pith. sign in

arxiv: 2506.08080 · v2 · submitted 2025-06-09 · ✦ hep-ph · cs.LG· physics.comp-ph· stat.ML

Towards AI-assisted Neutrino Flavor Theory Design

Jason Benjamin Baretz , Max Fieg , Vijay Ganesh , Aishik Ghosh , V. Knapp-Perez , Jake Rudolph , Daniel Whiteson This is my paper

Pith reviewed 2026-05-19 10:21 UTC · model grok-4.3

classification ✦ hep-ph cs.LGphysics.comp-phstat.ML
keywords neutrino flavor modelsreinforcement learningmodel buildingsymmetry groupsparticle representationsautonomous model builderflavor mixingtheory space
0
0 comments X p. Extension

The pith

A reinforcement learning agent constructs viable neutrino flavor models by choosing symmetry groups and representations to minimize free parameters.

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

The paper introduces AMBer, a framework where a reinforcement learning agent interacts with a physics software pipeline to explore possible symmetry groups, particle contents, and representation assignments for neutrino flavor theories. This replaces much of the manual intuition-driven model building with an automated search that favors models using fewer free parameters while still matching data. A sympathetic reader would care because the space of possible models is huge and current methods depend heavily on individual theorist choices. The work validates the approach on familiar cases and applies it to an unexplored symmetry group. If the method works, it could make systematic exploration of theory space routine rather than ad hoc.

Core claim

AMBer is a framework in which a reinforcement learning agent interacts with a streamlined physics software pipeline to search model spaces efficiently, constructing viable neutrino flavor models while minimizing the number of free parameters introduced. The approach is validated in well-studied regions of theory space and extended to a novel symmetry group.

What carries the argument

The Autonomous Model Builder (AMBer): a reinforcement learning agent that selects symmetry groups, particle content, and group representation assignments using feedback from a physics software pipeline.

If this is right

  • Viable neutrino flavor models become reachable through automated search rather than exhaustive manual enumeration.
  • The same pipeline can be pointed at previously unexamined symmetry groups to generate candidate models.
  • Models produced this way are explicitly constructed to use fewer free parameters than typical hand-built alternatives.
  • The reinforcement learning plus software feedback pattern can be reused for other particle theory construction tasks.

Where Pith is reading between the lines

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

  • An RL agent free of human priors could surface viable models that theorists have not yet considered.
  • Coupling the same agent to more precise simulation codes would allow direct comparison against upcoming neutrino oscillation data.
  • The framework offers a template for autonomous model building in adjacent areas such as lepton number violation or dark sector theories.

Load-bearing premise

The streamlined physics software pipeline must supply accurate and unbiased signals about model viability and the number of free parameters without missing important constraints or adding artifacts.

What would settle it

Test whether the trained agent rediscovers a set of previously known viable neutrino models when started from scratch and whether its proposed new models survive full experimental fits without extra parameters.

Figures

Figures reproduced from arXiv: 2506.08080 by Aishik Ghosh, Daniel Whiteson, Jake Rudolph, Jason Benjamin Baretz, Max Fieg, Vijay Ganesh, V. Knapp-Perez.

Figure 1
Figure 1. Figure 1: A diagram illustrating how the reinforcement learning agent, AMBer, searches the space of models, taking actions to modify the model. Each new model is then evaluated using a pipeline of physics software, which produces a reward depending on the χ 2 of the fit to data and the number of model parameters. The reward and model inform the agent’s selection of the next action. This structure could be generalize… view at source ↗
Figure 2
Figure 2. Figure 2: Training variables of interest over time for searches in three spaces: A4 ×Z4 (top), A4 ×ZN (middle), and T19 ×Z4 (bottom). The left column shows the evolution of χ 2 in blue (where the curve indicates the median log10 χ 2 over all environments) and the mean number of parameters ⟨np⟩ as training progresses in orange. The right column shows the number of valid models in orange and good (χ 2 ≤ 10 and np ≤ 7)… view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of the order of ZN versus training step, with (left) and without (right) the penalty RZ that encourages higher orders in models [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: A single environment trajectory in the A4 ×Z4 latent space over the course of the full run. Each panel displays the distribution of models in a different quartile of the run (where t=1 represents 1000 timesteps). Early on the agent searches more broadly throughout the space before honing in on specific promising regions. Good models (χ 2 ≤ 10 and np ≤ 7) are indicated by magenta diamonds, showing that the … view at source ↗
Figure 5
Figure 5. Figure 5: Number of parameters np and χ 2 for a representative distribution of found models for the three theory spaces searched: A4 ×Z4(left), A4 ×ZN (middle), and T19 ×Z4 (right). The region within the dashed black lines contains models with ≤ 7 parameters, and good fits, χ 2 ≤ 10. shows the distribution of models found for a separate run that did not have the RZ(N) in the reward function and AMBer is found to foc… view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of the number of flavon fields (upper left), the A4 representations (upper right), and vacuum alignments (lower) in good models found in the search of the A4 ×Z4 space. The vacuum alignment is normalized by the flavon breaking scale, 0.1Λ, i.e. ⟨φ⟩/(0.1Λ) as in equation (4). certain flexibility in this search space, as there are generally many ways to form group invariant interactions, particu… view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of the number of flavon fields (top left), the A4 representations (top right), the distribution of vacuum alignments (bottom left) and the ZN symmetry (bottom right) in good models found in the search of the A4 ×ZN space. For comparison, the bottom right panel shows a configuration with and without the RZ(N) reward in equation (8), and AMBer is found to be more efficient at exploring large N w… view at source ↗
Figure 8
Figure 8. Figure 8: Distribution of the number of flavon fields (upper left), the T19 representations (upper right), and the distribution of vacuum alignments (lower) of good models found in the search of the T19 ×Z4 space. The vacuum alignment is normalized by the flavon breaking scale, 0.1Λ, i.e. ⟨⃗φ⟩/(0.1Λ) as in equation (4). (9) where αˆ (C) , αˆ (M) and αˆ (D) are constants that can be factored out, because only dimensi… view at source ↗
Figure 9
Figure 9. Figure 9: A particular model found in the T19 ×Z4 search with the star indicating the best fit point. A Markov-Chain Monte-Carlo is performed near the best fit to map the χ 2 distribution. The left panel shows the χ 2 distribution and correlations of the model’s parameters, the right panel shows the distribution of the model’s predictions, along with the corresponding predictions with 2σ error bars. pipeline. Existi… view at source ↗
Figure 10
Figure 10. Figure 10: Evolution of the reward function for different targets χ 2 target and n target p terms of equation (5). As the targets shift, the requirements for a terminal state become more stringent, allowing the agent to focus on finding valid models early and good models late. In all plots, the solid blue line indicates the threshold, the solid black line the invalid model penalty crank, and the red line the invalid… view at source ↗
Figure 11
Figure 11. Figure 11: Neural network level metrics that demonstrate how the agent is learning from an individual A4 ×Z4 run. The policy loss is shown in the top left panel. When this loss decreases, it indicates the agent is selecting more advantageous actions. The top right panel shows the entropy loss. The increase in this term indicates that the agent is focusing on certain actions more, however the use of this term in the … view at source ↗
Figure 12
Figure 12. Figure 12: The left panel shows 500k background points are used to visualize the A4 ×Z4 latent space. Contour lines generated via kernel density estimation denote the approximate area the agent searched. Good models (χ 2 ≤ 10 and np ≤ 7) are included by magenta diamonds for reference. The right panel shows a 2D histogram of the full run in the same latent space. While the agent spends the majority of its time honing… view at source ↗
read the original abstract

Particle physics theories, such as those which explain neutrino flavor mixing, arise from a vast landscape of model-building possibilities. A model's construction typically relies on the intuition of theorists. It also requires considerable effort to identify appropriate symmetry groups, assign field representations, and extract predictions for comparison with experimental data. We develop an Autonomous Model Builder (AMBer), a framework in which a reinforcement learning agent interacts with a streamlined physics software pipeline to search these spaces efficiently. AMBer selects symmetry groups, particle content, and group representation assignments to construct viable models while minimizing the number of free parameters introduced. We validate our approach in well-studied regions of theory space and extend the exploration to a novel, previously unexamined symmetry group. While demonstrated in the context of neutrino flavor theories, this approach of reinforcement learning with physics software feedback may be extended to other theoretical model-building problems in the future.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces AMBer (Autonomous Model Builder), a reinforcement learning framework in which an agent interacts with a streamlined physics software pipeline to select symmetry groups, particle content, and representation assignments for neutrino flavor models. The central goal is to construct viable models while minimizing the number of free parameters. The approach is validated in well-studied regions of theory space and extended to exploration of a novel, previously unexamined symmetry group.

Significance. If the RL agent reliably identifies models with demonstrably fewer free parameters than those constructed by hand in the literature, the framework could meaningfully assist theorists in navigating the large space of possible neutrino flavor models. The integration of external physics software for independent viability feedback is a constructive direction that could generalize to other model-building problems. However, the absence of quantitative benchmarks currently limits the assessed impact.

major comments (2)
  1. [Abstract] Abstract: the validation in well-studied regions is stated without quantitative metrics, success rates, achieved free-parameter counts, or side-by-side comparisons against established literature models (e.g., standard A4 or S4 assignments). This leaves open whether AMBer systematically improves on or merely rediscovers known minimal constructions.
  2. [Abstract] Abstract and extension section: the exploration of the novel symmetry group reports no specific results on model viability, satisfied experimental constraints, or the number of free parameters obtained, so the claim of successful extension cannot be evaluated.
minor comments (2)
  1. The reward function and the precise heuristic used by the physics pipeline to count free parameters should be specified explicitly to permit reproducibility and to rule out implicit bias toward already-known assignments.
  2. A table tabulating parameter counts for AMBer-generated models versus representative literature models in the validated regions would strengthen the minimization claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for their detailed and constructive feedback on our manuscript. Their comments have helped us identify areas where the presentation of our results can be strengthened. We respond to each major comment below and indicate the revisions made to the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the validation in well-studied regions is stated without quantitative metrics, success rates, achieved free-parameter counts, or side-by-side comparisons against established literature models (e.g., standard A4 or S4 assignments). This leaves open whether AMBer systematically improves on or merely rediscovers known minimal constructions.

    Authors: We thank the referee for this observation. While the abstract is intended to be concise, we agree that it should include key quantitative indicators to allow immediate assessment of the validation. The main text already contains detailed results on success rates for rediscovering minimal models, achieved free-parameter counts, and direct comparisons with standard A4 and S4 constructions from the literature. In the revised version, we have updated the abstract to summarize these metrics and comparisons. revision: yes

  2. Referee: [Abstract] Abstract and extension section: the exploration of the novel symmetry group reports no specific results on model viability, satisfied experimental constraints, or the number of free parameters obtained, so the claim of successful extension cannot be evaluated.

    Authors: We appreciate the referee highlighting the need for more explicit details on the novel symmetry group. This extension demonstrates the framework's ability to explore unexamined groups while using the physics pipeline for viability feedback. We acknowledge that specific quantitative outcomes were not sufficiently elaborated in the section. We have revised the extension section to include concrete examples of constructed models, their satisfaction of experimental neutrino constraints, and the minimized free-parameter counts obtained by the agent. revision: yes

Circularity Check

0 steps flagged

No circularity detected in derivation or claims

full rationale

The paper introduces AMBer as an RL-based framework that interacts with an external, streamlined physics software pipeline to evaluate model viability and count free parameters. This feedback mechanism is presented as independent of the agent's internal definitions, with validation performed against well-studied regions of theory space and extension to a novel symmetry group. No equations, fitted parameters, or self-citations are shown to reduce the central results to inputs by construction. The approach does not rename known results, smuggle ansatzes via prior work, or import uniqueness theorems from the authors' own citations in a load-bearing way. The derivation chain relies on external software feedback rather than self-referential fitting or prediction, making the framework self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the effectiveness of the RL search and the reliability of the physics evaluation pipeline; no free parameters or invented entities are explicitly introduced beyond the framework itself.

axioms (1)
  • domain assumption The streamlined physics software pipeline accurately evaluates model predictions, viability, and free parameter count for any proposed symmetry and representation assignment.
    This is invoked when the agent interacts with the pipeline to receive feedback during model construction and validation.
invented entities (1)
  • AMBer (Autonomous Model Builder) framework no independent evidence
    purpose: To automate selection of symmetry groups, particle content, and representations for neutrino flavor models via RL.
    New tool and workflow introduced by the authors to address the model-building landscape.

pith-pipeline@v0.9.0 · 5700 in / 1409 out tokens · 84949 ms · 2026-05-19T10:21:40.931685+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages · 11 internal anchors

  1. [1]

    Fox, P. J.et al. Tf08 snowmass report: Bsm model building (2022). 2210.03075

    work page arXiv 2022

  2. [2]

    Feruglio and A

    Feruglio, F. & Romanino, A. Lepton flavour symmetries (2021). 1912.06028

    work page arXiv 2021

  3. [3]

    Neutrino Flavor Model Building and the Origins of Flavor and CP Violation

    Almumin, Y .et al. Neutrino Flavor Model Building and the Origins of Flavor and CP Violation. Universe 9, 512, DOI: 10.3390/universe9120512 (2023). 2204.08668

    work page doi:10.3390/universe9120512 2023

  4. [4]

    de Gouvêa, A. et al. Theory of neutrino physics – snowmass tf11 (aka nf08) topical group report (2022). 2209.07983

    work page arXiv 2022

  5. [5]

    & Ruehle, F

    Halverson, J., Nelson, B. & Ruehle, F. Branes with brains: exploring string vacua with deep reinforcement learning. J. High Energy Phys. 2019, DOI: 10.1007/jhep06(2019)003 (2019)

    work page doi:10.1007/jhep06(2019)003 2019

  6. [6]

    & Otsuka, H

    Nishimura, S., Miyao, C. & Otsuka, H. Exploring the flavor structure of quarks and leptons with reinforcement learning. JHEP 23, 021, DOI: 10.1007/JHEP12(2023)021 (2020). 2304.14176

    work page doi:10.1007/jhep12(2023)021 2023

  7. [7]

    Harvey, T. R. & Lukas, A. Quark Mass Models and Reinforcement Learning. JHEP 08, 161, DOI: 10.1007/JHEP08(2021) 161 (2021). 2103.04759

    work page doi:10.1007/jhep08(2021 2021

  8. [8]

    & Otsuka, H

    Nishimura, S., Miyao, C. & Otsuka, H. Reinforcement learning-based statistical search strategy for an axion model from flavor. (2024). 2409.10023

    work page arXiv 2024

  9. [9]

    N., Eu, S

    Wojcik, G. N., Eu, S. T. & Everett, L. L. Graph Reinforcement Learning for Exploring BSM Model Spaces. (2024). 2407.07203

    work page arXiv 2024

  10. [10]

    & He, Y .-H

    Carta, F., Gauntlett, A., Griffin, F. & He, Y .-H. BPS spectroscopy with reinforcement learning. (2025). 2501.14863

    work page arXiv 2025

  11. [11]

    & Feruglio, F

    Altarelli, G. & Feruglio, F. Discrete flavor symmetries and models of neutrino mixing. Rev. Mod. Phys. 82, 2701–2729, DOI: 10.1103/revmodphys.82.2701 (2010)

    work page doi:10.1103/revmodphys.82.2701 2010

  12. [12]

    Softly Broken A_4 Symmetry for Nearly Degenerate Neutrino Masses

    Ma, E. & Rajasekaran, G. Softly broken A(4) symmetry for nearly degenerate neutrino masses. Phys. Rev. D 64, 113012, DOI: 10.1103/PhysRevD.64.113012 (2001). hep-ph/0106291

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1103/physrevd.64.113012 2001

  13. [13]

    Tri-Bimaximal Neutrino Mixing, A4 and the Modular Symmetry

    Altarelli, G. & Feruglio, F. Tri-bimaximal neutrino mixing, A(4) and the modular symmetry. Nucl. Phys. B 741, 215–235, DOI: 10.1016/j.nuclphysb.2006.02.015 (2006). hep-ph/0512103

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1016/j.nuclphysb.2006.02.015 2006

  14. [14]

    µ → eγ at a Rate of One Out of109 Muon Decays? Phys

    Minkowski, P. µ → eγ at a Rate of One Out of109 Muon Decays? Phys. Lett. B 67, 421–428, DOI: 10.1016/0370-2693(77) 90435-X (1977)

    work page doi:10.1016/0370-2693(77 1977

  15. [15]

    Horizontal gauge symmetry and masses of neutrinos

    Yanagida, T. Horizontal gauge symmetry and masses of neutrinos. Conf. Proc. C 7902131, 95–99 (1979)

    work page 1979

  16. [16]

    Glashow, S. L. The Future of Elementary Particle Physics.NATO Sci. Ser. B61, 687, DOI: 10.1007/978-1-4684-7197-7_15 (1980). 24/25

    work page doi:10.1007/978-1-4684-7197-7_15 1980

  17. [17]

    Complex Spinors and Unified Theories

    Gell-Mann, M., Ramond, P. & Slansky, R. Complex Spinors and Unified Theories. Conf. Proc. C 790927, 315–321 (1979). 1306.4669

    work page internal anchor Pith review Pith/arXiv arXiv 1979

  18. [18]

    Ishimori, H. et al. Non-Abelian Discrete Symmetries in Particle Physics. Prog. Theor. Phys. Suppl. 183, 1–163, DOI: 10.1143/PTPS.183.1 (2010). 1003.3552

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1143/ptps.183.1 2010

  19. [19]

    & Romanino, A

    Feruglio, F. & Romanino, A. Lepton flavor symmetries.Rev. Mod. Phys.93, 015007, DOI: 10.1103/RevModPhys.93.015007 (2021). 1912.06028

    work page doi:10.1103/revmodphys.93.015007 2021

  20. [20]

    C., Maltoni, M., Schwetz, T

    Esteban, I., Gonzalez-Garcia, M. C., Maltoni, M., Schwetz, T. & Zhou, A. The fate of hints: updated global analysis of three-flavor neutrino oscillations. JHEP 09, 178, DOI: 10.1007/JHEP09(2020)178 (2020). 2007.14792

    work page doi:10.1007/jhep09(2020)178 2020

  21. [21]

    Unification and Fermion Mass Structure

    Ross, G. & Serna, M. Unification and fermion mass structure. Phys. Lett. B 664, 97–102, DOI: 10.1016/j.physletb.2008.05. 014 (2008). 0704.1248

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1016/j.physletb.2008.05 2008

  22. [22]

    A Model for Tri-bimaximal Mixing from a Completely Broken $A_4$

    Ding, G.-J. & Meloni, D. A Model for Tri-bimaximal Mixing from a Completely Broken A4. Nucl. Phys. B 855, 21–45, DOI: 10.1016/j.nuclphysb.2011.10.001 (2012). 1108.2733

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1016/j.nuclphysb.2011.10.001 2011

  23. [23]

    Ludl, P. O. On the finite subgroups of U(3) of order smaller than 512. J. Phys. A 43, 395204, DOI: 10.1088/1751-8113/43/ 39/395204 (2010). [Erratum: J.Phys.A 44, 139501 (2011)], 1006.1479

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/1751-8113/43/ 2010

  24. [24]

    Natural Vacuum Alignment from Group Theory: The Minimal Case

    Holthausen, M. & Schmidt, M. A. Natural Vacuum Alignment from Group Theory: The Minimal Case. JHEP 01, 126, DOI: 10.1007/JHEP01(2012)126 (2012). 1111.1730

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1007/jhep01(2012)126 2012

  25. [25]

    Navas, S. et al. Review of particle physics. Phys. Rev. D 110, 030001, DOI: 10.1103/PhysRevD.110.030001 (2024)

    work page doi:10.1103/physrevd.110.030001 2024

  26. [26]

    FlavorPy, DOI: 10.5281/zenodo.11060597 (2024)

    Baur, A. FlavorPy, DOI: 10.5281/zenodo.11060597 (2024)

    work page doi:10.5281/zenodo.11060597 2024

  27. [27]

    Running Neutrino Mass Parameters in See-Saw Scenarios

    Antusch, S., Kersten, J., Lindner, M., Ratz, M. & Schmidt, M. A. Running neutrino mass parameters in see-saw scenarios. JHEP 03, 024, DOI: 10.1088/1126-6708/2005/03/024 (2005). hep-ph/0501272

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/1126-6708/2005/03/024 2005

  28. [28]

    Criado, J. C. & Feruglio, F. Modular Invariance Faces Precision Neutrino Data. SciPost Phys. 5, 042, DOI: 10.21468/ SciPostPhys.5.5.042 (2018). 1807.01125

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  29. [29]

    GAP – Groups, Algorithms, and Programming, Version 4.14.0(2024)

    The GAP Group. GAP – Groups, Algorithms, and Programming, Version 4.14.0(2024)

    work page 2024

  30. [30]

    van Den Broek, P. M. & Cornwell, J. F. Clebsch-gordan coefficients of symmetry groups. physica status solidi (b) 90, 211–224, DOI: https://doi.org/10.1002/pssb.2220900123 (1978). https://onlinelibrary.wiley.com/doi/pdf/10.1002/pssb. 2220900123

    work page doi:10.1002/pssb.2220900123 1978

  31. [31]

    Sutton, R. S. & Barto, A. G. Reinforcement Learning: An Introduction (A Bradford Book, Cambridge, MA, USA, 2018)

    work page 2018

  32. [32]

    Silver, D. et al. Mastering the game of go without human knowledge. Nature 550, 354–359, DOI: 10.1038/nature24270 (2017)

    work page doi:10.1038/nature24270 2017

  33. [33]

    K., Khandelwal, D., Sahu, V

    Inamdar, R., Sundarr, S. K., Khandelwal, D., Sahu, V . D. & Katal, N. A comprehensive review on safe reinforcement learning for autonomous vehicle control in dynamic environments. e-Prime - Adv. Electr. Eng. Electron. Energy 10, 100810, DOI: https://doi.org/10.1016/j.prime.2024.100810 (2024)

    work page doi:10.1016/j.prime.2024.100810 2024

  34. [34]

    Baydin, A. G. et al. Etalumis: bringing probabilistic programming to scientific simulators at scale. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, SC ’19, 1–24, DOI: 10.1145/3295500.3356180 (ACM, 2019)

    work page doi:10.1145/3295500.3356180 2019

  35. [35]

    Proximal Policy Optimization Algorithms

    Schulman, J., Wolski, F., Dhariwal, P., Radford, A. & Klimov, O. Proximal policy optimization algorithms. CoRR abs/1707.06347 (2017). 1707.06347

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  36. [36]

    & Vaudrevange, P

    Mütter, A., Parr, E. & Vaudrevange, P. K. Deep learning in the heterotic orbifold landscape.Nucl. Phys. B 940, 113–129, DOI: 10.1016/j.nuclphysb.2019.01.013 (2019)

    work page doi:10.1016/j.nuclphysb.2019.01.013 2019

  37. [37]

    & Whiteson, D

    Baretz, J., Carrara, N., Hollingsworth, J. & Whiteson, D. Visualization and efficient generation of constrained high- dimensional theoretical parameter spaces. J. High Energy Phys. 2023, 62 (2023)

    work page 2023

  38. [38]

    & Giryes, R

    Bank, D., Koenigstein, N. & Giryes, R. Autoencoders, 353–374 (Springer International Publishing, Cham, 2023)

    work page 2023

  39. [39]

    Cognitive Science , author =

    Ackley, D. H., Hinton, G. E. & Sejnowski, T. J. A learning algorithm for boltzmann machines. Cogn. Sci. 9, 147–169, DOI: https://doi.org/10.1016/S0364-0213(85)80012-4 (1985). 25/25

    work page doi:10.1016/s0364-0213(85)80012-4 1985