Pith. sign in

REVIEW 3 major objections 1 cited by

Jets of initial data for BKM PDEs can be matched to arbitrary order by finite-gap solutions built from Stäckel systems.

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

Latent Grammar Flow embeds grammar-based ODE representations into a discrete latent space with a behavioural loss and samples candidate equations via discrete flow to fit observed data.

T0 review reviewed 2026-07-15 challenge →

load-bearing objection Cache mismatch: abstract promises LGF for ODE discovery; full text is a different math paper on BKM finite-gap jets, so the claimed method cannot be evaluated. the 3 major comments →

arxiv 2604.16232 v2 pith:5Y7AWNYE submitted 2026-04-17 cs.LG cs.AIcs.CEcs.SC

Neuro-Symbolic ODE Discovery with Latent Grammar Flow

classification cs.LG cs.AIcs.CEcs.SC MSC 37K1035Q5337J3558J53
keywords finite-gap solutionsBKM systemsStäckel systemsjet surjectivityKdVCamassa–HolmNijenhuis operatorsintegrable PDEs
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper proves that for large classes of integrable evolutionary PDEs known as BKM systems (including KdV, Kaup–Boussinesq and Camassa–Holm), the k-jet of essentially arbitrary spatial initial data can be realized, for any finite k, by a finite-gap solution. Finite-gap solutions are produced algebraically by a finite-reduction map that sends trajectories of an associated Stäckel integrable system into solutions of the PDE. When the polynomial that labels the BKM class has degree zero the reduction map is triangular, so every jet is attained exactly once N (the number of gaps) is large enough. When the polynomial has degree one and there is a single component, the same construction still covers an open set of jets over the reals and a Zariski-open set over the complexes. The result therefore supplies a concrete sense in which these PDEs are “integrable by explicit solutions”: their local Cauchy data lie in the closure of the finite-gap locus.

Core claim

For BKM systems with deg(m)=0 the finite-reduction map is jet-surjective for every order k once the gap number N is large enough; for deg(m)=1 and n=1 the image of the map from Stäckel initial data to k-jets is open (and Zariski-dense over C). Consequently every sufficiently regular initial datum can be approximated, jet-wise, by a finite-gap solution.

What carries the argument

The finite-reduction map R that algebraically embeds solutions of an N-dimensional Stäckel system (built from a monic polynomial c of degree 2N+n and the BKM data) into solutions of the BKM PDE; its triangular structure when deg(m)=0, and a graded analysis of the resulting Taylor-coefficient map when deg(m)=1, are what establish jet-surjectivity.

Load-bearing premise

That the algebraic conditions defining the finite-reduction map, once solved for the companion coordinates, leave enough free parameters (the Stäckel initial values) to control every jet coefficient once N is large.

What would settle it

For the Camassa–Holm case, compute the explicit polynomial map from (w(0),p(0)) to the 3-jet of u and check whether its image is open (or compute the Jacobian rank on a dense set); if the image has positive-codimension components the open-set claim fails.

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

If this is right

  • Any analytic initial datum for KdV or Kaup–Boussinesq is the jet-limit of a sequence of finite-gap solutions of increasing gap number.
  • Local well-posedness statements for these PDEs can be tested first on the dense subclass of finite-gap data, which admit explicit theta-function formulae.
  • The same reduction-map technique yields jet-density for any other PDE family that admits an analogous Stäckel embedding.
  • Numerical integrators can be validated by comparing their jets against the algebraically exact finite-gap jets.

Where Pith is reading between the lines

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

  • The triangular grading used for deg(m)=0 suggests a recursive algorithm that, given any jet, produces an explicit finite-gap initial condition realizing it.
  • If the open-set result for Camassa–Holm can be upgraded to full surjectivity, the same proof strategy would likely extend to multi-component systems with deg(m)=1.
  • Uniform convergence of the finite-gap sequence (beyond jets) would give a constructive approximation theory for the full Cauchy problem.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 0 minor

Summary. The abstract announces Latent Grammar Flow (LGF), a neuro-symbolic generative method that embeds grammar-based ODE representations in a discrete latent space, uses a behavioural loss to cluster semantically similar equations, and samples candidates with a discrete flow model (optionally conditioned on domain constraints such as stability). The body of the provided manuscript is an unrelated pure-mathematics paper (arXiv:2604.16233, math.AP): it studies jet-density of finite-gap solutions for Bolsinov–Konyaev–Matveev (BKM) systems of PDEs, constructs finite-reduction maps from Stäckel systems, and proves jet-surjectivity results (Theorems 1–2) for deg(m)=0 (e.g. KdV, Kaup–Boussinesq) and for deg(m)=1, n=1 on an open set (e.g. Camassa–Holm). No LGF architecture, loss definitions, algorithms, datasets, baselines, or experiments appear anywhere in the text.

Significance. The claimed LGF contribution cannot be assessed: none of its components (grammar latents, behavioural loss, discrete flow, ODE-discovery experiments) are present. The mathematical content that is present is a specialized, carefully structured result on formal jet approximation by finite-gap solutions in the BKM/Nijenhuis-geometry setting and may be of interest to the integrable-systems community, but it is not the paper described by the title and abstract under review. Consequently the submission as provided does not establish any of the neuro-symbolic claims.

major comments (3)
  1. Abstract vs. full text: the abstract and paper_id claim a cs.LG neuro-symbolic ODE-discovery framework (LGF); the entire manuscript body (title page through Appendix B) is instead the math.AP paper on jet-density of finite-gap solutions for BKM systems. No section, equation, theorem, or experiment belonging to LGF exists in the document. The central claims of the abstract are therefore completely unsupported by the submitted text and cannot be refereed.
  2. Because the body is a different paper, load-bearing elements required for any method paper of this type—definition of the behavioural loss, discrete latent grammar, flow model, training procedure, data-fitting objective, validity under grammar/constraints, and empirical evaluation—are absent. There is nothing to verify against the abstract’s strongest claim that behavioural loss plus discrete flow yields valid, data-fitting ODEs.
  3. Internal identifiers confirm the mismatch: the body carries arXiv:2604.16233v1 [math.AP] and theorems about Stäckel systems, finite-reduction maps R, and jet-surjectivity (Theorems 1–2, §§3–6), none of which relate to Latent Grammar Flow or ODE discovery from data.

Circularity Check

0 steps flagged

No circularity in the supplied manuscript; it is a self-contained pure-math derivation of jet-surjectivity for BKM finite-gap solutions, unrelated to the LGF abstract.

full rationale

The CACHEABLE full text is the arXiv:2604.16233 manuscript on jet-density of finite-gap solutions for BKM systems (Theorems 1–2 on approximating k-jets of initial data via the finite-reduction map R from Stäckel systems). Its derivation chain proceeds from the algebraic definition of R (divisibility and degree conditions on σ_u(μ)·σ_w(μ)^2 − c(μ)), the triangular structure for deg(m)=0 (Proposition 1), grading of Hamiltonians/potentials (Section 4), and direct recursive calculations showing the map from (w(0),p(0)) to j^k_0 u is surjective (or open) (Sections 5–6). These steps are independent algebraic/ODE arguments; they do not redefine outputs as inputs, fit parameters then re-label them predictions, or rest load-bearing uniqueness on unverified self-citations. Prior BKM/Stäckel constructions are cited externally. Because the text contains none of the LGF grammar embeddings, behavioural loss, or discrete flow of the claimed abstract, no circular reduction for that method can be exhibited. Honest finding: score 0, empty steps.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 2 invented entities

Abstract-only ledger. The paper postulates a neuro-symbolic pipeline whose load-bearing pieces are a grammar for ODEs, a discrete latent embedding with behavioural similarity, and a discrete flow sampler. No free parameters, proofs, or independent evidence are given in the abstract. Domain assumptions include that target systems are well-modeled by ODEs expressible in the chosen grammar and that data suffice to identify them.

axioms (3)
  • domain assumption Target dynamics can be expressed as ordinary differential equations generable by a formal grammar over a chosen operator/basis set.
    Implicit in the claim that grammar-based representations and recursive generation recover the governing ODEs from data.
  • ad hoc to paper A behavioural loss can arrange discrete latent codes so that semantically similar equations are nearby, improving generative search.
    Core design claim of LGF in the abstract; not a standard theorem and not evidenced in the provided text.
  • domain assumption Domain constraints (e.g., stability) can be encoded in grammar rules or as conditional predictors without destroying discoverability of true equations.
    Stated capability in the abstract; success depends on how constraints interact with the search space.
invented entities (2)
  • Latent Grammar Flow (LGF) no independent evidence
    purpose: Neuro-symbolic generative framework that embeds grammar-based ODE representations in a discrete latent space and samples fitting equations via discrete flow.
    Named central method introduced in the abstract; no independent evidence outside this paper is provided in the available text.
  • Behavioural loss for equation embeddings no independent evidence
    purpose: Force semantically similar equations to lie closer in discrete latent space.
    Key mechanism claimed to structure the latent space; definition and validation not available in the abstract.

reviewed 2026-07-15 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Neuro-Symbolic ODE Discovery with Latent Grammar Flow." pith.science (2026). https://pith.science/paper/5Y7AWNYE

@misc{pith2026260416232,
  author       = {Pith},
  title        = {Pith review of: Neuro-Symbolic ODE Discovery with Latent Grammar Flow},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5Y7AWNYE}},
  note         = {Machine review of arXiv:2604.16232}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Understanding natural and engineered systems often relies on symbolic formulations, such as differential equations, which provide interpretability and transferability beyond black-box models. We introduce Latent Grammar Flow (LGF), a neuro-symbolic generative framework for discovering ordinary differential equations from data. LGF embeds equations as grammar-based representations into a discrete latent space and forces semantically similar equations to be positioned closer together with a behavioural loss. Then, a discrete flow model guides the sampling process to recursively generate candidate equations that best fit the observed data. Domain knowledge and constraints, such as stability, can be either embedded into the rules or used as conditional predictors.

Figures

Figures reproduced from arXiv: 2604.16232 by Eleni Chatzi, Georgios Kissas, Karin Yu.

Figure 1
Figure 1. Figure 1: Overview of the training of the GQAE with the semantic loss of the Wasserstein distance [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Violin plots of Benchmark 1, where the distribution is shown in blue and single values as red dots. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The numerical trajectories of the ground truth and predicted ODEs of Benchmark 2. [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The numerical trajectories of the ground truth and predicted ODEs of Benchmark 3. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. Verifier-Guided Model Discovery for Physical Dynamical Systems with Pretrained Symbolic Transformers

    cs.LG 2026-08 conditional novelty 6.0

    Verifier-guided selection lets a pretrained symbolic transformer transfer from synthetic ODEs to high-dimensional cylinder-flow data, recovering symbolic vortex-shedding models that generalize across Reynolds numbers.

Reference graph

Works this paper leans on

42 extracted references · 1 canonical work pages · cited by 1 Pith paper

  1. [1]

    Zembowicz and J

    R. Zembowicz and J. M. Zytkow, ``Discovery of equations: experimental evaluation of convergence,'' in Proceedings of the Tenth National Conference on Artificial Intelligence , ser. AAAI '92. 1em plus 0.5em minus 0.4em AAAI Press, 1992, pp. 70--75, place: San Jose, California

  2. [2]

    Schmidt and H

    M. Schmidt and H. Lipson, `` en Distilling Free - Form Natural Laws from Experimental Data ,'' en Science , vol. 324, no. 5923, pp. 81--85, Apr. 2009. [Online]. Available: https://www.science.org/doi/10.1126/science.1165893

  3. [3]

    R. T. Q. Chen, Y. Rubanova, J. Bettencourt, and D. K. Duvenaud, ``Neural Ordinary Differential Equations ,'' in Advances in Neural Information Processing Systems , S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, Eds., vol. 31. 1em plus 0.5em minus 0.4em Curran Associates, Inc., 2018. [Online]. Available: https://proceedi...

  4. [4]

    Lample and F

    G. Lample and F. Charton, ``Deep Learning for Symbolic Mathematics ,'' 2019, version Number: 1. [Online]. Available: https://arxiv.org/abs/1912.01412

  5. [5]

    d'Ascoli, P.-A

    S. d'Ascoli, P.-A. Kamienny, G. Lample, and F. Charton, ``Deep Symbolic Regression for Recurrent Sequences ,'' in Proceedings of the 39th International Conference on Machine Learning . 1em plus 0.5em minus 0.4em PMLR, 2022, version Number: 2. [Online]. Available: https://arxiv.org/abs/2201.04600

  6. [6]

    Kamienny, G

    P.-A. Kamienny, G. Lample, S. Lamprier, and M. Virgolin, ``Deep Generative Symbolic Regression with Monte - Carlo - Tree - Search ,'' in Proceedings of the 40th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research , A. Krause, E. Brunskill, K. Cho, B. Engelhardt, S. Sabato, and J. Scarlett, Eds., vol. 202. 1em plus ...

  7. [7]

    Becker, M

    S. Becker, M. Klein, A. Neitz, G. Parascandolo, and N. Kilbertus, ``Predicting Ordinary Differential Equations with Transformers ,'' in Proceedings of the 40th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research , A. Krause, E. Brunskill, K. Cho, B. Engelhardt, S. Sabato, and J. Scarlett, Eds., vol. 202. 1em plus 0...

  8. [8]

    J. R. Koza, eng Genetic programming. 1: On the programming of computers by means of natural selection , ser. Complex adaptive systems. 1em plus 0.5em minus 0.4em Cambridge, Mass.: MIT Press, 1992

  9. [9]

    Bongard and H

    J. Bongard and H. Lipson, `` en Automated reverse engineering of nonlinear dynamical systems ,'' en Proceedings of the National Academy of Sciences , vol. 104, no. 24, pp. 9943--9948, Jun. 2007. [Online]. Available: https://pnas.org/doi/full/10.1073/pnas.0609476104

  10. [10]

    I. G. Tsoulos and I. E. Lagaris, `` en Solving differential equations with genetic programming ,'' en Genetic Programming and Evolvable Machines , vol. 7, no. 1, pp. 33--54, Mar. 2006. [Online]. Available: http://link.springer.com/10.1007/s10710-006-7009-y

  11. [11]

    Cranmer, A

    M. Cranmer, A. Sanchez-Gonzalez, P. Battaglia, R. Xu, K. Cranmer, D. Spergel, and S. Ho, ``Discovering Symbolic Models from Deep Learning with Inductive Biases ,'' 2020, version Number: 2. [Online]. Available: https://arxiv.org/abs/2006.11287

  12. [12]

    Cranmer, ``Interpretable Machine Learning for Science with PySR and SymbolicRegression .jl,'' 2023, version Number: 3

    M. Cranmer, ``Interpretable Machine Learning for Science with PySR and SymbolicRegression .jl,'' 2023, version Number: 3. [Online]. Available: https://arxiv.org/abs/2305.01582

  13. [13]

    Brence, L

    J. Brence, L. Todorovski, and S. Džeroski, `` en Probabilistic grammars for equation discovery ,'' en Knowledge-Based Systems , vol. 224, p. 107077, Jul. 2021. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S0950705121003403

  14. [14]

    Omejc, B

    N. Omejc, B. Gec, J. Brence, L. Todorovski, and S. Džeroski, `` en Probabilistic grammars for modeling dynamical systems from coarse, noisy, and partial data ,'' en Machine Learning , vol. 113, no. 10, pp. 7689--7721, Oct. 2024. [Online]. Available: https://link.springer.com/10.1007/s10994-024-06522-1

  15. [15]

    B. K. Petersen, M. Landajuela, T. N. Mundhenk, C. P. Santiago, S. K. Kim, and J. T. Kim, ``Deep symbolic regression: Recovering mathematical expressions from data via risk-seeking policy gradients,'' in Proceedings of the International Conference on Learning Representations ( ICLR ) , 2019. [Online]. Available: https://arxiv.org/abs/1912.04871

  16. [16]

    Crochepierre, L

    L. Crochepierre, L. Boudjeloud, and V. Barbesant, ``A Reinforcement Learning Approach to Domain - Knowledge Inclusion Using Grammar Guided Symbolic Regression ,'' ArXiv, vol. abs/2202.04367, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID:246679968

  17. [17]

    N. Hoai, R. McKay, D. Essam, and R. Chau, ``Solving the symbolic regression problem with tree-adjunct grammar guided genetic programming: the comparative results,'' in Proceedings of the 2002 Congress on Evolutionary Computation . CEC '02 ( Cat . No . 02TH8600 ) , vol. 2. 1em plus 0.5em minus 0.4em Honolulu, HI, USA: IEEE, 2002, pp. 1326--1331. [Online]. ...

  18. [18]

    Sahoo, C

    S. Sahoo, C. Lampert, and G. Martius, ``Learning Equations for Extrapolation and Control ,'' in Proceedings of the 35th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research , J. Dy and A. Krause, Eds., vol. 80. 1em plus 0.5em minus 0.4em PMLR, Jul. 2018, pp. 4442--4450. [Online]. Available: https://proceedings.mlr.p...

  19. [19]

    Garmaev, S

    S. Garmaev, S. Mishra, and O. Fink, `` NOMTO : Neural Operator -based symbolic Model approximaTion and discOvery ,'' 2025, version Number: 1. [Online]. Available: https://arxiv.org/abs/2501.08086

  20. [20]

    S. L. Brunton, J. L. Proctor, and J. N. Kutz, `` en Discovering governing equations from data by sparse identification of nonlinear dynamical systems ,'' en Proceedings of the National Academy of Sciences , vol. 113, no. 15, pp. 3932--3937, Apr. 2016. [Online]. Available: https://pnas.org/doi/full/10.1073/pnas.1517384113

  21. [21]

    Kaheman, J

    K. Kaheman, J. N. Kutz, and S. L. Brunton, `` en SINDy - PI : a robust algorithm for parallel implicit sparse identification of nonlinear dynamics ,'' en Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , vol. 476, no. 2242, p. 20200279, Oct. 2020. [Online]. Available: https://royalsocietypublishing.org/doi/10.1098/rspa....

  22. [22]

    Kacprzyk, Z

    K. Kacprzyk, Z. Qian, and M. van der Schaar, ``D- CIPHER : Discovery of Closed -form Partial Differential Equations ,'' in Advances in Neural Information Processing Systems , A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, Eds., vol. 36. 1em plus 0.5em minus 0.4em Curran Associates, Inc., 2023, pp. 27\,609--27\,644. [Online]. Availabl...

  23. [23]

    K. Yu, E. Chatzi, and G. Kissas, `` en Grammar-based ordinary differential equation discovery ,'' en Mechanical Systems and Signal Processing , vol. 240, p. 113395, Nov. 2025. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S0888327025010969

  24. [24]

    Q. Li, Y. Hu, J. Liu, and Y. Chen, `` GenSR : Symbolic Regression Based in Equation Generative Space ,'' in Proceedings of the International Conference on Learning Representations ( ICLR ) , 2026, version Number: 1

  25. [25]

    Novikov, N

    A. Novikov, N. Vũ, M. Eisenberger, E. Dupont, P.-S. Huang, A. Z. Wagner, S. Shirobokov, B. Kozlovskii, F. J. R. Ruiz, A. Mehrabian, M. P. Kumar, A. See, S. Chaudhuri, G. Holland, A. Davies, S. Nowozin, P. Kohli, and M. Balog, `` AlphaEvolve : A coding agent for scientific and algorithmic discovery,'' 2025, version Number: 1. [Online]. Available: https://a...

  26. [26]

    Shojaee, K

    P. Shojaee, K. Meidani, S. Gupta, A. B. Farimani, and C. K. Reddy, `` LLM - SR : Scientific Equation Discovery via Programming with Large Language Models ,'' in Proceedings of the International Conference on Learning Representations ( ICLR ) , 2025, version Number: 3. [Online]. Available: https://arxiv.org/abs/2404.18400

  27. [27]

    d'Ascoli, S

    S. d'Ascoli, S. Becker, A. Mathis, P. Schwaller, and N. Kilbertus, `` ODEFormer : Symbolic Regression of Dynamical Systems with Transformers ,'' in The Twelfth International Conference on Learning Representations , ICLR 2024, Vienna , Austria , May 7-11, 2024 . 1em plus 0.5em minus 0.4em OpenReview.net, 2023, version Number: 1. [Online]. Available: https:...

  28. [28]

    Chomsky, Aspects of the theory of syntax

    N. Chomsky, Aspects of the theory of syntax. 1em plus 0.5em minus 0.4em The MIT Press, Mar. 1969. [Online]. Available: https://www.jstor.org/stable/j.ctt17kk81z

  29. [29]

    M. J. Kusner, B. Paige, and J. M. Hernández-Lobato, ``Grammar Variational Autoencoder ,'' in Proceedings of the 34th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research , D. Precup and Y. W. Teh, Eds., vol. 70. 1em plus 0.5em minus 0.4em PMLR, Aug. 2017, pp. 1945--1954. [Online]. Available: https://proceedings.mlr....

  30. [30]

    Mentzer, D

    F. Mentzer, D. Minnen, E. Agustsson, and M. Tschannen, ``Finite Scalar Quantization : VQ - VAE Made Simple ,'' 2023, version Number: 2. [Online]. Available: https://arxiv.org/abs/2309.15505

  31. [31]

    A. v. d. Oord, O. Vinyals, and K. Kavukcuoglu, ``Neural Discrete Representation Learning ,'' 2017, version Number: 2. [Online]. Available: https://arxiv.org/abs/1711.00937

  32. [32]

    Mežnar, S

    S. Mežnar, S. Džeroski, and L. Todorovski, ``Quantifying Behavioural Distance Between Mathematical Expressions ,'' 2024, version Number: 1. [Online]. Available: https://arxiv.org/abs/2408.11515

  33. [33]

    L. V. Kantorovich, `` en Mathematical Methods of Organizing and Planning Production ,'' en Management Science , vol. 6, no. 4, pp. 366--422, Jul. 1960. [Online]. Available: https://pubsonline.informs.org/doi/10.1287/mnsc.6.4.366

  34. [34]

    D. P. Kingma and J. Ba, ``Adam: A Method for Stochastic Optimization ,'' 2014, version Number: 9. [Online]. Available: https://arxiv.org/abs/1412.6980

  35. [35]

    Ansel, E

    J. Ansel, E. Yang, H. He, N. Gimelshein, A. Jain, M. Voznesensky, B. Bao, P. Bell, D. Berard, E. Burovski, G. Chauhan, A. Chourdia, W. Constable, A. Desmaison, Z. DeVito, E. Ellison, W. Feng, J. Gong, M. Gschwind, B. Hirsh, S. Huang, K. Kalambarkar, L. Kirsch, M. Lazos, M. Lezcano, Y. Liang, J. Liang, Y. Lu, C. K. Luk, B. Maher, Y. Pan, C. Puhrsch, M. Res...

  36. [36]

    Nisonoff, J

    H. Nisonoff, J. Xiong, S. Allenspach, and J. Listgarten, ``Unlocking Guidance for Discrete State - Space Diffusion and Flow Models ,'' in International Conference on Learning Representations , Y. Yue, A. Garg, N. Peng, F. Sha, and R. Yu, Eds., vol. 2025, 2025, pp. 36\,052--36\,106. [Online]. Available: https://proceedings.iclr.cc/paper_files/paper/2025/fi...

  37. [37]

    Campbell, J

    A. Campbell, J. Yim, R. Barzilay, T. Rainforth, and T. Jaakkola, ``Generative Flows on Discrete State - Spaces : Enabling Multimodal Flows with Applications to Protein Co - Design ,'' in Proceedings of the 41st International Coference on Machine Learning . ICML . 1em plus 0.5em minus 0.4em Vienna, Austria: PMLR 235, 2024. [Online]. Available: https://arxi...

  38. [38]

    J. A. Nelder and R. Mead, `` en A Simplex Method for Function Minimization ,'' en The Computer Journal , vol. 7, no. 4, pp. 308--313, Jan. 1965. [Online]. Available: https://academic.oup.com/comjnl/article-lookup/doi/10.1093/comjnl/7.4.308

  39. [39]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. Van Der Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones, R. Kern, E. Larson, C. J. Carey, I. Polat, Y. Feng, E. W. Moore, J. VanderPlas, D. Laxalde, J. Perktold, R. Cimrman, I. Henriksen, ...

  40. [40]

    A. M. Lyapunov, `` en The general problem of the stability of motion ,'' en International Journal of Control , vol. 55, no. 3, pp. 531--534, Mar. 1992. [Online]. Available: http://www.tandfonline.com/doi/abs/10.1080/00207179208934253

  41. [41]

    H. K. Khalil and J. W. Grizzle, Nonlinear systems. 1em plus 0.5em minus 0.4em Prentice hall Upper Saddle River, NJ, 2002, vol. 3

  42. [42]

    Clevert, T

    D.-A. Clevert, T. Unterthiner, and S. Hochreiter, ``Fast and Accurate Deep Network Learning by Exponential Linear Units ( ELUs ),'' 2015, version Number: 5. [Online]. Available: https://arxiv.org/abs/1511.07289

This paper was first reviewed by grok-4.5 on July 15, 2026.