, pre-train a likelihood surrogate on a parameter grid and then use standard Bayesian or frequentist inference on real data.

Load-bearing premise

The load-bearing premise is that the official JUNO simulation faithfully reproduces the real detector's energy response, because every training spectrum and every 'observed' dataset in the tests come from that same simulation; if the simulation is wrong, the near-zero biases measured here would not hold on real data.

Editorial extensions

If this is right

  • JUNO's energy-response parameters can be tuned with near-zero systematic bias and uncertainties that shrink as $1/\sqrt{N}$, so repeated fits across many parameter settings become practical.
  • The same trained surrogates work under frequentist minimization and two Bayesian samplers, giving consistent best-fit values and interval estimates.
  • A spectrum evaluation that took hours per parameter point drops to milliseconds, making experiment-wide systematic uncertainty quantification feasible.
  • The modeled conditional densities capture the non-linear and strongly correlated effects of $k_B$, $f_C$, and $Y$, so combined multi-source fits can break parameter degeneracies.
  • The approach carries over to other large neutrino and particle-physics experiments whose likelihoods are intractable.

Reading between the lines

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

  • A real test the closure tests cannot provide: fit actual JUNO calibration data with the trained surrogates and check whether the pull distributions remain centered; any simulator-to-detector mismatch would show up there, not in the simulated-data tests.
  • The same training scheme could be extended to time-dependent calibration or to include the 3-inch PMT system; the paper names these directions but does not demonstrate them.
  • The unbinned normalizing-flow estimator may show a clearer advantage at lower statistics than the exposure levels used here, where its precision already matches the binned transformer.
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

2 major / 4 minor

Summary. The paper proposes two neural likelihood estimators, TEDE (a transformer-based histogram density estimator) and NFDE (a conditional normalizing flow), as surrogates for the expensive JUNO Monte Carlo simulation in the tuning of the energy-response parameters (kB, fC, Y). The estimators are trained on roughly 600 million simulated calibration events from five sources and are subsequently used in Bayesian nested sampling, Metropolis-Hastings MCMC, and MIGRAD+MINOS fits. The central result is a set of closure tests on parameter grids deliberately offset from the training points, with 1,000 independent fits per statistical exposure and 100 fits per parameter-space point, showing median biases below about 0.25% in kB, 0.55% in fC, and 0.02% in Y, and estimated 1-sigma uncertainties that match the spread of best-fit values.

Significance. If the validation holds, the paper provides a valuable methodological template for simulation-based inference in large liquid-scintillator neutrino detectors, where analytic likelihoods are unavailable. The internal validation is unusually thorough: validation and testing grids are offset from training points to test interpolation, 1,000 fits per exposure are used to assess bias and calibration, and three independent inference methods give consistent results. The code is publicly available. The main caveat is that all training, validation, and 'observed' data are produced by the same official JUNO software, so the demonstrated unbiasedness is conditional on simulator fidelity; this is a boundary on the physical claim rather than an internal inconsistency.

major comments (2)
  1. [Abstract and Section IV.A] The abstract's claim of 'near-zero systematic biases' and 'uncertainties limited only by statistics' is established entirely through closure tests in which the 'observed' datasets are generated by the same official JUNO software [37] used for the training data (Section IV A). Any mismatch between the simulator and the physical detector would bias the inferred parameters in a real application, and no test in Sections II B or II B 2 can detect that mismatch. The abstract and conclusions should explicitly qualify these statements as simulation-based validation and identify simulator fidelity as the key assumption that must be tested on real calibration data.
  2. [Section IV.E.1.a, Eq. (14)] Equation (14) defines the expected bin count as mu_{s,i} = N_s \hat{p}_i^{TEDE}, but \hat{p}_i^{TEDE} is a probability density per Np.e. (with the 1/Delta x factor in Eq. (1)), so sum_i \hat{p}_i = 1/Delta x rather than 1. The formula should either use the bin probability (softmax output without the 1/Delta x factor) or read mu_{s,i} = N_s \hat{p}_i^{TEDE} \Delta x. Because N_s is a free parameter, the profile likelihood for (kB, fC, Y) is invariant under this rescaling and the central parameter estimates are likely unaffected; nevertheless, the equation as written contradicts the stated interpretation of N_s as the expected total number of events and must be corrected.
minor comments (4)
  1. [Section IV.D and Figure 13] The early-stopping patience for NFDE is stated as 50 in the first paragraph of Section IV.D but as 100 in the later training description and in the caption of Figure 13; these values should be reconciled.
  2. [Section IV.E.3] The prior bounds for the flat priors used in the Bayesian fits are not stated; specifying them (presumably the extent of the training grid) would improve reproducibility.
  3. [Figure 8] The y-axis ranges in Figure 8 are dominated by the uncertainty bands, which makes the claimed sub-percent-level bias curves visually indistinguishable; a zoomed inset or a separate axis for the median bias would improve readability.
  4. [Section I and Table I caption] There are minor language issues: 'both methods provides' should be 'both methods provide', and the Table I caption says 'three statistical distance' instead of 'three statistical distances'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the bias and uncertainty closure tests are genuine interpolation tests against held-out simulator points; simulator fidelity is a scope boundary, not a circular input.

full rationale

The claimed derivation chain is a three-stage surrogate-inference pipeline: (i) generate roughly 10^9 events with the official JUNO Geant4 simulation over a 21^3 training grid; (ii) train TEDE and NFDE as conditional density estimators on training points only; and (iii) fit independent pseudo-observed data drawn from validation/testing grids with nested sampling, comparing the recovered (kB, fC, Y) to the known simulation inputs. This evaluation is not a refit of the training data: the paper states that 'validation and testing datasets are positioned exactly in the middle between the training points' (Section II) and that the testing grid is 'shifted with respect to both the validation grid and the training grid' (Section IV A), so the near-zero median biases and matching uncertainty widths in Figures 8 and 9 test interpolation and statistical calibration at unseen parameter values. The only load-bearing external premise is set out in Section IV A: 'The data used in this study were generated using the official JUNO software [37], based on a detailed Geant4 detector simulation.' Because both the surrogate training data and the fitted 'observed' datasets come from that same simulator, the closure tests cannot certify simulator-to-detector fidelity; that is a scope boundary on the abstract's wording, not a circular reduction of the inference. The recovered parameters are the simulator's input parameters, and no fitted parameter is renamed as a prediction. The only self-citation, Ref. [52], is a pointer to a predecessor study and carries no load in the derivation. Accordingly, no circular step is identified.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim does not introduce new physical entities. It rests on the fidelity of the JUNO simulation, standard density-estimation mathematics, and the ML hyperparameters listed. The main practical caveat is that all ground truth is generated by the same simulator used to train the surrogates, so the method is validated in a closed loop.

free parameters (4)
  • Per-source normalization Ns = fit-dependent
    Five free normalization constants in the joint likelihood, one per calibration source (Section IV E 1). They absorb source statistics and are profiled or marginalized in every fit.
  • Softmax temperature T (TEDE) = 2.03
    Hyperparameter tuned via Optuna on validation Cramér-von Mises distance (Table III). Controls the sharpness of the output PDF.
  • Number of planar flows nflows (NFDE) = 120
    Hyperparameter chosen by hyperparameter optimization (Table III). Sets the capacity of the density transformation.
  • NFDE conditioning network width nunits = 30
    Hyperparameter controlling the conditioning network capacity, selected via Optuna (Table III).
assumptions (5)
  • standard math Change-of-variables formula for invertible normalizing flows yields exact densities.
    Used in Eq. (7) and Eq. (8) to define the NFDE likelihood and training loss.
  • domain assumption The JUNO software simulation faithfully represents the detector energy response as a function of (kB, fC, Y).
    All training, validation, and testing data come from this simulator (Section IV A). Any simulator bias is invisible to the closure test and would bias real-data inference.
  • domain assumption The five calibration sources are statistically independent.
    Assumed when constructing the total likelihood as a product of per-source likelihoods in Eq. (19)-(20).
  • domain assumption Flat priors are used for (kB, fC, Y) and Ns.
    Bayesian inference in Section IV E 2b uses flat priors; posterior medians and credible intervals depend on this choice.
  • ad hoc to paper The learned conditional densities interpolate smoothly between training grid points.
    The method relies on generalization from a 21^3 grid. Validation and testing datasets are positioned at midpoints, but behavior near edges or beyond the grid is not tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Simulation-based inference for Precision Neutrino Physics through Neural Monte Carlo tuning." pith.science (2026). https://pith.science/paper/XSH3HMHH

@misc{pith2026250723297,
  author       = {Pith},
  title        = {Pith review of: Simulation-based inference for Precision Neutrino Physics through Neural Monte Carlo tuning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XSH3HMHH}},
  note         = {Machine review of arXiv:2507.23297}
}
read the original abstract

Precise modeling of detector energy response is crucial for next-generation neutrino experiments which present computational challenges due to lack of analytical likelihoods. We propose a solution using neural likelihood estimation within the simulation-based inference framework. We develop two complementary neural density estimators that model likelihoods of calibration data: conditional normalizing flows and a transformer-based regressor. We adopt JUNO - a large neutrino experiment - as a case study. The energy response of JUNO depends on several parameters, all of which should be tuned, given their non-linear behavior and strong correlations in the calibration data. To this end, we integrate the modeled likelihoods with Bayesian nested sampling for parameter inference, achieving uncertainties limited only by statistics with near-zero systematic biases. The normalizing flows model enables unbinned likelihood analysis, while the transformer provides an efficient binned alternative. By providing both options, our framework offers flexibility to choose the most appropriate method for specific needs. Finally, our approach establishes a template for similar applications across experimental neutrino and broader particle physics.

Figures

Figures reproduced from arXiv: 2507.23297 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9 [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10 [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: FIG. 11 [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: FIG. 12 [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]
Figure 13
Figure 13. Figure 13: FIG. 13 [PITH_FULL_IMAGE:figures/full_fig_p018_13.png]
Figure 14
Figure 14. Figure 14: FIG. 14 [PITH_FULL_IMAGE:figures/full_fig_p023_14.png]
Figure 15
Figure 15. Figure 15: FIG. 15 [PITH_FULL_IMAGE:figures/full_fig_p023_15.png]
Figure 16
Figure 16. Figure 16: FIG. 16 [PITH_FULL_IMAGE:figures/full_fig_p024_16.png]
Figure 17
Figure 17. Figure 17: FIG. 17 [PITH_FULL_IMAGE:figures/full_fig_p025_17.png]

Discussion (0). Continue with ORCID 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. Flow-Based Surrogates for High-Dimensional Likelihoods in Experimental Neutrino Physics

    hep-ex 2026-07 accept novelty 6.0 of 10

    A hybrid coupling-plus-autoregressive normalizing flow trained on a 110-parameter T2K-like near-detector likelihood reaches 98% relative ESS versus 5% for the post-fit Gaussian and matches MCMC flux predictions.

Reference graph

Works this paper leans on

113 extracted references · 69 canonical work pages · cited by 1 Pith paper

  1. [37]

    An et al

    F. An et al. , Neutrino Physics with JUNO, J. Phys. G 43, 030401 (2016)

    work page 2016

  2. [1]

    Validation dataset 1. This dataset uses a 103 = 1000 point grid, shifted relative to the training grid to lie exactly midway between training points, probing the models interpo- lation capability in an unbiased manner under the most diffi- cult conditions: ΘVal1 kB = {7.35 + 1.8j} g · cm−2 · GeV−1, ΘVal1 fC = {0.075 + 0.1j}, ΘVal1 Y = {8300 + 400j} MeV−1,...

  3. [2]

    Aggregated results and the robustness analysis To assess the robustness and potential biases of our ap- proach, we conducted a comprehensive analysis across vary- ing statistical conditions and throughout the parameter space. a. Stability under varying statistical exposure levels. The analysis based on testing data 2 corresponds to results obtained for a ...

  4. [3]

    black box

    Differences between the parameter minimization methods As discussed in Section IV E, we employed three differ- ent methods for parameter estimation: the Frequentist MI- GRAD optimization algorithm (from iminuit [60]) and its MINOS uncertainty estimation, alongside two Bayesian approaches (nested sampling from ultranest and a self- written Metropolis-Hasti...

  5. [4]

    Validation dataset 2. This dataset comprises three spe- cific parameter points chosen from the validation dataset 1 grid space, but simulated with high statistics ( 107 events per source per point, ∼200 million total events): ΘVal2 = {(kB,i, fC,i, Yi) | i = 1, 2, 3} = = {( 7 .35 g · cm−2 · GeV−1, 0.075, 8300 MeV−1), (14.55 g · cm−2 · GeV−1, 0.475, 9900 Me...

  6. [5]

    Testing dataset 1. Similar to validation dataset 1, this uses a 103 = 1000 point grid, again shifted to lie midway between training points, but different from the validation grid: ΘTest1 kB = {6.45 + 1.8j} g · cm−2 · GeV−1, ΘTest1 fC = {0.025 + 0.1j}, ΘTest1 Y = {8100 + 400j} MeV−1, where j : j ∈ Z, 0 ≤ j ≤ 9. Thus, ΘTest1 = ΘTest1 kB × ΘTest1 fC × ΘTest1...

  7. [6]

    Testing dataset 2. This dataset focuses on a sin- gle parameter point near the center of the parameter space ({15.45 g · cm−2 · GeV−1, 0.525, 10100 MeV−1}), also lo- cated midway between training points. For this point, we pro- duce 1000 independent simulations (using different random seeds) for each of five different event exposures per source: {103, 2 ×...

  8. [7]

    JUNO software

    Transformer Encoder Density Estimator (TEDE) a) The method. This direct regression approach begins by approximating the one-dimensional target continuous PDF p(x|ψ) with a histogram using a fixed binning scheme. Let this histogram-based approximation, derived from the JUNO software output for a given condition ψ, be represented by the vector pJSW = {pJSW ...

Show all 113 references
  1. [8]

    This approach directly models the contin- uous conditional density p(x|ψ) using the principle of nor- malizing flows [47]

    Normalizing Flows Density Estimator (NFDE) a) The method. This approach directly models the contin- uous conditional density p(x|ψ) using the principle of nor- malizing flows [47]. In out study, we consider conditional normalizing flows that learn an invertible and differentia...

  2. [9]

    hyperparam- eters

    [89]. This metric balances sensitivity to both global distribution shifts and local deformations, effec- tively measuring theL2-norm of the difference between 16 CDFs: d2(u, v) = sZ +∞ −∞ |U (x) − V (x)|2 dx (11) • For p = ∞, we obtain the Kolmogorov-Smirnov dis- tance, which ...

  3. [10]

    related to the training procedure. The architecture hyper- parameters are focused on values defining the transformer encoder structure [46]: the number of tokens that encode each input MC parameter ntokens, the embedding dimension dmodel, the number of attention heads nhead, t...

  4. [11]

    For a specific source, we define the corresponding subset of data as Ds and the overall nor- malization constant Ns as the total number of expected events for that source

    Likelihood functions for statistical inference Let the parameter vector be denoted by ϕ = {kB, fC, Y}, and the total dataset by D. For a specific source, we define the corresponding subset of data as Ds and the overall nor- malization constant Ns as the total number of expecte...

  5. [12]

    In our study, two different approaches are employed: a

    Inference approaches After constructing the appropriate likelihood functions, the next step is to infer the parameter vector ϕ. In our study, two different approaches are employed: a. Frequentist inference. For the Frequentist perspec- tive, the parameter estimation is perform...

  6. [13]

    Parameter estimation discussion The dual approach, Frequentist optimization with iminuit (using the MINOS algorithm for robust uncertainty estimation) and Bayesian sampling with ultranest and Metropolis-Hastings, allows for cross-validation of parameter estimates while providi...

    2020

  7. [14]

    Capozzi et al., Neutrino masses and mixing: Entering the era of subpercent precision, Phys

    F. Capozzi et al., Neutrino masses and mixing: Entering the era of subpercent precision, Phys. Rev. D111, 093006 (2025)

    2025

  8. [15]

    Agostini et al., The Monte Carlo simulation of the Borex- ino detector, Astropart

    M. Agostini et al., The Monte Carlo simulation of the Borex- ino detector, Astropart. Phys. 97, 8 (2018)

    2018

  9. [16]

    F. P. An et al., Measurement of the reactor antineutrino flux and spectrum at Daya Bay, Phys. Rev. Lett. 116, 061801 (2016)

    2016

  10. [17]

    S. N. Wood, Statistical inference for noisy nonlinear ecologi- cal dynamic systems, Nature 466, 1102 (2010)

    2010

  11. [18]

    Papamakarios and I

    G. Papamakarios and I. Murray, Fast ε-free inference of sim- ulation models with bayesian conditional density estimation, NIPS 29 (2016)

    2016

  12. [19]

    Lueckmann et al., Likelihood-free inference with emu- lator networks, PMLR 96, 32 (2019)

    J.-M. Lueckmann et al., Likelihood-free inference with emu- lator networks, PMLR 96, 32 (2019)

    2019

  13. [20]

    Papamakarios, D

    G. Papamakarios, D. Sterratt, and I. Murray, Sequential neural likelihood: Fast likelihood-free inference with autoregressive flows, PMLR 89, 837 (2019)

    2019

  14. [21]

    Cranmer, J

    K. Cranmer, J. Brehmer, and G. Louppe, The frontier of simulation-based inference, PNAS 117, 30055 (2020)

    2020

  15. [22]

    Lueckmann et al., Benchmarking simulation-based in- ference, PMLR 130, 343 (2021)

    J.-M. Lueckmann et al., Benchmarking simulation-based in- ference, PMLR 130, 343 (2021)

    2021

  16. [23]

    Alsing, B

    J. Alsing, B. Wandelt, and S. Feeney, Massive optimal data compression and density estimation for scalable, likelihood- free inference in cosmology, MNRAS 477, 2874 (2018)

    2018

  17. [24]

    Alsing et al

    J. Alsing et al. , Fast likelihood-free cosmology with neural density estimators and active learning, MNRAS 488, 4440 (2019)

    2019

  18. [25]

    Kosiba et al., The cosmological analysis of X-ray cluster surveys - VI

    M. Kosiba et al., The cosmological analysis of X-ray cluster surveys - VI. Inference based on analytically simulated ob- servable diagrams, Astron. Astrophys. 693, A46 (2025)

    2025

  19. [26]

    Brehmer et al., Mining for Dark Matter Substructure: In- ferring subhalo population properties from strong lenses with machine learning, Astrophys

    J. Brehmer et al., Mining for Dark Matter Substructure: In- ferring subhalo population properties from strong lenses with machine learning, Astrophys. J. 886, 49 (2019)

    2019

  20. [27]

    Dax et al

    M. Dax et al. , Real-Time Gravitational Wave Science with Neural Posterior Estimation, Phys. Rev. Lett. 127, 241103 (2021)

    2021

  21. [28]

    Jeffrey et al., Dark energy survey year 3 results: likelihood- free, simulation-based wCDM inference with neural compres- sion of weak-lensing map statistics, Mon

    N. Jeffrey et al., Dark energy survey year 3 results: likelihood- free, simulation-based wCDM inference with neural compres- sion of weak-lensing map statistics, Mon. Not. Roy. Astron. Soc. 536, 1303 (2024)

    2024

  22. [29]

    Brehmer et al., MadMiner: Machine learning-based infer- ence for particle physics, Comput

    J. Brehmer et al., MadMiner: Machine learning-based infer- ence for particle physics, Comput. Softw. Big Sci.4, 3 (2020)

    2020

  23. [30]

    Lazzarin, S

    M. Lazzarin, S. Alioli, and S. Carrazza, MCNNTUNES: Tun- ing Shower Monte Carlo generators with machine learning, Comput. Phys. Commun. 263, 107908 (2021)

    2021

  24. [31]

    Heller et al

    N. Heller et al. , Rejection Sampling with Autodifferentia- tion - Case study: Fitting a Hadronization Model (2024), arXiv:2411.02194

    2024 arXiv

  25. [32]

    Aad et al., An implementation of neural simulation-based inference for parameter estimation in ATLAS, Rept

    G. Aad et al., An implementation of neural simulation-based inference for parameter estimation in ATLAS, Rept. Prog. Phys. 88, 067801 (2025)

    2025

  26. [33]

    Pina-Otey et al., Likelihood-free inference of experimental neutrino oscillations using neural spline flows, Phys

    S. Pina-Otey et al., Likelihood-free inference of experimental neutrino oscillations using neural spline flows, Phys. Rev. D 101, 113001 (2020)

    2020

  27. [34]

    El Baz and F

    M. El Baz and F. S´anchez, Fast posterior probability sampling with normalizing flows and its applicability in Bayesian anal- ysis in particle physics, Phys. Rev. D 109, 032008 (2024)

    2024

  28. [35]

    Brehmer et al

    J. Brehmer et al. , Mining gold from implicit models to im- prove likelihood-free inference, PNAS 117, 5242 (2020)

    2020

  29. [36]

    Brehmer et al

    J. Brehmer et al. , Constraining effective field theories with machine learning, Phys. Rev. Lett. 121, 111801 (2018)

    2018

  30. [38]

    Abusleme et al., JUNO Physics and Detector, PPNP 123, 103927 (2022)

    A. Abusleme et al., JUNO Physics and Detector, PPNP 123, 103927 (2022)

    2022

  31. [39]

    Abusleme et al

    A. Abusleme et al. , Potential to identify neutrino mass or- dering with reactor antineutrinos at JUNO, CPC 49, 033104 (2025)

    2025

  32. [40]

    Abusleme et al

    A. Abusleme et al. , Sub-percent precision measurement of neutrino oscillation parameters with JUNO, CPC 46, 123001 (2022)

    2022

  33. [41]

    Abusleme et al., JUNO sensitivity to 7Be, pep, and CNO solar neutrinos, JCAP (10), 022

    A. Abusleme et al., JUNO sensitivity to 7Be, pep, and CNO solar neutrinos, JCAP (10), 022

  34. [42]

    Zhao et al., Model-independent Approach of the JUNO 8B Solar Neutrino Program, Astrophys

    J. Zhao et al., Model-independent Approach of the JUNO 8B Solar Neutrino Program, Astrophys. J. 965, 122 (2024)

    2024

  35. [43]

    Abusleme et al., Real-time monitoring for the next core- collapse supernova in JUNO, JCAP (01), 057

    A. Abusleme et al., Real-time monitoring for the next core- collapse supernova in JUNO, JCAP (01), 057

  36. [44]

    Abusleme et al., Prospects for detecting the diffuse super- nova neutrino background with JUNO, JCAP (10), 033

    A. Abusleme et al., Prospects for detecting the diffuse super- nova neutrino background with JUNO, JCAP (10), 033

  37. [45]

    Abusleme et al

    A. Abusleme et al. , JUNO sensitivity to low energy atmo- spheric neutrino spectra, EPJ C 81, 10 (2021)

    2021

  38. [46]

    Abusleme et al., JUNO sensitivity on proton decay p → νK + searches, CPC 47, 113002 (2023)

    A. Abusleme et al., JUNO sensitivity on proton decay p → νK + searches, CPC 47, 113002 (2023). 21

    2023

  39. [47]

    Abusleme et al., JUNO sensitivity to invisible decay modes of neutrons, EPJ C 85, 5 (2025)

    A. Abusleme et al., JUNO sensitivity to invisible decay modes of neutrons, EPJ C 85, 5 (2025)

    2025

  40. [48]

    Abusleme et al

    A. Abusleme et al. , JUNO sensitivity to the annihilation of MeV dark matter in the galactic halo, JCAP (09), 001

  41. [49]

    Abusleme et al

    A. Abusleme et al. , Prediction of energy resolution in the JUNO experiment, CPC 49, 013003 (2025)

    2025

  42. [50]

    Lin et al

    T. Lin et al. , Simulation software of the JUNO experiment, EPJ C 83, 382 (2023)

    2023

  43. [51]

    Qian et al., Vertex and energy reconstruction in JUNO with machine learning methods, NIM-A 1010, 165527 (2021)

    Z. Qian et al., Vertex and energy reconstruction in JUNO with machine learning methods, NIM-A 1010, 165527 (2021)

    2021

  44. [52]

    Gavrikov, Y

    A. Gavrikov, Y . Malyshkin, and F. Ratnikov, Energy recon- struction for large liquid scintillator detectors with machine learning techniques: aggregated features approach, Eur. Phys. J. C 82, 1021 (2022)

    2022

  45. [53]

    Yang et al., First attempt of directionality reconstruction for atmospheric neutrinos in a large homogeneous liquid scintil- lator detector, Phys

    Z. Yang et al., First attempt of directionality reconstruction for atmospheric neutrinos in a large homogeneous liquid scintil- lator detector, Phys. Rev. D109, 052005 (2024)

    2024

  46. [54]

    Gavrikov et al., Interpretable machine learning approach for electron antineutrino selection in a large liquid scintillator detector, PLB 860, 139141 (2025)

    A. Gavrikov et al., Interpretable machine learning approach for electron antineutrino selection in a large liquid scintillator detector, PLB 860, 139141 (2025)

    2025

  47. [55]

    W. Jiang et al., Machine-learning based photon counting for PMT waveforms and its application to the improvement of the energy resolution in large liquid scintillator detectors, EPJ C 85, 69 (2025)

    2025

  48. [56]

    Malyshkin, Novel fitting approach based on a neural net- work for juno, in CHEP2024 (Krak´ow, Poland, 2024)

    Y . Malyshkin, Novel fitting approach based on a neural net- work for juno, in CHEP2024 (Krak´ow, Poland, 2024)

    2024

  49. [57]

    Liu et al., Neutrino type identification for atmospheric neu- trinos in a large homogeneous liquid scintillation detector (2025), arXiv:2503.21353

    J. Liu et al., Neutrino type identification for atmospheric neu- trinos in a large homogeneous liquid scintillation detector (2025), arXiv:2503.21353

    2025 arXiv

  50. [58]

    Abusleme et al., Calibration Strategy of the JUNO Experi- ment, JHEP 03, 004

    A. Abusleme et al., Calibration Strategy of the JUNO Experi- ment, JHEP 03, 004

  51. [59]

    Vaswani et al., Attention Is All You Need, NIPS30 (2017)

    A. Vaswani et al., Attention Is All You Need, NIPS30 (2017)

    2017

  52. [60]

    E. G. Tabak and C. V . Turner, A family of nonparametric den- sity estimation algorithms, Communications on Pure and Ap- plied Mathematics 66, 145 (2013)

    2013

  53. [61]

    Rezende and S

    D. Rezende and S. Mohamed, Variational inference with nor- malizing flows, PMLR 37, 1530 (2015)

    2015

  54. [62]

    B. L. Trippe and R. E. Turner, Conditional density estimation with bayesian normalising flows (2018), arXiv:1802.04908

    2018 arXiv

  55. [63]

    Buchner, UltraNest - a robust, general purpose Bayesian in- ference engine, The Journal of Open Source Software 6, 3001 (2021)

    J. Buchner, UltraNest - a robust, general purpose Bayesian in- ference engine, The Journal of Open Source Software 6, 3001 (2021)

    2021

  56. [64]

    Buchner, Nested sampling methods, Statistic Surveys 17, 169 (2023)

    J. Buchner, Nested sampling methods, Statistic Surveys 17, 169 (2023)

    2023

  57. [65]

    Gavrikov and A

    A. Gavrikov and A. Serafini, NeuroMCT: Fast Monte Carlo Tuning with Generative Machine Learning in the JUNO Ex- periment, in CHEP2024 (Krak´ow, Poland, 2024)

    2024

  58. [66]

    I. J. Goodfellow et al. , Generative Adversarial Networks, NIPS 27 (2014)

    2014

  59. [67]

    Gulrajani et al., Improved Training of Wasserstein GANs, NIPS 30 (2017)

    I. Gulrajani et al., Improved Training of Wasserstein GANs, NIPS 30 (2017)

    2017

  60. [68]

    Abi et al

    B. Abi et al. , Deep Underground Neutrino Experiment (DUNE), far detector technical design report, volume I: In- troduction to DUNE, JINST 15 (08), T08008

  61. [69]

    Abe et al

    K. Abe et al. , Hyper-Kamiokande design report (2018), arXiv:1805.04163

    2018 arXiv

  62. [70]

    Cowan, Statistical data analysis (Clarendon Press, 1998)

    G. Cowan, Statistical data analysis (Clarendon Press, 1998)

    1998

  63. [71]

    Barlow, Extended maximum likelihood, NIM-A 297, 496 (1990)

    R. Barlow, Extended maximum likelihood, NIM-A 297, 496 (1990)

    1990

  64. [72]

    Cowan, Statistical Data Analysis (Oxford University Press, 1998)

    G. Cowan, Statistical Data Analysis (Oxford University Press, 1998)

    1998

  65. [73]

    Dembinski et al., scikit-hep/iminuit, Zenodo 10.5281/zen- odo.3949207 (2020)

    H. Dembinski et al., scikit-hep/iminuit, Zenodo 10.5281/zen- odo.3949207 (2020)

    2020 doi

  66. [74]

    W. K. Hastings, Monte Carlo sampling methods using Markov chains and their applications, Biometrika 57, 97 (1970)

    1970

  67. [75]

    Metropolis et al

    N. Metropolis et al. , Equation of state calculations by fast computing machines, The Journal of Chemical Physics 21, 1087 (1953)

    1953

  68. [76]

    Serafini et al

    A. Serafini et al. , Accelerating Modern Neutrino Physics Analysis with GPU Parallelization, In preparation (2025)

    2025

  69. [77]

    Belle and I

    V . Belle and I. Papantonis, Principles and practice of explain- able machine learning, Frontiers in big Data4, 688969 (2021)

    2021

  70. [78]

    HEP ML Community, A Living Review of Machine Learning for Particle Physics

  71. [79]

    Dirmeier, C

    S. Dirmeier, C. Albert, and F. Perez-Cruz, Simulation-based Inference for High-dimensional Data using Surjective Sequen- tial Neural Likelihood Estimation (2023), arXiv:2308.01054

    2023 arXiv

  72. [80]

    Gelman et al., Bayesian Data Analysis, 3rd ed

    A. Gelman et al., Bayesian Data Analysis, 3rd ed. (CRC Press, 2013)

    2013

  73. [81]

    Agostinelli et al., Geant4 — a simulation toolkit, NIM-A 506, 250 (2003)

    S. Agostinelli et al., Geant4 — a simulation toolkit, NIM-A 506, 250 (2003)

    2003

  74. [82]

    Allison et al., Recent developments in Geant4, NIM-A 835, 186 (2016)

    J. Allison et al., Recent developments in Geant4, NIM-A 835, 186 (2016)

    2016

  75. [83]

    Huang et al., The Flash ADC system and PMT waveform reconstruction for the Daya Bay Experiment, NIM-A 895, 48 (2018)

    Y . Huang et al., The Flash ADC system and PMT waveform reconstruction for the Daya Bay Experiment, NIM-A 895, 48 (2018)

    2018

  76. [84]

    Lin et al., A survey of transformers, AI open 3, 111 (2022)

    T. Lin et al., A survey of transformers, AI open 3, 111 (2022)

    2022

  77. [85]

    Islam et al

    S. Islam et al. , A comprehensive survey on applications of transformers for deep learning tasks, Expert Systems with Ap- plications 241, 122666 (2024)

    2024

  78. [86]

    Butter et al., Jet diffusion versus JetGPT — modern net- works for the LHC, SciPost Physics Core 8, 026 (2025)

    A. Butter et al., Jet diffusion versus JetGPT — modern net- works for the LHC, SciPost Physics Core 8, 026 (2025)

    2025

  79. [87]

    Gavrikov, J

    A. Gavrikov, J. G. Pardi ˜nas, and A. Garfagnini, DINAMO: Dynamic and INterpretable Anomaly MOnitoring for Large- Scale Particle Physics Experiments (2025), arXiv:2501.19237

    2025

  80. [88]

    Spinner et al., Lorentz-equivariant geometric algebra trans- formers for high-energy physics, NeurIPS 37 (2024)

    J. Spinner et al., Lorentz-equivariant geometric algebra trans- formers for high-energy physics, NeurIPS 37 (2024)

    2024

  81. [89]

    Wang et al

    A. Wang et al. , Interpreting Transformers for Jet Tagging (2024), arXiv:2412.03673

    2024 arXiv

  82. [90]

    Pata et al., Improved particle-flow event reconstruction with scalable neural networks for current and future particle detec- tors, Commun

    J. Pata et al., Improved particle-flow event reconstruction with scalable neural networks for current and future particle detec- tors, Commun. Phys. 7, 124 (2024)

    2024

  83. [91]

    Mokhtar et al., Fine-tuning machine-learned particle-flow reconstruction for new detector geometries in future colliders, Phys

    F. Mokhtar et al., Fine-tuning machine-learned particle-flow reconstruction for new detector geometries in future colliders, Phys. Rev. D 111, 092015 (2025)

    2025

  84. [92]

    Dreyer et al., Automated Approach to Accurate, Precise, and Fast Detector Simulation and Reconstruction, Phys

    E. Dreyer et al., Automated Approach to Accurate, Precise, and Fast Detector Simulation and Reconstruction, Phys. Rev. Lett. 133, 211902 (2024)

    2024

  85. [93]

    J. L. Ba, J. R. Kiros, and G. E. Hinton, Layer normalization (2016), arXiv:1607.06450

    2016 arXiv

  86. [94]

    Srivastava et al., Dropout: a simple way to prevent neural networks from overfitting, Journal of Machine Learning Re- search 15, 1929 (2014)

    N. Srivastava et al., Dropout: a simple way to prevent neural networks from overfitting, Journal of Machine Learning Re- search 15, 1929 (2014)

    2014

  87. [95]

    Kullback and R

    S. Kullback and R. A. Leibler, On information and sufficiency, Ann. Math. Statist. 22, 79 (1951)

    1951

  88. [96]

    M. I. Jordan et al., An introduction to variational methods for graphical models, Machine learning 37, 183 (1999)

    1999

  89. [97]

    Dembo, T

    A. Dembo, T. M. Cover, and J. A. Thomas, Information the- oretic inequalities, IEEE Transactions on Information theory 37, 1501 (1991)

    1991

  90. [98]

    T. M. Cover and J. A. Thomas, Elements of Information The- ory (Wiley-Interscience, 2006)

    2006

  91. [99]

    Papamakarios et al., Normalizing Flows for Probabilistic Modeling and Inference, Journal of Machine Learning Re- search 22, 1 (2021)

    G. Papamakarios et al., Normalizing Flows for Probabilistic Modeling and Inference, Journal of Machine Learning Re- search 22, 1 (2021)

    2021

  92. [100]

    Paszke et al

    A. Paszke et al. , Pytorch: An imperative style, high- performance deep learning library, NeurIPS 32 (2019). 22

    2019

  93. [101]

    Falcon et al

    W. Falcon et al. , Pytorch lightning, https://github. com/PyTorchLightning/pytorch-lightning (2019)

    2019

  94. [102]

    M. L. Rizzo and G. J. Sz ´ekely, Energy distance, Wiley Inter- disciplinary Reviews: Computational statistics 8, 27 (2016)

    2016

  95. [103]

    Nair and G

    V . Nair and G. E. Hinton, Rectified linear units improve re- stricted boltzmann machines, in Proceedings of the 27th inter- national conference on machine learning (ICML) (2010) pp. 807–814

    2010

  96. [104]

    Hendrycks and K

    D. Hendrycks and K. Gimpel, Gaussian Error Linear Units (GELUs) (2016), arXiv:1606.08415

    2016 arXiv

  97. [105]

    Chen, Back-propagation neural networks for nonlinear self-tuning adaptive control, IEEE Control Systems Magazine 10, 44 (1990)

    F.-C. Chen, Back-propagation neural networks for nonlinear self-tuning adaptive control, IEEE Control Systems Magazine 10, 44 (1990)

    1990

  98. [106]

    Elfwing, E

    S. Elfwing, E. Uchibe, and K. Doya, Sigmoid-weighted linear units for neural network function approximation in reinforce- ment learning, Neural Networks 107, 3 (2018)

    2018

  99. [107]

    Loshchilov and F

    I. Loshchilov and F. Hutter, Decoupled weight decay regular- ization, in 7th International Conference on Learning Repre- sentations (2019)

    2019

  100. [108]

    Tieleman and G

    T. Tieleman and G. Hinton, Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude, COURSERA: Neural networks for machine learning 4, 26 (2012)

    2012

  101. [109]

    Li and A

    Z. Li and A. Sanjeev, An Exponential Learning Rate Schedule for Deep Learning, in 8th International Conference on Learn- ing Representations (2020)

    2020

  102. [110]

    Loshchilov and F

    I. Loshchilov and F. Hutter, SGDR: Stochastic gradient de- scent with warm restarts, in 7th International Conference on Learning Representations (2019)

    2019

  103. [111]

    Akiba et al

    T. Akiba et al. , Optuna: A next-generation hyperparameter optimization framework, in 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (2019)

    2019

  104. [112]

    Bergstra et al., Algorithms for Hyper-Parameter Optimiza- tion, NIPS 24 (2011)

    J. Bergstra et al., Algorithms for Hyper-Parameter Optimiza- tion, NIPS 24 (2011)

    2011

  105. [113]

    B. A. Murtagh and M. A. Saunders, Large-scale linearly constrained optimization, Mathematical programming 14, 41 (1978). 23 Appendix A: Energy response parameters effects with the TEDE model 10 3 10 1 101 PDF: f(Np. e. kB, fC, Y) 2 4 6 8 10 12Number of photo-electrons: Np. e. ...

    1978

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.