pith. machine review for the scientific record. sign in

arxiv: 2604.11929 · v2 · submitted 2026-04-13 · 💻 cs.LG · math.DS· physics.comp-ph

Recognition: unknown

Fast and principled equation discovery from chaos to climate

Yuzheng Zhang , Weizhen Li , Rui Carvalho
Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:07 UTC · model grok-4.3

classification 💻 cs.LG math.DSphysics.comp-ph
keywords equation discoveryBayesian inferencesparse regressionchaotic systemsclimate modelinguncertainty quantificationdata-driven dynamicsSINDy
0
0 comments X

The pith

Bayesian-ARGOS discovers governing equations from noisy observations by screening candidates quickly then applying Bayesian inference for rigor and uncertainty.

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

The paper introduces Bayesian-ARGOS to identify the differential equations that drive complex systems directly from scarce and noisy time-series data. It pairs a fast frequentist step that narrows possible terms with a targeted Bayesian step that quantifies uncertainty and checks validity, avoiding the accuracy-efficiency trade-off in earlier library-based methods. On seven benchmark chaotic systems the approach uses less data than SINDy, tolerates more noise in most cases, and runs two orders of magnitude faster than bootstrap-based alternatives while supplying standard statistical diagnostics. When the same pipeline is coupled with representation learning for high-dimensional sea-surface-temperature fields, it produces more valid latent equations that remain stable over long forecast horizons.

Core claim

Bayesian-ARGOS reconciles automation, statistical rigor, and speed by first using frequentist screening to prune an overcomplete library of candidate terms and then performing focused Bayesian inference on the surviving models, thereby delivering governing equations together with principled uncertainty estimates at far lower computational cost than existing sparse-regression or bootstrap methods.

What carries the argument

The hybrid Bayesian-ARGOS pipeline, which uses frequentist screening to select a reduced candidate set followed by Bayesian posterior inference to rank models and produce uncertainty measures.

If this is right

  • The method scales equation discovery to real climate data by increasing the fraction of valid latent equations and their forecast stability.
  • Standard diagnostics for influence and multicollinearity become routine parts of the workflow, exposing when a discovered equation is unreliable.
  • Computational cost drops by roughly two orders of magnitude relative to bootstrap-based Bayesian alternatives, enabling repeated application on large ensembles.
  • Data efficiency improves across all tested systems, allowing usable equations to be recovered from shorter observation windows.
  • The same screening-plus-Bayesian structure can be inserted into other sparse-regression pipelines without redesigning the core library.

Where Pith is reading between the lines

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

  • The efficiency gain could allow equation discovery to be rerun frequently as new observations arrive, supporting adaptive modeling in operational forecasting.
  • The diagnostics may generalize to detect when representation-learning preprocessing itself introduces spurious terms in latent space.
  • If the screening threshold is tuned per system, the framework might extend to non-chaotic domains such as biological regulatory networks where data are similarly sparse.
  • Combining the approach with symbolic regression variants could further reduce reliance on a pre-specified library.

Load-bearing premise

The frequentist screening stage does not discard terms that would have been retained under a fully Bayesian search, preserving completeness when data are limited or noisy.

What would settle it

On a new high-dimensional chaotic system with 20 percent observation noise, Bayesian-ARGOS yields equations with lower long-horizon prediction accuracy than SINDy while using comparable data volume.

read the original abstract

Our ability to predict, control, and ultimately understand complex systems rests on discovering the equations that govern their dynamics. Identifying these equations directly from noisy, limited observations has therefore become a central challenge in data-driven science, yet existing library-based sparse regression methods force a compromise between automation, statistical rigor, and computational efficiency. Here we develop Bayesian-ARGOS, a hybrid framework that reconciles these demands by combining rapid frequentist screening with focused Bayesian inference, enabling automated equation discovery with principled uncertainty quantification at a fraction of the computational cost of existing methods. Tested on seven chaotic systems under varying data scarcity and noise levels, Bayesian-ARGOS outperforms two state-of-the-art methods in most scenarios. It surpasses SINDy in data efficiency for all systems and noise tolerance for six out of the seven, with a two-order-of-magnitude reduction in computational cost compared to bootstrap-based ARGOS. The probabilistic formulation additionally enables a suite of standard statistical diagnostics, including influence analysis and multicollinearity detection that expose failure modes otherwise opaque. When integrated with representation learning (SINDy-SHRED) for high dimensional sea surface temperature reconstruction, Bayesian-ARGOS increases the yield of valid latent equations with significantly improved long horizon stability. Bayesian-ARGOS thus provides a principled, automated, and computationally efficient route from scarce and noisy observations to interpretable governing equations, offering a practical framework for equation discovery across scales, from benchmark chaotic systems to the latent dynamics underlying global climate patterns.

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 / 1 minor

Summary. The manuscript introduces Bayesian-ARGOS, a hybrid framework combining rapid frequentist screening with focused Bayesian inference for automated discovery of governing equations from noisy, limited data. It claims to outperform SINDy and bootstrap-based ARGOS on seven chaotic systems in data efficiency (all systems), noise tolerance (six of seven), and computational cost (two-order-of-magnitude reduction), while enabling standard statistical diagnostics such as influence analysis and multicollinearity detection. The method is further integrated with representation learning (SINDy-SHRED) for high-dimensional sea surface temperature reconstruction, increasing the yield of valid latent equations and improving long-horizon stability.

Significance. If the central claims hold after addressing validation gaps, the work offers a computationally efficient route to principled equation discovery with uncertainty quantification, bridging benchmark chaotic systems to real climate applications. The reported efficiency gains and diagnostic tools represent practical advances for data-driven science in dynamical systems.

major comments (2)
  1. [Abstract, paragraph 2; method pipeline description] Abstract and method description: The hybrid pipeline applies frequentist screening to prune the library before Bayesian inference, yet no experiments or analysis quantify the screening step's false-negative rate for ground-truth terms across noise levels and data scarcity. If screening excludes true terms (as is common in sparse regression under noise), the Bayesian posterior is conditioned on an incomplete model space, undermining the claim of 'principled uncertainty quantification' and rendering reported noise-tolerance gains potentially attributable to screening artifacts rather than the framework.
  2. [Abstract, paragraph 2; experimental results] Results section on chaotic systems: Superiority is asserted over two state-of-the-art methods on seven systems, but the abstract provides no exact metrics, error bars, or details on handling post-hoc choices (e.g., library construction, thresholding). This absence makes it impossible to evaluate whether the two-order-of-magnitude cost reduction and outperformance are robust or sensitive to implementation details.
minor comments (1)
  1. [Abstract, paragraph 3] The abstract mentions 'standard statistical diagnostics' but does not specify which ones are implemented or how they are computed in the hybrid setting; a brief enumeration would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which highlight important aspects of our hybrid framework that merit further clarification and validation. We address each major comment below and indicate the revisions we will incorporate.

read point-by-point responses

  1. Referee: [Abstract, paragraph 2; method pipeline description] Abstract and method description: The hybrid pipeline applies frequentist screening to prune the library before Bayesian inference, yet no experiments or analysis quantify the screening step's false-negative rate for ground-truth terms across noise levels and data scarcity. If screening excludes true terms (as is common in sparse regression under noise), the Bayesian posterior is conditioned on an incomplete model space, undermining the claim of 'principled uncertainty quantification' and rendering reported noise-tolerance gains potentially attributable to screening artifacts rather than the framework.

    Authors: We agree that a dedicated quantification of the frequentist screening step's false-negative rate is necessary to fully substantiate the hybrid pipeline's reliability and to ensure the Bayesian inference operates on a model space that includes the ground-truth terms. The original manuscript emphasized end-to-end performance but did not isolate screening errors. In the revision we will add a new analysis (in the Methods or an appendix) that reports false-negative rates for known ground-truth terms across the tested noise levels and data-scarcity regimes on the seven chaotic systems. This will demonstrate that the chosen screening threshold maintains high recall, thereby supporting the validity of the reported uncertainty quantification and noise-tolerance improvements. revision: yes

  2. Referee: [Abstract, paragraph 2; experimental results] Results section on chaotic systems: Superiority is asserted over two state-of-the-art methods on seven systems, but the abstract provides no exact metrics, error bars, or details on handling post-hoc choices (e.g., library construction, thresholding). This absence makes it impossible to evaluate whether the two-order-of-magnitude cost reduction and outperformance are robust or sensitive to implementation details.

    Authors: The abstract is deliberately concise and therefore omits the precise numerical values, error bars, and implementation specifics that appear in the Results section, figures, and supplementary material. To improve evaluability directly from the abstract, we will revise it to include the key quantitative claims (e.g., data-efficiency and noise-tolerance gains with associated variability measures) while preserving its brevity. We will also expand the Methods section with an explicit description of library construction, thresholding procedure, and any post-hoc choices, including brief sensitivity checks that confirm robustness of the reported computational-cost reduction. revision: partial

Circularity Check

0 steps flagged

No significant circularity in Bayesian-ARGOS derivation chain

full rationale

The abstract and description present Bayesian-ARGOS as a hybrid pipeline of rapid frequentist screening followed by focused Bayesian inference on the pruned library, with empirical outperformance claims on benchmark systems. No load-bearing step reduces a claimed prediction or result to its own inputs by construction, self-definition, or renaming. Performance metrics (data efficiency, noise tolerance) are reported as external comparisons to SINDy and bootstrap-ARGOS rather than tautological fits. Self-citations, if present for prior ARGOS work, are not invoked as uniqueness theorems or ansatzes that force the central result. The derivation remains self-contained against external benchmarks without the enumerated circular patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the framework appears to rest on standard sparse regression and Bayesian assumptions without new postulates.

pith-pipeline@v0.9.0 · 5564 in / 1057 out tokens · 64041 ms · 2026-05-10T15:07:25.204386+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

73 extracted references · 48 canonical work pages

  1. [1]

    Notices of the Amer- ican Mathematical Society68(04), 1 (2021) https://doi.org/10.1090/noti2259

    E, W.: The Dawning of a New Era in Applied Mathematics. Notices of the Amer- ican Mathematical Society68(04), 1 (2021) https://doi.org/10.1090/noti2259

    work page doi:10.1090/noti2259 2021

  2. [2]

    Proceedings of the National Academy of Sciences113(15), 3932–3937 (2016) https://doi.org/10

    Brunton, S.L., Proctor, J.L., Kutz, J.N.: Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the National Academy of Sciences113(15), 3932–3937 (2016) https://doi.org/10. 1073/pnas.1517384113

    2016

  3. [3]

    Science Advances3(4), 1602614 (2017) https://doi

    Rudy, S.H., Brunton, S.L., Proctor, J.L., Kutz, J.N.: Data-driven discovery of partial differential equations. Science Advances3(4), 1602614 (2017) https://doi. org/10.1126/sciadv.1602614

    work page doi:10.1126/sciadv.1602614 2017

  4. [4]

    Europhysics Letters142(1), 11001 (2023) https://doi.org/10.1209/ 0295-5075/acc3bf

    Gao, T.-T., Yan, G.: Data-driven inference of complex system dynamics: A mini-review. Europhysics Letters142(1), 11001 (2023) https://doi.org/10.1209/ 0295-5075/acc3bf

    2023

  5. [5]

    Nature Computational Science (2024) https://doi.org/10

    Brunton, S.L., Kutz, J.N.: Promising directions of machine learning for partial differential equations. Nature Computational Science (2024) https://doi.org/10. 1038/s43588-024-00643-2

    2024

  6. [6]

    Champion, B

    Champion, K., Lusch, B., Kutz, J.N., Brunton, S.L.: Data-driven discovery of coordinates and governing equations. Proceedings of the National Academy of Sciences116(45), 22445–22451 (2019) https://doi.org/10.1073/pnas.1906995116

    work page doi:10.1073/pnas.1906995116 2019

  7. [7]

    Nature (2025) https: //doi.org/10.1038/s41586-025-09544-4

    Raut, R.V., Rosenthal, Z.P., Wang, X., Miao, H., Zhang, Z., Lee, J.-M., Raichle, M.E., Bauer, A.Q., Brunton, S.L., Brunton, B.W., Kutz, J.N.: Arousal as a universal embedding for spatiotemporal brain dynamics. Nature (2025) https: //doi.org/10.1038/s41586-025-09544-4

    work page doi:10.1038/s41586-025-09544-4 2025

  8. [8]

    Nature Computational Science2(3), 160–168 (2022) https://doi.org/10.1038/s43588-022-00217-0

    Gao, T.-T., Yan, G.: Autonomous inference of complex network dynamics from incomplete and noisy data. Nature Computational Science2(3), 160–168 (2022) https://doi.org/10.1038/s43588-022-00217-0

    work page doi:10.1038/s43588-022-00217-0 2022

  9. [9]

    IEEE Access7, 1404–1423 (2019) https://doi.org/10.1109/ACCESS.2018.2886528

    Zheng, P., Askham, T., Brunton, S.L., Kutz, J.N., Aravkin, A.Y.: A Unified Framework for Sparse Relaxed Regularized Regression: SR3. IEEE Access7, 1404–1423 (2019) https://doi.org/10.1109/ACCESS.2018.2886528

    work page doi:10.1109/access.2018.2886528 2019

  10. [10]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences473(2204), 20170009 (2017) https://doi.org/10.1098/rspa.2017.0009

    Mangan, N.M., Kutz, J.N., Brunton, S.L., Proctor, J.L.: Model selection for dynamical systems via sparse regression and information criteria. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences473(2204), 20170009 (2017) https://doi.org/10.1098/rspa.2017.0009

    work page doi:10.1098/rspa.2017.0009 2017

  11. [11]

    Proceedings of the Royal Society A: Mathematical, Physical and 29 Engineering Sciences478(2260), 20210904 (2022) https://doi.org/10.1098/rspa

    Fasel, U., Kutz, J.N., Brunton, B.W., Brunton, S.L.: Ensemble-SINDy: Robust sparse model discovery in the low-data, high-noise limit, with active learning and control. Proceedings of the Royal Society A: Mathematical, Physical and 29 Engineering Sciences478(2260), 20210904 (2022) https://doi.org/10.1098/rspa. 2021.0904

    work page doi:10.1098/rspa 2022

  12. [13]

    Neural Computation4(3), 415–447 (1992) https://doi.org/10.1162/neco.1992.4.3.415

    MacKay, D.J.C.: Bayesian Interpolation. Neural Computation4(3), 415–447 (1992) https://doi.org/10.1162/neco.1992.4.3.415

    work page doi:10.1162/neco.1992.4.3.415 1992

  13. [14]

    IEEE Transactions on Automatic Control61(1), 182–187 (2016) https://doi.org/10.1109/TAC.2015

    Pan, W., Yuan, Y., Goncalves, J., Stan, G.-B.: A Sparse Bayesian Approach to the Identification of Nonlinear State-Space Systems. IEEE Transactions on Automatic Control61(1), 182–187 (2016) https://doi.org/10.1109/TAC.2015. 2426291

    work page doi:10.1109/tac.2015 2016

  14. [15]

    Proceedings of the Royal Society A: Mathematical, Physical and Engi- neering Sciences474(2217), 20180305 (2018) https://doi.org/10.1098/rspa.2018

    Zhang, S., Lin, G.: Robust data-driven discovery of governing physical laws with error bars. Proceedings of the Royal Society A: Mathematical, Physical and Engi- neering Sciences474(2217), 20180305 (2018) https://doi.org/10.1098/rspa.2018. 0305

    work page doi:10.1098/rspa.2018 2018

  15. [16]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences481(2319), 20240620 (2025) https://doi.org/10.1098/rspa.2024.0620

    Champneys, M.D., Rogers, T.J.: BINDy: Bayesian identification of nonlinear dynamics with reversible-jump Markov-chain Monte Carlo. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences481(2319), 20240620 (2025) https://doi.org/10.1098/rspa.2024.0620

    work page doi:10.1098/rspa.2024.0620 2025

  16. [17]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences477(2254), 20210426 (2021) https://doi.org/10.1098/rspa.2021.0426

    Martina-Perez, S., Simpson, M.J., Baker, R.E.: Bayesian uncertainty quantifi- cation for data-driven equation learning. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences477(2254), 20210426 (2021) https://doi.org/10.1098/rspa.2021.0426

    work page doi:10.1098/rspa.2021.0426 2021

  17. [18]

    In: International Workshop on Artificial Intelligence and Statistics, pp

    Tipping, M.E., Faul, A.C.: Fast Marginal Likelihood Maximisation for Sparse Bayesian Models. In: International Workshop on Artificial Intelligence and Statistics, pp. 276–283. PMLR, Key West, FL, USA (2003)

    2003

  18. [19]

    Nonlinear Dynamics111(14), 13143–13164 (2023) https://doi.org/10.1007/ s11071-023-08525-4

    Kaptanoglu, A.A., Zhang, L., Nicolaou, Z.G., Fasel, U., Brunton, S.L.: Benchmarking sparse system identification with low-dimensional chaos. Nonlinear Dynamics111(14), 13143–13164 (2023) https://doi.org/10.1007/ s11071-023-08525-4

    2023

  19. [20]

    Statistics and Computing , author =

    Vehtari, A., Gelman, A., Gabry, J.: Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Statistics and Computing27(5), 1413– 1432 (2017) https://doi.org/10.1007/s11222-016-9696-4

    work page doi:10.1007/s11222-016-9696-4 2017

  20. [21]

    The American Statistician29(1), 3–20 (1975) https://doi.org/10.2307/2683673 2683673 30

    Marquardt, D.W., Snee, R.D.: Ridge Regression in Practice. The American Statistician29(1), 3–20 (1975) https://doi.org/10.2307/2683673 2683673 30

    work page doi:10.2307/2683673 1975

  21. [22]

    arXiv (2025)

    Gao, M.L., Williams, J.P., Kutz, J.N.: Sparse Identification of Nonlinear Dynam- ics and Koopman Operators with Shallow Recurrent Decoder Networks. arXiv (2025). https://doi.org/10.48550/arXiv.2501.13329

    work page doi:10.48550/arxiv.2501.13329 2025

  22. [23]

    Benner, S

    Benner, P., Gugercin, S., Willcox, K.: A Survey of Projection-Based Model Reduc- tion Methods for Parametric Dynamical Systems. SIAM Review57(4), 483–531 (2015) https://doi.org/10.1137/130932715

    work page doi:10.1137/130932715 2015

  23. [24]

    Applied Mathematical Sciences, vol

    Guckenheimer, J., Holmes, P.: Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Applied Mathematical Sciences, vol. 42. Springer, New York, NY (1983). https://doi.org/10.1007/978-1-4612-1140-2

    work page doi:10.1007/978-1-4612-1140-2 1983

  24. [25]

    Journal of the Ameri- can Statistical Association101(476), 1418–1429 (2006) https://doi.org/10.1198/ 016214506000000735

    Zou, H.: The Adaptive Lasso and its oracle properties. Journal of the Ameri- can Statistical Association101(476), 1418–1429 (2006) https://doi.org/10.1198/ 016214506000000735

    2006

  25. [26]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , author =

    Tibshirani, R.: Regression shrinkage and selection via the Lasso. Journal of the Royal Statistical Society. Series B (Methodological)58, 267–288 (1996) https: //doi.org/10.1111/j.2517-6161.1996.tb02080.x

    work page doi:10.1111/j.2517-6161.1996.tb02080.x 1996

  26. [27]

    The Annals of Statistics7(1), 1–26 (1979) https://doi.org/10.1214/aos/1176344552

    Schwarz, G.: Estimating the dimension of a model. The Annals of Statistics6(2), 461–464 (1978) https://doi.org/10.1214/aos/1176344136

    work page doi:10.1214/aos/1176344136 1978

  27. [28]

    Hoerl and Robert W

    Hoerl, A.E., Kennard, R.W.: Ridge regression: Biased estimation for nonorthog- onal problems. Technometrics12(1), 55–67 (1970) https://doi.org/10.1080/ 00401706.1970.10488634

    work page arXiv 1970

  28. [29]

    Current trends in Bayesian methodology with applications79(30), 2–4 (2015)

    Betancourt, M., Girolami, M.: Hamiltonian monte carlo for hierarchical models. Current trends in Bayesian methodology with applications79(30), 2–4 (2015)

    2015

  29. [30]

    Machine Learning: Science and Technology5(3), 035046 (2024) https://doi.org/10.1088/2632-2153/ad682f

    Li, W., Carvalho, R.: Automating the discovery of partial differential equations in dynamical systems. Machine Learning: Science and Technology5(3), 035046 (2024) https://doi.org/10.1088/2632-2153/ad682f

    work page doi:10.1088/2632-2153/ad682f 2024

  30. [31]

    arXiv (2025)

    Gao, M.L., Kutz, J.N., Font, B.: Mesh-Free Sparse Identification of Nonlinear Dynamics. arXiv (2025). https://doi.org/10.48550/arXiv.2505.16058

    work page doi:10.48550/arxiv.2505.16058 2025

  31. [32]

    Quality & Quantity41(5), 673–690 (2007) https://doi.org/10.1007/ s11135-006-9018-6

    O’brien, R.M.: A Caution Regarding Rules of Thumb for Variance Inflation Factors. Quality & Quantity41(5), 673–690 (2007) https://doi.org/10.1007/ s11135-006-9018-6

    2007

  32. [33]

    Journal of Machine Learning Research25(72), 1–58 (2024)

    Vehtari, A., Simpson, D., Gelman, A., Yao, Y., Gabry, J.: Pareto Smoothed Importance Sampling. Journal of Machine Learning Research25(72), 1–58 (2024)

    2024

  33. [34]

    Physical Review Research3(3), 033270 (2021) https://doi.org/ 10.1103/PhysRevResearch.3.033270 31

    Xu, H., Zhang, D.: Robust discovery of partial differential equations in com- plex situations. Physical Review Research3(3), 033270 (2021) https://doi.org/ 10.1103/PhysRevResearch.3.033270 31

    work page doi:10.1103/physrevresearch.3.033270 2021

  34. [35]

    Communications Physics5(1), 1–7 (2022) https: //doi.org/10.1038/s42005-022-00987-z

    Lu, P.Y., Ari˜ no Bernad, J., Soljaˇ ci´ c, M.: Discovering sparse interpretable dynam- ics from partial observations. Communications Physics5(1), 1–7 (2022) https: //doi.org/10.1038/s42005-022-00987-z

    work page doi:10.1038/s42005-022-00987-z 2022

  35. [36]

    Physical Review Research6(1), 013182 (2024) https://doi.org/10.1103/ PhysRevResearch.6.013182

    Du, M., Chen, Y., Zhang, D.: DISCOVER: Deep identification of symboli- cally concise open-form partial differential equations via enhanced reinforcement learning. Physical Review Research6(1), 013182 (2024) https://doi.org/10.1103/ PhysRevResearch.6.013182

    2024

  36. [37]

    Nature Communications12, 6136 (2021) https://doi.org/10.1038/ s41467-021-26434-1

    Chen, Z., Liu, Y., Sun, H.: Physics-informed learning of governing equations from scarce data. Nature Communications12, 6136 (2021) https://doi.org/10.1038/ s41467-021-26434-1

    2021

  37. [38]

    Nature Communications12, 3219 (2021) https: //doi.org/10.1038/s41467-021-23479-0

    Reinbold, P.A.K., Kageorge, L.M., Schatz, M.F., Grigoriev, R.O.: Robust learn- ing from noisy, incomplete, high-dimensional experimental data via physically constrained symbolic regression. Nature Communications12, 3219 (2021) https: //doi.org/10.1038/s41467-021-23479-0

    work page doi:10.1038/s41467-021-23479-0 2021

  38. [39]

    PLOS Computational Biology21(1), 1012762 (2025) https://doi.org/10.1371/journal.pcbi.1012762

    Wu, X., McDermott, M., MacLean, A.L.: Data-driven model discovery and model selection for noisy biological systems. PLOS Computational Biology21(1), 1012762 (2025) https://doi.org/10.1371/journal.pcbi.1012762

    work page doi:10.1371/journal.pcbi.1012762 2025

  39. [40]

    Nature Communications15(1), 6029 (2024) https://doi.org/10.1038/s41467-024-50378-x

    Gao, T.-T., Barzel, B., Yan, G.: Learning interpretable dynamics of stochastic complex systems from experimental data. Nature Communications15(1), 6029 (2024) https://doi.org/10.1038/s41467-024-50378-x

    work page doi:10.1038/s41467-024-50378-x 2024

  40. [41]

    Nature Communications 16(1), 10260 (2025) https://doi.org/10.1038/s41467-025-65126-y

    Tomasetto, M., Williams, J.P., Braghin, F., Manzoni, A., Kutz, J.N.: Reduced order modeling with shallow recurrent decoder networks. Nature Communications 16(1), 10260 (2025) https://doi.org/10.1038/s41467-025-65126-y

    work page doi:10.1038/s41467-025-65126-y 2025

  41. [42]

    Journal of Climate 15(13), 1609–1625 (2002) https://doi.org/10.1175/1520-0442(2002)015⟨1609: AIISAS⟩2.0.CO;2

    Reynolds, R.W., Rayner, N.A., Smith, T.M., Stokes, D.C., Wang, W.: An Improved In Situ and Satellite SST Analysis for Climate. Journal of Climate 15(13), 1609–1625 (2002) https://doi.org/10.1175/1520-0442(2002)015⟨1609: AIISAS⟩2.0.CO;2 . Chap. Journal of Climate

    work page doi:10.1175/1520-0442(2002)015 2002

  42. [43]

    Monthly Weather Review117(10), 2165–2185 (1989) https://doi.org/ 10.1175/1520-0493(1989)117⟨2165:RFAFUP⟩2.0.CO;2

    Penland, C.: Random Forcing and Forecasting Using Principal Oscillation Pattern Analysis. Monthly Weather Review117(10), 2165–2185 (1989) https://doi.org/ 10.1175/1520-0493(1989)117⟨2165:RFAFUP⟩2.0.CO;2 . Chap. Monthly Weather Review

    work page doi:10.1175/1520-0493(1989)117 1989

  43. [44]

    Physica D: Nonlinear Phenomena98(2), 534–558 (1996) https://doi.org/10.1016/ 0167-2789(96)00124-8

    Penland, C.: A stochastic model of IndoPacific sea surface temperature anomalies. Physica D: Nonlinear Phenomena98(2), 534–558 (1996) https://doi.org/10.1016/ 0167-2789(96)00124-8

    1996

  44. [45]

    Journal of Climate 21(2), 385–402 (2008) https://doi.org/10.1175/2007JCLI1849.1

    Alexander, M.A., Matrosova, L., Penland, C., Scott, J.D., Chang, P.: Forecasting Pacific SSTs: Linear Inverse Model Predictions of the PDO. Journal of Climate 21(2), 385–402 (2008) https://doi.org/10.1175/2007JCLI1849.1 . Chap. Journal 32 of Climate

    work page doi:10.1175/2007jcli1849.1 2008

  45. [46]

    John Wiley & Sons, Hoboken, NJ, USA (2005)

    Belsley, D.A., Kuh, E., Welsch, R.E.: Regression Diagnostics: Identifying Influ- ential Data and Sources of Collinearity. John Wiley & Sons, Hoboken, NJ, USA (2005)

    2005

  46. [47]

    Statistica Sinica6(4), 733–760 (1996) 24306036

    Gelman, A., Meng, X.-L., Stern, H.: Posterior Predictive Assessment of Model Fit- ness Via Realized Discrepancies. Statistica Sinica6(4), 733–760 (1996) 24306036

    1996

  47. [48]

    Springer, Dordrecht (1982)

    Cook, R.D., Weisberg, S.: Residuals and Influence in Regression. Springer, Dordrecht (1982)

    1982

  48. [49]

    The Journal of Chemical Physics148(24), 241723 (2018) https://doi

    Boninsegna, L., N¨ uske, F., Clementi, C.: Sparse learning of stochastic dynamical equations. The Journal of Chemical Physics148(24), 241723 (2018) https://doi. org/10.1063/1.5018409

    work page doi:10.1063/1.5018409 2018

  49. [50]

    Representation learning: a review and new perspectives

    Bengio, Y., Courville, A., Vincent, P.: Representation Learning: A Review and New Perspectives. IEEE Transactions on Pattern Analysis and Machine Intelligence35(8), 1798–1828 (2013) https://doi.org/10.1109/TPAMI.2013.50

    work page doi:10.1109/tpami.2013.50 2013

  50. [51]

    Physical Review E100(2), 022220 (2019) https://doi.org/10

    Li, S., Kaiser, E., Laima, S., Li, H., Brunton, S.L., Kutz, J.N.: Discovering time- varying aerodynamics of a prototype bridge by sparse identification of nonlinear dynamical systems. Physical Review E100(2), 022220 (2019) https://doi.org/10. 1103/PhysRevE.100.022220

    2019

  51. [52]

    Nature Communications16(1) (2025) https://doi

    Hu, J., Cui, J., Yang, B.: Learning interpretable network dynamics via universal neural symbolic regression. Nature Communications16(1) (2025) https://doi. org/10.1038/s41467-025-61575-7

    work page doi:10.1038/s41467-025-61575-7 2025

  52. [53]

    Proceedings of the National Academy of Sciences , author =

    Taylor, J., Tibshirani, R.J.: Statistical learning and selective inference. Proceed- ings of the National Academy of Sciences112(25), 7629–7634 (2015) https: //doi.org/10.1073/pnas.1507583112

    work page doi:10.1073/pnas.1507583112 2015

  53. [54]

    NAR Genomics and Bioinformatics7(2), 048 (2025) https://doi.org/10.1093/nargab/lqaf048

    Sadria, M., Swaroop, V.: Discovering governing equations of biological systems through representation learning and sparse model discovery. NAR Genomics and Bioinformatics7(2), 048 (2025) https://doi.org/10.1093/nargab/lqaf048

    work page doi:10.1093/nargab/lqaf048 2025

  54. [55]

    The Annals of Applied Statistics2(4), 1360–1383 (2008) 30245139

    Gelman, A., Jakulin, A., Pittau, M.G., Su, Y.-S.: A Weakly Informative Default Prior Distribution for Logistic and Other Regression Models. The Annals of Applied Statistics2(4), 1360–1383 (2008) 30245139

    2008

  55. [56]

    Learning- Based Link Anomaly Detection in Continuous-Time Dynamic Graphs,

    Yang, J., Rao, W., Dehmamy, N., Walters, R., Yu, R.: Symmetry-Informed Gov- erning Equation Discovery. arXiv (2024). https://doi.org/10.48550/arXiv.2405. 16756

    work page doi:10.48550/arxiv.2405 2024

  56. [57]

    Journal of Statistical Software33(1), 1–22 33 (2010) https://doi.org/10.18637/jss.v033.i01

    Friedman, J., Hastie, T., Tibshirani, R.: Regularization paths for generalized linear models via Coordinate Descent. Journal of Statistical Software33(1), 1–22 33 (2010) https://doi.org/10.18637/jss.v033.i01

    work page doi:10.18637/jss.v033.i01 2010

  57. [58]

    Bernoulli19(2), 521–547 (2013) https://doi.org/10

    Belloni, A., Chernozhukov, V.: Least squares after model selection in high- dimensional sparse models. Bernoulli19(2), 521–547 (2013) https://doi.org/10. 3150/11-BEJ410

    2013

  58. [59]

    Hastie, T., Tibshirani, R., Friedman, J.: Elements of statistical learning: Data mining, inference, and prediction (2009)

    2009

  59. [60]

    The Annals of Statistics41(3), 1111–1141 (2013) https://doi.org/10.1214/ 13-AOS1096

    Bien, J., Taylor, J., Tibshirani, R.: A lasso for hierarchical interactions. The Annals of Statistics41(3), 1111–1141 (2013) https://doi.org/10.1214/ 13-AOS1096

    2013

  60. [61]

    Journal of Statistical Software76, 1–32 (2017) https://doi.org/10

    Carpenter, B., Gelman, A., Hoffman, M.D., Lee, D., Goodrich, B., Betancourt, M., Brubaker, M., Guo, J., Li, P., Riddell, A.: Stan: A Probabilistic Programming Language. Journal of Statistical Software76, 1–32 (2017) https://doi.org/10. 18637/jss.v076.i01

    2017

  61. [62]

    Statistical Science7(4), 457–472 (1992) https://doi.org/10.1214/ss/ 1177011136

    Gelman, A., Rubin, D.B.: Inference from Iterative Simulation Using Multiple Sequences. Statistical Science7(4), 457–472 (1992) https://doi.org/10.1214/ss/ 1177011136

    work page doi:10.1214/ss/ 1992

  62. [63]

    Advances in Methods and Practices in Psychological Science1(2), 270–280 (2018) https://doi.org/10.1177/2515245918771304

    Kruschke, J.K.: Rejecting or Accepting Parameter Values in Bayesian Estimation. Advances in Methods and Practices in Psychological Science1(2), 270–280 (2018) https://doi.org/10.1177/2515245918771304

    work page doi:10.1177/2515245918771304 2018

  63. [64]

    1980 , issn =

    Dormand, J.R., Prince, P.J.: A family of embedded Runge-Kutta formulae. Journal of Computational and Applied Mathematics6(1), 19–26 (1980) https: //doi.org/10.1016/0771-050X(80)90013-3

    work page doi:10.1016/0771-050x(80)90013-3 1980

  64. [65]

    Chapman and Hall/CRC, New York (1994)

    Efron, B., Tibshirani, R.J.: An Introduction to the Bootstrap. Chapman and Hall/CRC, New York (1994). https://doi.org/10.1201/9780429246593 34 Supplementary Information S1 Success-rate benchmarking for governing-equation identification on Sprott and Halvorsen systems 0.0 0.8 1.6 x1 0.8 0.0 0.8 x2 0.6 0.0 0.6 x3 0 0.2 0.4 0.6 0.8 1 102 102.5 103 103.5 104 ...

    work page doi:10.1201/9780429246593 1994

  65. [66]

    and mixed products, probing identification performance when the true dynamics are not limited to low-degree polynomials. To generate diverse trajectories, we performed 100 trials with initial conditions sampled independently from uniform distributions:x 1(0)∼ U[−2,2],x 2(0)∼ U[−2,2], andx 3(0)∼ U[−1,2]. Figure S6 summarizes the frequency with which each c...

    2000

  66. [67]

    Proceedings of the National Academy of Sciences113(15), 3932–3937 (2016) https://doi.org/10.1073/ pnas.1517384113

    Brunton, S.L., Proctor, J.L., Kutz, J.N.: Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the National Academy of Sciences113(15), 3932–3937 (2016) https://doi.org/10.1073/ pnas.1517384113

    2016

  67. [68]

    Communications Physics7(1), 1–10 (2024) https://doi.org/10.1038/s42005-023-01516-2

    Egan, K., Li, W., Carvalho, R.: Automatically discovering ordinary differential equations from data with sparse regression. Communications Physics7(1), 1–10 (2024) https://doi.org/10.1038/s42005-023-01516-2

    work page doi:10.1038/s42005-023-01516-2 2024

  68. [69]

    International Journal of Bifurcation and Chaos17(06), 2097–2108 (2007) https://doi.org/10.1142/ S0218127407018245

    Sprott, J.C., Chlouverakis, K.E.: Labyrinth chaos. International Journal of Bifurcation and Chaos17(06), 2097–2108 (2007) https://doi.org/10.1142/ S0218127407018245

    2097

  69. [70]

    Multiscale Modeling & Simulation15(3), 1108–1129 (2017) https://doi.org/ 10.1137/16M1086637

    Tran, G., Ward, R.: Exact Recovery of Chaotic Systems from Highly Corrupted Data. Multiscale Modeling & Simulation15(3), 1108–1129 (2017) https://doi.org/ 10.1137/16M1086637

    work page doi:10.1137/16m1086637 2017

  70. [71]

    Physics Letters A373(40), 3637– 3642 (2009) https://doi.org/10.1016/j.physleta.2009.07.088

    Dadras, S., Momeni, H.R.: A novel three-dimensional autonomous chaotic system generating two, three and four-scroll attractors. Physics Letters A373(40), 3637– 3642 (2009) https://doi.org/10.1016/j.physleta.2009.07.088

    work page doi:10.1016/j.physleta.2009.07.088 2009

  71. [72]

    In: International Conference on Learning Representations (ICLR) (2024)

    d’Ascoli, S., Becker, S., Mathis, A., Schwaller, P., Kilbertus, N.: ODEFormer: Symbolic regression of dynamical systems with transformers. In: International Conference on Learning Representations (ICLR) (2024)

    2024

  72. [73]

    Kuznetsova, D.V

    Sprott, J.C.: A dynamical system with a strange attractor and invariant tori. Physics Letters A378(20), 1361–1363 (2014) https://doi.org/10.1016/j.physleta. 2014.03.028

    work page doi:10.1016/j.physleta 2014

  73. [74]

    Functions of Bounded Variation and Free Discontinuity Problems

    Sprott, J.C.: Chaos and Time-Series Analysis. Oxford University Press, Oxford (2003). https://doi.org/10.1093/oso/9780198508397.001.0001 53

    work page doi:10.1093/oso/9780198508397.001.0001 2003