arxiv: 2605.07262 · v1 · submitted 2026-05-08 · ✦ hep-lat

Recognition: no theorem link

Testing machine-learned distributions against Monte Carlo data for the QCD chiral phase transition

Christian Schmidt, Frithjof Karsch, Jan Philipp Klinger, Owe Philipsen, Reinhold Kaiser, Simran Singh

Pith reviewed 2026-05-11 01:50 UTC · model grok-4.3

classification ✦ hep-lat

keywords lattice QCDchiral phase transitionmasked autoregressive flowsmachine learninginterpolationcritical quark massfirst-order transitionsMonte Carlo

0 comments

The pith

Conditional Masked Autoregressive Flows reproduce reweighting and interpolate lattice QCD data in quark mass and volume to estimate critical points with fewer ensembles.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether machine-learned conditional distributions can stand in for direct Monte Carlo simulations when exploring the chiral phase structure of five-flavor QCD. These flows are trained on existing ensembles and then produce new samples for unsimulated values of the gauge coupling, quark mass, and lattice volume. Where standard reweighting is possible, the generated distributions match the Monte Carlo results; the method also works in regimes where reweighting cannot be applied. The resulting samples supply first estimates of the critical quark mass that ends the first-order transition region, cutting the total number of lattice runs required while remaining limited by training bias near the transition.

Core claim

Conditional Masked Autoregressive Flows trained on Monte Carlo data for five degenerate quarks reproduce the distributions of lattice observables and extend them across the gauge coupling, quark mass, and spatial volume, yielding consistent first estimates of the critical quark mass that terminates the region of first-order chiral transitions.

What carries the argument

Conditional Masked Autoregressive Flows that learn the distribution of observables conditioned on bare lattice parameters and then sample from the interpolated distributions.

If this is right

The flows match standard reweighting in the gauge coupling to high accuracy.
Interpolation works for quark mass and spatial volume where reweighting is prohibitive or impossible.
Full parameter-space samples are generated in minutes once training is complete.
First estimates of the critical quark mass become available without simulating every intermediate point.
The approach supports localising phase boundaries and identifying universal scaling axes ahead of high-precision runs.

Where Pith is reading between the lines

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

The same conditional-flow technique could be applied to other multi-parameter lattice models to reduce the number of ensembles needed for phase-diagram mapping.
Replacing or augmenting the maximum-likelihood objective with a loss that penalises mode-covering might remove the current precision barrier near transitions.
Interpolated distributions could be used to choose the exact parameter values for a final high-precision Monte Carlo campaign, shortening the overall workflow.

Load-bearing premise

The mode-covering bias from maximum-likelihood training stays small enough near first-order transitions that the generated samples still produce useful first estimates of the critical quark mass.

What would settle it

New Monte Carlo simulations performed at an interpolated quark mass near the transition, compared directly to histograms or critical-mass estimates from the flow-generated samples, would show agreement or systematic mismatch.

read the original abstract

We demonstrate that conditional Masked Autoregressive Flows constitute a flexible interpolation tool for lattice QCD observables, conditioned on bare lattice parameters. As a benchmark, we use the chiral phase structure of QCD with five degenerate light quark flavours, which on coarse lattices exhibits a region of first-order chiral transitions terminating in a critical quark mass. The method successfully reproduces standard reweighting in the gauge coupling, and naturally extends to interpolation in quark mass and spatial volume, for which reweighting is computationally prohibitive or inapplicable, respectively. Once trained, the model generates samples across the full parameter space in minutes, which can be used to obtain consistent first estimates of the critical quark mass without simulating all intermediate parameter values. This offers a concrete reduction in the number of lattice ensembles required. Precision on the critical mass from learned distributions is so far prohibited by the mode-covering effect inherent to maximum-likelihood-based training, which introduces a systematic bias near first-order transitions. At the current stage, the method is well-suited for a range of practical applications: localising phase boundaries, identifying the universal scaling axes at a critical point, and accelerating informed determinations of parameter values ahead of high-precision Monte Carlo campaigns.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Conditional MAFs reproduce reweighting and extend to mass/volume interpolation in 5-flavor QCD, but mode-covering bias limits reliability near the transition.

read the letter

The paper trains conditional masked autoregressive flows on Monte Carlo ensembles for the five-flavor QCD chiral transition and shows they can generate samples across gauge coupling, quark mass, and volume. It reproduces standard reweighting results in the coupling and then uses the same model to interpolate in the other two directions, where reweighting is either expensive or impossible. Once trained, the model produces configurations in minutes and yields first estimates of the critical quark mass without filling in every intermediate ensemble. That part is concrete and directly addresses the cost of mapping phase structure on the lattice. The authors are upfront that maximum-likelihood training introduces a mode-covering bias that prevents precision near first-order transitions, which matches the stress-test concern. If that bias moves the apparent location of the critical point by an amount comparable to the spacing between training points, the claimed reduction in ensembles comes with an uncontrolled systematic. The abstract gives no error bars, no breakdown of training versus validation sets, and no quantitative measure of how large the bias actually is on the observables that matter, so it is difficult to judge whether the first estimates are useful in practice or just directionally correct. This work is aimed at lattice QCD groups already running phase-transition studies who might want a fast scouting tool before committing to dense Monte Carlo campaigns. A reader comfortable with both reweighting and basic normalizing flows will see the practical angle immediately. It deserves a serious referee because the application is new, the benchmark against reweighting is there, and the limitation is stated rather than hidden, even though the current evidence is still at the proof-of-concept level.

Referee Report

3 major / 2 minor

Summary. The manuscript demonstrates that conditional Masked Autoregressive Flows (MAFs), trained on Monte Carlo ensembles, can interpolate distributions of lattice QCD observables conditioned on bare parameters (gauge coupling, quark mass, volume). As a benchmark it reproduces standard reweighting results in the gauge coupling for the five-flavor chiral transition; it then extends the approach to quark-mass and volume interpolation (where reweighting is prohibitive or inapplicable) and uses the generated samples to locate the critical quark mass with fewer simulated ensembles. The work explicitly notes that maximum-likelihood training introduces a mode-covering bias that currently precludes high-precision determinations of the critical mass.

Significance. If the mode-covering bias can be quantified and shown to be sub-dominant to the spacing between training ensembles, the method would offer a practical reduction in the number of costly lattice simulations needed for exploratory mapping of phase boundaries and identification of universal scaling axes. The approach is therefore potentially valuable as a precursor tool that informs subsequent high-precision Monte Carlo campaigns, even if it does not yet replace them for final results.

major comments (3)

[§4] §4 (Results on critical-mass estimation): The claim that the generated samples yield 'consistent first estimates' of the critical quark mass is not supported by any quantitative metric. No table or figure reports the shift in the inferred critical mass relative to reweighting benchmarks, nor are error bars or deviation measures provided that would demonstrate the bias remains smaller than the ensemble spacing.
[§3] §3 (Training procedure): The manuscript provides no information on training/validation splits, batch sizes, or convergence diagnostics for the MAF. Without these, it is impossible to assess whether the reported reproduction of reweighting is robust or merely an artifact of overfitting to the specific ensembles used.
[§5] §5 (Discussion of volume interpolation): The extension to spatial-volume interpolation is presented as straightforward, yet the text does not address how the conditional model captures or extrapolates finite-volume corrections. This omission is load-bearing for the claim that the method works 'naturally' for volumes where direct simulation is expensive.

minor comments (2)

[Figure 3] Figure 3 and associated text: The caption should explicitly state the number of generated samples used for each histogram and whether any reweighting or post-processing was applied to the MAF output.
[Abstract] Abstract: The phrase 'precision ... is so far prohibited' is accurate but could be followed by a single sentence indicating the magnitude of the bias observed in the benchmark reweighting test.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed report. We address each major comment below and will revise the manuscript accordingly to strengthen the presentation.

read point-by-point responses

Referee: [§4] §4 (Results on critical-mass estimation): The claim that the generated samples yield 'consistent first estimates' of the critical quark mass is not supported by any quantitative metric. No table or figure reports the shift in the inferred critical mass relative to reweighting benchmarks, nor are error bars or deviation measures provided that would demonstrate the bias remains smaller than the ensemble spacing.

Authors: We agree that a quantitative metric would better support the claim. Although the manuscript already highlights the mode-covering bias as precluding high-precision results, the revised version will include a table (or figure) comparing the critical quark mass inferred from MAF-generated samples to the reweighting benchmark. This will report the numerical shift, associated uncertainties, and an explicit comparison to the spacing between training ensembles, thereby clarifying the regime in which the estimates remain useful for first-order exploration. revision: yes
Referee: [§3] §3 (Training procedure): The manuscript provides no information on training/validation splits, batch sizes, or convergence diagnostics for the MAF. Without these, it is impossible to assess whether the reported reproduction of reweighting is robust or merely an artifact of overfitting to the specific ensembles used.

Authors: We thank the referee for noting this omission. The revised manuscript will add a subsection (or appendix) specifying the training/validation split (80/20 on the available ensembles), the batch sizes used during optimization, and convergence diagnostics including training and validation negative-log-likelihood curves. These additions will allow readers to verify that the reproduction of reweighting results is not due to overfitting. revision: yes
Referee: [§5] §5 (Discussion of volume interpolation): The extension to spatial-volume interpolation is presented as straightforward, yet the text does not address how the conditional model captures or extrapolates finite-volume corrections. This omission is load-bearing for the claim that the method works 'naturally' for volumes where direct simulation is expensive.

Authors: We acknowledge that the current text does not explicitly discuss finite-volume corrections. In the revision we will expand §5 to explain that the conditional MAF learns volume dependence directly from training data spanning multiple volumes, thereby capturing finite-volume effects through the conditioning variable. We will also add a discussion of the model's ability (and limitations) to extrapolate to unsimulated volumes, including the assumptions underlying the 'natural' extension claim. revision: yes

Circularity Check

0 steps flagged

No circularity: ML model trained on independent Monte Carlo ensembles and benchmarked against separate reweighting results

full rationale

The paper trains conditional Masked Autoregressive Flows directly on existing lattice QCD Monte Carlo data for the chiral phase structure and validates the generated samples by reproducing standard reweighting results in the gauge coupling. Interpolation to quark mass and volume is presented as an extension that reduces the need for new simulations, but the critical-mass estimates are explicitly qualified as first estimates limited by mode-covering bias rather than claimed as exact derivations. No equations reduce to fitted parameters by construction, no self-citations form load-bearing uniqueness arguments, and the central workflow remains externally falsifiable against independent Monte Carlo benchmarks. The derivation chain is therefore self-contained and non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated beyond standard assumptions of neural density estimation and lattice QCD observables being representable by autoregressive flows.

pith-pipeline@v0.9.0 · 5524 in / 1022 out tokens · 41329 ms · 2026-05-11T01:50:45.400356+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

65 extracted references · 65 canonical work pages · 3 internal anchors

[1]

Albergo, G

M.S. Albergo, G. Kanwar and P.E. Shanahan,Flow-based generative models for Markov chain Monte Carlo in lattice field theory,Phys. Rev. D100(2019) 034515 [1904.12072]

work page arXiv 2019
[2]

Niedermayer, P

F. Niedermayer, P. Rufenacht and U. Wenger,Fixed point gauge actions with fat links: Scaling and glueballs,Nucl. Phys. B597(2001) 413 [hep-lat/0007007]

work page arXiv 2001
[3]

Abbott et al.,Sampling QCD field configurations with gauge-equivariant flow models,PoS LATTICE2022(2023) 036 [2208.03832]

R. Abbott et al.,Sampling QCD field configurations with gauge-equivariant flow models,PoS LATTICE2022(2023) 036 [2208.03832]

work page arXiv 2023
[4]

Albandea, L

D. Albandea, L. Del Debbio, P. Hern´ andez, R. Kenway, J. Marsh Rossney and A. Ramos, Learning trivializing flows,Eur. Phys. J. C83(2023) 676 [2302.08408]

work page arXiv 2023
[5]

Abbott, et al., Normalizing flows for lattice gauge theory in arbitrary space-time dimension (2023)

R. Abbott et al.,Normalizing flows for lattice gauge theory in arbitrary space-time dimension,2305.02402

work page arXiv
[6]

L. Wang, G. Aarts and K. Zhou,Diffusion models as stochastic quantization in lattice field theory,JHEP05(2024) 060 [2309.17082]

work page arXiv 2024
[7]

Gerdes, P

M. Gerdes, P. de Haan, R. Bondesan and M.C.N. Cheng,Continuous normalizing flows for lattice gauge theories,2410.13161

work page arXiv
[8]

Scaling flow-based approaches for topology sampling in $\mathrm{SU}(3)$ gauge theory

C. Bonanno, A. Bulgarelli, E. Cellini, A. Nada, D. Panfalone, D. Vadacchino et al.,Scaling flow-based approaches for topology sampling in SU(3) gauge theory,JHEP04(2026) 051 [2510.25704]

work page internal anchor Pith review Pith/arXiv arXiv 2026
[9]

Q. Zhu, G. Aarts, W. Wang, K. Zhou and L. Wang,Physics-Conditioned Diffusion Models for Lattice Gauge Theory,2502.05504

work page arXiv
[10]

Lehner and T

C. Lehner and T. Wettig,Gauge-equivariant pooling layers for preconditioners in lattice QCD,Phys. Rev. D110(2024) 034517 [2304.10438]

work page arXiv 2024
[11]

Kn¨ uttel, C

D. Kn¨ uttel, C. Lehner and T. Wettig,Gauge-equivariant multigrid neural networks,PoS LATTICE2023(2024) 037

work page 2024
[12]

Machine learning for four-dimensional SU(3) lattice gauge theories

U. Wenger,Machine learning for four-dimensional SU(3) lattice gauge theories, in42th International Symposium on Lattice Field Theory, 4, 2026 [2604.12416]

work page internal anchor Pith review Pith/arXiv arXiv 2026
[13]

K. Zhou, L. Wang, L.-G. Pang and S. Shi,Exploring QCD matter in extreme conditions with Machine Learning,Prog. Part. Nucl. Phys.135(2024) 104084 [2303.15136]

work page arXiv 2024
[14]

Larkoski,A step toward interpretability: smearing the likelihood,JHEP03(2025) 198 [2501.07643]

A.J. Larkoski,A step toward interpretability: smearing the likelihood,JHEP03(2025) 198 [2501.07643]

work page arXiv 2025
[15]

L. Wang, S. Shi and K. Zhou,Unsupervised learning spectral functions with neural networks, J. Phys. Conf. Ser.2586(2023) 012158

work page 2023
[16]

Kades, J.M

L. Kades, J.M. Pawlowski, A. Rothkopf, M. Scherzer, J.M. Urban, S.J. Wetzel et al.,Spectral Reconstruction with Deep Neural Networks,Phys. Rev. D102(2020) 096001 [1905.04305]

work page arXiv 2020
[17]

Del Debbio, M

L. Del Debbio, M. Naviglio and F. Tarantelli,Neural Networks Asymptotic Behaviours for the Resolution of Inverse Problems,2402.09338

work page arXiv
[18]

Buzzicotti, A

M. Buzzicotti, A. De Santis and N. Tantalo,Teaching to extract spectral densities from lattice correlators to a broad audience of learning-machines,Eur. Phys. J. C84(2024) 32 [2307.00808]

work page arXiv 2024
[19]

Wetzel and M

S.J. Wetzel and M. Scherzer,Machine Learning of Explicit Order Parameters: From the Ising Model to SU(2) Lattice Gauge Theory,Phys. Rev. B96(2017) 184410 [1705.05582]. – 32 –

work page arXiv 2017
[20]

Bachtis, G

D. Bachtis, G. Aarts and B. Lucini,Extending machine learning classification capabilities with histogram reweighting,Phys. Rev. E102(2020) 033303 [2004.14341]

work page arXiv 2020
[21]

Bachtis, G

D. Bachtis, G. Aarts and B. Lucini,Adding machine learning within Hamiltonians: Renormalization group transformations, symmetry breaking and restoration,Phys. Rev. Res. 3(2021) 013134 [2010.00054]

work page arXiv 2021
[22]

Karsch, A

F. Karsch, A. Lahiri, M. Neumann and C. Schmidt,A machine learning approach to the classification of phase transitions in many flavor QCD,PoSLATTICE2022(2023) 027 [2211.16232]

work page arXiv 2023
[23]

Shanahan, A

P.E. Shanahan, A. Trewartha and W. Detmold,Machine learning action parameters in lattice quantum chromodynamics,Phys. Rev. D97(2018) 094506 [1801.05784]

work page arXiv 2018
[24]

Cranmer, G

K. Cranmer, G. Kanwar, S. Racani` ere, D.J. Rezende and P.E. Shanahan,Advances in machine-learning-based sampling motivated by lattice quantum chromodynamics,Nature Rev. Phys.5(2023) 526 [2309.01156]

work page arXiv 2023
[25]

Lawrence,Machine-learning approaches to accelerating lattice simulations,PoS LATTICE2024(2025) 010 [2502.02670]

S. Lawrence,Machine-learning approaches to accelerating lattice simulations,PoS LATTICE2024(2025) 010 [2502.02670]

work page arXiv 2025
[26]

B.J. Choi, H. Ohno and A. Tomiya,Machine Learning-Based Estimation of Cumulants of Chiral Condensate via Multi-Ensemble Reweighting with Deborah.jl, in42th International Symposium on Lattice Field Theory, 2, 2026 [2602.21617]

work page arXiv 2026
[27]

Ferrenberg and R.H

A.M. Ferrenberg and R.H. Swendsen,New Monte Carlo Technique for Studying Phase Transitions,Phys. Rev. Lett.61(1988) 2635

work page 1988
[28]

Ferrenberg and R.H

A.M. Ferrenberg and R.H. Swendsen,Optimized Monte Carlo analysis,Phys. Rev. Lett.63 (1989) 1195

work page 1989
[29]

Cuteri, O

F. Cuteri, O. Philipsen and A. Sciarra,On the order of the QCD chiral phase transition for different numbers of quark flavours,JHEP11(2021) 141 [2107.12739]

work page arXiv 2021
[30]

Pisarski and F

R.D. Pisarski and F. Wilczek,Remarks on the Chiral Phase Transition in Chromodynamics, Phys. Rev. D29(1984) 338

work page 1984
[31]

Brown, F.P

F.R. Brown, F.P. Butler, H. Chen, N.H. Christ, Z.-h. Dong, W. Schaffer et al.,On the existence of a phase transition for QCD with three light quarks,Phys. Rev. Lett.65(1990) 2491

work page 1990
[32]

Iwasaki, K

Y. Iwasaki, K. Kanaya, S. Kaya, S. Sakai and T. Yoshie,Finite temperature transitions in lattice QCD with Wilson quarks: Chiral transitions and the influence of the strange quark, Phys. Rev. D54(1996) 7010 [hep-lat/9605030]

work page arXiv 1996
[33]

Karsch, E

F. Karsch, E. Laermann and C. Schmidt,The Chiral critical point in three-flavor QCD, Phys. Lett. B520(2001) 41 [hep-lat/0107020]

work page arXiv 2001
[34]

Klinger, R

J.P. Klinger, R. Kaiser and O. Philipsen,The order of the chiral phase transition in massless many-flavour lattice QCD,PoSLATTICE2024(2025) 172 [2501.19251]

work page arXiv 2025
[35]

Klinger, R

J.P. Klinger, R. Kaiser, O. Philipsen and J. Schaible,On the phase structure of massless many-flavour QCD with staggered fermions, in42th International Symposium on Lattice Field Theory, 3, 2026 [2603.20099]

work page arXiv 2026
[36]

D’Ambrosio, O

A. D’Ambrosio, O. Philipsen and R. Kaiser,The chiral phase transition at non-zero imaginary baryon chemical potential for different numbers of quark flavours,PoS LATTICE2022(2023) 172 [2212.03655]. – 33 –

work page arXiv 2023
[37]

D’Ambrosio, M

A. D’Ambrosio, M. Fromm, R. Kaiser and O. Philipsen,On the nature of the QCD chiral phase transition with imaginary chemical potential,2512.15418

work page arXiv
[38]

Neumann,Chiral phase transition in QCD with five degenerate quark flavors: Lattice simulations and machine learning approaches, Ph.D

M. Neumann,Chiral phase transition in QCD with five degenerate quark flavors: Lattice simulations and machine learning approaches, Ph.D. thesis, Bielefeld U., 2023. 10.4119/unibi/2983242

work page doi:10.4119/unibi/2983242 2023
[39]

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner et al.,An image is worth 16x16 words: Transformers for image recognition at scale,2010.11929

work page internal anchor Pith review Pith/arXiv arXiv 2010
[40]

L.-G. Pang, K. Zhou, N. Su, H. Petersen, H. St¨ ocker and X.-N. Wang,An equation-of-state-meter of quantum chromodynamics transition from deep learning,Nature Commun.9(2018) 210 [1612.04262]

work page arXiv 2018
[41]

MADE: masked autoencoder for distribution estimation

M. Germain, K. Gregor, I. Murray and H. Larochelle,MADE: Masked Autoencoder for Distribution Estimation,1502.03509

work page arXiv
[42]

2017, arXiv e-prints, arXiv:1705.07057

G. Papamakarios, T. Pavlakou and I. Murray,Masked Autoregressive Flow for Density Estimation,1705.07057

work page arXiv
[43]

Sciarra, C

A. Sciarra, C. Pinke, M. Bach, F. Cuteri, L. Zeidlewicz, C. Sch¨ afer et al., “Cl2qcd.” https://doi.org/10.5281/zenodo.5121917, Feb., 2021. 10.5281/zenodo.5121917

work page doi:10.5281/zenodo.5121917 2021
[44]

Kennedy, I

A.D. Kennedy, I. Horvath and S. Sint,A New exact method for dynamical fermion computations with nonlocal actions,Nucl. Phys. B Proc. Suppl.73(1999) 834 [hep-lat/9809092]

work page arXiv 1999
[45]

Clark and A.D

M.A. Clark and A.D. Kennedy,Accelerating dynamical fermion computations using the rational hybrid Monte Carlo (RHMC) algorithm with multiple pseudofermion fields,Phys. Rev. Lett.98(2007) 051601 [hep-lat/0608015]

work page arXiv 2007
[46]

X.-Y. Jin, Y. Kuramashi, Y. Nakamura, S. Takeda and A. Ukawa,Critical point phase transition for finite temperature 3-flavor QCD with non-perturbatively O(a) improved Wilson fermions atN t = 10,Phys. Rev. D96(2017) 034523 [1706.01178]

work page arXiv 2017
[47]

Takeda, X.-Y

S. Takeda, X.-Y. Jin, Y. Kuramashi, Y. Nakamura and A. Ukawa,Update on Nf=3 finite temperature QCD phase structure with Wilson-Clover fermion action,PoSLATTICE2016 (2017) 384 [1612.05371]

work page arXiv 2017
[48]

Critical phenomena and renormalization group theory,

A. Pelissetto and E. Vicari,Critical phenomena and renormalization group theory,Phys. Rept.368(2002) 549 [cond-mat/0012164]

work page arXiv 2002
[49]

Bengio and S

Y. Bengio and S. Bengio,Modeling high-dimensional discrete data with multi-layer neural networks, inAdvances in Neural Information Processing Systems, S. Solla, T. Leen and K. M¨ uller, eds., vol. 12, MIT Press, 1999, https://proceedings.neurips.cc/paper files/paper/1999/file/e6384711491713d29bc63fc5eeb5ba4f- Paper.pdf

work page 1999
[50]

Rumelhart, G.E

D.E. Rumelhart, G.E. Hinton and R.J. Williams,Learning representations by back-propagating errors,Nature323(1986) 533

work page 1986
[51]

Hinton and R.R

G.E. Hinton and R.R. Salakhutdinov,Reducing the Dimensionality of Data with Neural Networks,Science313(2006) 1127647

work page 2006
[52]

Jimenez Rezende and S

D.J. Rezende and S. Mohamed,Variational Inference with Normalizing Flows, 5, 2015 [1505.05770]. – 34 –

work page arXiv 2015
[53]

Minka et al.,Divergence measures and message passing, Technical Report MSR-TR-2005-173, Microsoft Research (2005)

T. Minka et al.,Divergence measures and message passing, Technical Report MSR-TR-2005-173, Microsoft Research (2005)

work page 2005
[54]

Hackett, C.-C

D.C. Hackett, C.-C. Hsieh, S. Pontula, M.S. Albergo, D. Boyda, J.-W. Chen et al., Flow-based sampling for multimodal and extended-mode distributions in lattice field theory, 2107.00734

work page arXiv
[55]

Nicoli, C.J

K.A. Nicoli, C.J. Anders, T. Hartung, K. Jansen, P. Kessel and S. Nakajima,Detecting and mitigating mode-collapse for flow-based sampling of lattice field theories,Phys. Rev. D108 (2023) 114501 [2302.14082]

work page arXiv 2023
[56]

Hotelling,Analysis of a complex of statistical variables into principal components,Journal of Educational Psychology24(1933) 417

H. Hotelling,Analysis of a complex of statistical variables into principal components,Journal of Educational Psychology24(1933) 417

work page 1933
[57]

Wilding and A.D

N.B. Wilding and A.D. Bruce,Density fluctuations and field mixing in the critical fluid, Journal of Physics: Condensed Matter4(1992) 3087

work page 1992
[58]

Rummukainen, M

K. Rummukainen, M. Tsypin, K. Kajantie, M. Laine and M.E. Shaposhnikov,The Universality class of the electroweak theory,Nucl. Phys. B532(1998) 283 [hep-lat/9805013]

work page arXiv 1998
[59]

Kajantie, M

K. Kajantie, M. Laine, K. Rummukainen and M. Shaposhnikov,A non-perturbative analysis of the finite-t phase transition in su (2)×u (1) electroweak theory,Nuclear Physics B493 (1997) 413

work page 1997
[60]

TensorFlow: Large-scale machine learning on heterogeneous systems

M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro et al., “TensorFlow: Large-scale machine learning on heterogeneous systems.”https://www.tensorflow.org/, 2015

work page 2015
[61]

Bishop,Pattern Recognition and Machine Learning, Springer, New York, NY (2006)

C.M. Bishop,Pattern Recognition and Machine Learning, Springer, New York, NY (2006). [62]Code and data for figures available at https: // github. com/ ssimrandr/ lqcd-density-interpolation-maf, 2026

work page 2006
[62]

Kullback and R.A

S. Kullback and R.A. Leibler,On information and sufficiency,The Annals of Mathematical Statistics22(1951) 79

work page 1951
[63]

Gretton, K.M

A. Gretton, K.M. Borgwardt, M.J. Rasch, B. Sch¨ olkopf and A. Smola,A kernel two-sample test,Journal of Machine Learning Research13(2012) 723

work page 2012
[64]

Maximum mean discrepancy (mmd) in machine learning

O. Tunali, “Maximum mean discrepancy (mmd) in machine learning.” https://www.onurtunali.com/ml/2019/03/08/ maximum-mean-discrepancy-in-machine-learning.html, 2019

work page 2019
[65]

Waskom,seaborn: statistical data visualization,Journal of Open Source Software6 (2021) 3021

M.L. Waskom,seaborn: statistical data visualization,Journal of Open Source Software6 (2021) 3021. – 35 –

work page 2021