pith. sign in

arxiv: 2606.26727 · v1 · pith:NCRHGSHWnew · submitted 2026-06-25 · 🌊 nlin.CD

One-shot prediction of noise-induced bifurcations with reservoir computing

Nozomi Akashi , Takayuki Watanabe , Masato Hara , Takao Namiki , Hiroshi Kokubu , Ichiro Tsuda , Kohei Nakajima This is my paper

Pith reviewed 2026-06-26 01:55 UTC · model grok-4.3

classification 🌊 nlin.CD
keywords reservoir computingnoise-induced bifurcationdynamical systemsbifurcation diagramnoise cancellationchaosorder
0
0 comments X

The pith

Reservoir computing trained on one noise condition reconstructs the full noise-induced bifurcation structure.

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

A reservoir computing model can be trained using time series data collected from a dynamical system at only one level of noise. The trained model then predicts how the system's qualitative behavior changes across a range of noise intensities, capturing transitions such as noise-induced chaos or order. This approach uses dynamic noise cancellation to separate the noise effects from the underlying deterministic dynamics. The result is a reconstruction of the entire bifurcation diagram without needing data from other noise conditions. This matters for analyzing noisy real-world systems where varying the noise level may be difficult or costly.

Core claim

A simple reservoir computing framework can predict the noise-induced bifurcation structure from the time series at a single noise condition by demonstrating dynamic noise cancellation and the reconstruction of entire noise-induced bifurcation structures, including noise-induced chaos and noise-induced order, in representative dynamical systems.

What carries the argument

Dynamic noise cancellation performed by the reservoir computing network, allowing generalization across noise intensities.

If this is right

  • The full bifurcation structure can be obtained from data at one noise intensity.
  • Noise-induced chaos and order can be reconstructed without additional sampling.
  • The method extends to neuromorphic spintronics devices.
  • A theoretical explanation for the noise cancellation is provided.

Where Pith is reading between the lines

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

  • This suggests the reservoir learns an effective noise-free representation of the system.
  • Similar one-shot methods might apply to other parameter variations beyond noise.
  • Experimental validation on physical systems could confirm applicability to real devices.

Load-bearing premise

The reservoir trained at one noise intensity generalizes to predict the bifurcation structure at all other intensities through dynamic noise cancellation.

What would settle it

Run the dynamical system at a second noise intensity, collect the true bifurcation points, and check if the reservoir's predictions match those points.

read the original abstract

Dynamical systems can exhibit complex responses when noise is injected. In particular, dynamics can be qualitatively altered by dynamic noise, a phenomenon known as noise-induced bifurcation. Predicting noise-induced bifurcations is a critical challenge in nonlinear physics. Recently, it has been reported that reservoir computing, a machine learning framework, can reconstruct the unseen global structure of a dynamical system, including bifurcations, from limited time series data. However, learning global structures in random dynamical systems has not yet been systematically addressed. In this study, we report that a simple reservoir computing framework can predict the noise-induced bifurcation structure from the time series at a single noise condition. We demonstrate dynamic noise cancellation and the reconstruction of entire noise-induced bifurcation structures, including noise-induced chaos and noise-induced order, in representative dynamical systems. Additionally, we provide a theoretical explanation for noise cancellation and demonstrate noise cancellation of a neuromorphic spintronics device. Our results provide significant insights into understanding and harnessing real-world noisy complex dynamics.

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 paper claims that a reservoir computing model trained solely on time series data generated at one fixed noise intensity can reconstruct the full noise-induced bifurcation diagram—including transitions to noise-induced chaos and noise-induced order—for representative dynamical systems. It asserts that this occurs via an internal dynamic noise cancellation mechanism, supplies a theoretical account of the cancellation, and demonstrates the approach on a neuromorphic spintronics device.

Significance. If the one-shot generalization holds, the result would provide a practical route to forecasting noise-induced qualitative changes from limited data, with direct relevance to experimental nonlinear dynamics and hardware implementations. The combination of an empirical demonstration across multiple systems, a theoretical explanation, and a hardware example constitutes a concrete advance over purely data-driven reconstruction methods.

major comments (2)
  1. [§3.2 and §4.1] §3.2 and §4.1: The central one-shot claim requires that the reservoir dynamics trained at a single σ_train separate the deterministic skeleton from additive noise in a manner that remains valid when σ_test qualitatively alters the attractor structure. The reported reconstructions are shown only for a fixed choice of σ_train; no systematic ablation across multiple training intensities is presented to confirm that the cancellation is independent of the particular σ_train selected.
  2. [§4.3, Figure 7] §4.3, Figure 7: The reconstruction of noise-induced chaos and order transitions is demonstrated for the chosen test systems, yet the paper does not report quantitative error bars or cross-validation metrics that would establish whether the predicted bifurcation points remain accurate when the reservoir is retrained at a different base noise level.
minor comments (1)
  1. [§2] Notation for the reservoir update rule and the noise intensity parameter is introduced without an explicit comparison table to prior reservoir-computing literature on noisy systems; adding such a table would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and indicate the revisions we will incorporate.

read point-by-point responses

  1. Referee: [§3.2 and §4.1] §3.2 and §4.1: The central one-shot claim requires that the reservoir dynamics trained at a single σ_train separate the deterministic skeleton from additive noise in a manner that remains valid when σ_test qualitatively alters the attractor structure. The reported reconstructions are shown only for a fixed choice of σ_train; no systematic ablation across multiple training intensities is presented to confirm that the cancellation is independent of the particular σ_train selected.

    Authors: We agree that systematic variation of σ_train would strengthen the claim of independence. In the revised manuscript we will add an ablation study in §3.2 and §4.1 that retrains the reservoir at several distinct training intensities (both lower and higher than the original choice) and verifies that the reconstructed bifurcation diagrams remain qualitatively and quantitatively consistent. This will directly test whether the internal noise-cancellation mechanism is robust to the specific σ_train. revision: yes

  2. Referee: [§4.3, Figure 7] §4.3, Figure 7: The reconstruction of noise-induced chaos and order transitions is demonstrated for the chosen test systems, yet the paper does not report quantitative error bars or cross-validation metrics that would establish whether the predicted bifurcation points remain accurate when the reservoir is retrained at a different base noise level.

    Authors: We accept that the absence of error bars and cross-validation statistics limits the quantitative assessment of robustness. We will augment §4.3 and Figure 7 with (i) error bars obtained from an ensemble of independent reservoir realizations and (ii) a cross-validation table that reports the variation in predicted bifurcation locations when the model is retrained at alternative base noise intensities. These additions will quantify the stability of the identified transition points. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claim rests on empirical demonstration

full rationale

The abstract and available text present the one-shot prediction as an empirical result of reservoir training on single-noise time series, followed by demonstrated generalization via dynamic noise cancellation across representative systems. No equations, fitted parameters, or derivations are exhibited that reduce the bifurcation reconstruction to the training input by construction. Prior RC work is cited for context but is not invoked as a load-bearing uniqueness theorem or ansatz that forces the present result. The generalization step is asserted via demonstration rather than self-definition, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the central claim rests on the unstated modeling assumption that reservoir dynamics can separate noise effects from the underlying deterministic skeleton.

pith-pipeline@v0.9.1-grok · 5719 in / 1028 out tokens · 15110 ms · 2026-06-26T01:55:38.149286+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

50 extracted references · 11 canonical work pages

  1. [1]

    Crutchfield, J.D

    J.P. Crutchfield, J.D. Farmer, B.A. Huberman, Fluctuations and simple chaotic dynamics. Physics Reports92(2), 45–82 (1982)

    1982

  2. [2]

    Matsumoto, I

    K. Matsumoto, I. Tsuda, Noise-induced order. Journal of Statistical Physics 31(1), 87–106 (1983). https://doi.org/10.1007/BF01010923

    work page doi:10.1007/bf01010923 1983

  3. [3]

    Toral, C.R

    R. Toral, C.R. Mirasso, E. Hern´ andez-Garcıa, O. Piro, Analytical and numer- ical studies of noise-induced synchronization of chaotic systems. Chaos: An Interdisciplinary Journal of Nonlinear Science11(3), 665–673 (2001)

    2001

  4. [4]

    Horsthemke, R

    W. Horsthemke, R. Lefever,Noise-induced transitions: theory and applications in physics, chemistry, and biology(Springer, 1984)

    1984

  5. [5]

    Gammaitoni, P

    L. Gammaitoni, P. H¨ anggi, P. Jung, F. Marchesoni, Stochastic resonance. Reviews of modern physics70(1), 223 (1998)

    1998

  6. [6]

    Akashi, T

    N. Akashi, T. Yamaguchi, S. Tsunegi, T. Taniguchi, M. Nishida, R. Sakurai, Y. Wakao, K. Nakajima, Input-driven bifurcations and information processing capacity in spintronics reservoirs. Phys. Rev. Res.2, 043303 (2020). https://doi. org/10.1103/PhysRevResearch.2.043303

    work page doi:10.1103/physrevresearch.2.043303 2020

  7. [7]

    echo state

    H. Jaeger, The“echo state”approach to analysing and training recurrent neural networks-with an erratum note. Bonn, Germany: German National Research Center for Information Technology GMD Technical Report148(34), 13 (2001)

    2001

  8. [8]

    Maass, T

    W. Maass, T. Natschl¨ ager, H. Markram, Real-time computing without stable states: A new framework for neural computation based on perturbations. Neural computation14(11), 2531–2560 (2002)

    2002

  9. [9]

    Nakajima, I

    K. Nakajima, I. Fischer,Reservoir Computing—Theory, Physical Implementa- tions, and Applications(Springer, Singapore, 2021) 18

    2021

  10. [10]

    Pathak, B

    J. Pathak, B. Hunt, M. Girvan, Z. Lu, E. Ott, Model-free prediction of large spatiotemporally chaotic systems from data: A reservoir computing approach. Physical review letters120(2), 024102 (2018)

    2018

  11. [11]

    Pathak, A

    J. Pathak, A. Wikner, R. Fussell, S. Chandra, B.R. Hunt, M. Girvan, E. Ott, Hybrid forecasting of chaotic processes: Using machine learning in conjunction with a knowledge-based model. Chaos: An interdisciplinary journal of nonlinear science28(4) (2018)

    2018

  12. [12]

    Vlachas, J

    P.R. Vlachas, J. Pathak, B.R. Hunt, T.P. Sapsis, M. Girvan, E. Ott, P. Koumout- sakos, Backpropagation algorithms and reservoir computing in recurrent neural networks for the forecasting of complex spatiotemporal dynamics. Neural Networks126, 191–217 (2020)

    2020

  13. [13]

    and Girvan, Michelle and Ott, Edward , title =

    J. Pathak, Z. Lu, B.R. Hunt, M. Girvan, E. Ott, Using machine learning to replicate chaotic attractors and calculate Lyapunov exponents from data. Chaos: An Interdisciplinary Journal of Nonlinear Science27(12), 121102 (2017). https: //doi.org/10.1063/1.5010300

    work page doi:10.1063/1.5010300 2017

  14. [14]

    Z. Lu, B.R. Hunt, E. Ott, Attractor reconstruction by machine learning. Chaos: An Interdisciplinary Journal of Nonlinear Science28(6), 061104 (2018)

    2018

  15. [15]

    M. Hara, H. Kokubu, Learning dynamics by reservoir computing. Journal of Dynamics and Differential Equations36(Suppl 1), 515–540 (2024)

    2024

  16. [16]

    Y. Itoh, Y. Tada, M. Adachi, Reconstructing bifurcation diagrams with lyapunov exponents from only time-series data using an extreme learning machine. Non- linear Theory and Its Applications, IEICE8(1), 2–14 (2017). https://doi.org/10. 1587/nolta.8.2

    2017

  17. [17]

    Y. Itoh, S. Uenohara, M. Adachi, T. Morie, K. Aihara, Reconstructing bifurcation diagrams only from time-series data generated by electronic circuits in discrete- time dynamical systems. Chaos: An Interdisciplinary Journal of Nonlinear Science 30(1), 013128 (2020). https://doi.org/10.1063/1.5119187

    work page doi:10.1063/1.5119187 2020

  18. [18]

    J.Z. Kim, Z. Lu, E. Nozari, G.J. Pappas, D.S. Bassett, Teaching recurrent neural networks to infer global temporal structure from local examples. Nature Machine Intelligence3(4), 316–323 (2021)

    2021

  19. [19]

    L.W. Kong, Y. Weng, B. Glaz, M. Haile, Y.C. Lai, Reservoir computing as digital twins for nonlinear dynamical systems. Chaos: An Interdisciplinary Journal of Nonlinear Science33(3) (2023)

    2023

  20. [20]

    Tokuda, Y

    K. Tokuda, Y. Katori, Prediction of unobserved bifurcation by unsupervised extraction of slowly time-varying system parameter dynamics from time series using reservoir computing. Frontiers in Artificial Intelligence7(2024). https: //doi.org/10.3389/frai.2024.1451926 19

    work page doi:10.3389/frai.2024.1451926 2024

  21. [21]

    Tadokoro, A

    S. Tadokoro, A. Yamaguchi, T. Namiki, I. Tsuda, Trans-bifurcation prediction of dynamics in terms of extreme learning machines with control inputs. arXiv preprint arXiv:2410.13289 (2024)

    arXiv 2024

  22. [22]

    J. Shen, R. Al Ajroudi, N. Akashi, T. Jo, M. Nishida, Y. Wakao, R. Sakurai, Y. Kuniyoshi, K. Nakajima, Predicting bifurcation of mechanical systems using reservoir computing: Case studies on legged locomotion and pneumatic soft actu- ator. Chaos: An Interdisciplinary Journal of Nonlinear Science36(1), 013103 (2026). https://doi.org/10.1063/5.0283456

    work page doi:10.1063/5.0283456 2026

  23. [23]

    Semenova, X

    N. Semenova, X. Porte, L. Andreoli, M. Jacquot, L. Larger, D. Brunner, Fun- damental aspects of noise in analog-hardware neural networks. Chaos: An Interdisciplinary Journal of Nonlinear Science29(10), 103128 (2019). https: //doi.org/10.1063/1.5120824

    work page doi:10.1063/1.5120824 2019

  24. [24]

    Est´ ebanez, I

    I. Est´ ebanez, I. Fischer, M.C. Soriano, Constructive role of noise for high- quality replication of chaotic attractor dynamics using a hardware-based reservoir computer. Phys. Rev. Appl.12, 034058 (2019). https://doi.org/10.1103/ PhysRevApplied.12.034058

    2019

  25. [25]

    Wikner, J

    A. Wikner, J. Harvey, M. Girvan, B.R. Hunt, A. Pomerance, T. Antonsen, E. Ott, Stabilizing machine learning prediction of dynamics: Novel noise-inspired regularization tested with reservoir computing. Neural Networks170, 94–110 (2024)

    2024

  26. [26]

    H. Luo, Y. Du, H. Fan, X. Wang, J. Guo, X. Wang, Reconstructing bifurcation diagrams of chaotic circuits with reservoir computing. Phys. Rev. E109, 024210 (2024). https://doi.org/10.1103/PhysRevE.109.024210

    work page doi:10.1103/physreve.109.024210 2024

  27. [27]

    Nathe, C

    C. Nathe, C. Pappu, N.A. Mecholsky, J. Hart, T. Carroll, F. Sorrentino, Reservoir computing with noise. Chaos: An Interdisciplinary Journal of Nonlinear Science 33(4) (2023)

    2023

  28. [28]

    Sedehi, M

    O. Sedehi, M. Yadav, M. Stender, S. Oberst, Denoising and reconstruction of nonlinear dynamics using truncated reservoir computing. Chaos: An Interdisci- plinary Journal of Nonlinear Science35(9), 093103 (2025). https://doi.org/10. 1063/5.0273505

    2025

  29. [29]

    J. Choi, P. Kim, Signal–noise separation using unsupervised reservoir computing. Chaos: An Interdisciplinary Journal of Nonlinear Science35(8) (2025)

    2025

  30. [30]

    Z. Lin, Z. Lu, Z. Di, Y. Tang, Learning noise-induced transitions by multi-scaling reservoir computing. Nature Communications15(1), 6584 (2024). https://doi. org/10.1038/s41467-024-50905-w

    work page doi:10.1038/s41467-024-50905-w 2024

  31. [31]

    Brunton, J.L

    S.L. Brunton, J.L. Proctor, J.N. Kutz, Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the 20 National Academy of Sciences113(15), 3932–3937 (2016). https://doi.org/10. 1073/pnas.1517384113

    2016

  32. [32]

    Verstraeten, B

    D. Verstraeten, B. Schrauwen, M. d’Haene, D. Stroobandt, An experimental unification of reservoir computing methods. Neural networks20(3), 391–403 (2007)

    2007

  33. [33]

    Jaeger, H

    H. Jaeger, H. Haas, Harnessing nonlinearity: Predicting chaotic systems and saving energy in wireless communication. Science304(5667), 78–80 (2004). https://doi.org/10.1126/science.1091277

    work page doi:10.1126/science.1091277 2004

  34. [34]

    K. Sano, T. Mitsui, T. Akimoto, Reduction of the synchronization time in random logistic maps. Physical Review E102(6), 062209 (2020)

    2020

  35. [35]

    Grigoryeva, J.P

    L. Grigoryeva, J.P. Ortega, Echo state networks are universal. Neural Networks 108, 495–508 (2018)

    2018

  36. [36]

    Gonon, J.P

    L. Gonon, J.P. Ortega, Fading memory echo state networks are universal. Neural Networks138, 10–13 (2021)

    2021

  37. [37]

    Sugiura, R

    S. Sugiura, R. Ariizumi, T. Asai, S.I. Azuma, Nonessentiality of reservoir’s fading memory for universality of reservoir computing. IEEE Transactions on Neural Networks and Learning Systems (2023)

    2023

  38. [38]

    Yasumoto, T

    H. Yasumoto, T. Tanaka, Universality of reservoir systems with recurrent neural networks. Neural Networks188, 107413 (2025)

    2025

  39. [39]

    Bishop, Training with noise is equivalent to Tikhonov regularization

    C.M. Bishop, Training with noise is equivalent to Tikhonov regularization. Neural computation7(1), 108–116 (1995)

    1995

  40. [40]

    Grollier, D

    J. Grollier, D. Querlioz, K. Camsari, K. Everschor-Sitte, S. Fukami, M.D. Stiles, Neuromorphic spintronics. Nature electronics3(7), 360–370 (2020)

    2020

  41. [41]

    Akashi, Y

    N. Akashi, Y. Kuniyoshi, S. Tsunegi, T. Taniguchi, M. Nishida, R. Sakurai, Y. Wakao, K. Kawashima, K. Nakajima, A coupled spintronics neuromor- phic approach for high-performance reservoir computing. Advanced Intelligent Systems4(10), 2200123 (2022)

    2022

  42. [42]

    Galatolo, C.L

    S. Galatolo, C.L. Vereau, L. Marangio, I. Nisoli, Efficient computation of station- ary measures and the Lyapunov landscape for families random dynamical systems with smooth additive noise. arXiv preprint arXiv:2508.03895 (2025)

    arXiv 2025

  43. [43]

    Barrientos, F

    P.G. Barrientos, F. Nakamura, Y. Nakano, H. Toyokawa, Finitude of physical measures for random maps. Ast´ erisque459(2025)

    2025

  44. [44]

    Tomita, I

    K. Tomita, I. Tsuda, Towards the interpretation of Hudson’s experiment on the Belousov-Zhabotinsky reaction: Chaos due to delocalization. Progress of Theoretical Physics64(4), 1138–1160 (1980) 21

    1980

  45. [45]

    Hudson, M

    J. Hudson, M. Hart, D. Marinko, An experimental study of multiple peak periodic and nonperiodic oscillations in the Belousov–Zhabotinskii reaction. The Journal of Chemical Physics71(4), 1601–1606 (1979)

    1979

  46. [46]

    Galatolo, M

    S. Galatolo, M. Monge, I. Nisoli, Existence of noise induced order, a computer aided proof. Nonlinearity33(9), 4237 (2020)

    2020

  47. [47]

    L. Shi, H. Wang, S. Wang, R. Du, S.X. Qu, Predicting nonsmooth chaotic dynamics by reservoir computing. Phys. Rev. E109, 014214 (2024). https: //doi.org/10.1103/PhysRevE.109.014214

    work page doi:10.1103/physreve.109.014214 2024

  48. [48]

    Shimada, T

    I. Shimada, T. Nagashima, A numerical approach to ergodic problem of dis- sipative dynamical systems. Progress of theoretical physics61(6), 1605–1616 (1979)

    1979

  49. [49]

    Kubota, K

    H. Kubota, K. Yakushiji, A. Fukushima, S. Tamaru, M. Konoto, T. Nozaki, S. Ishibashi, T. Saruya, S. Yuasa, T. Taniguchi, et al., Spin-torque oscillator based on magnetic tunnel junction with a perpendicularly magnetized free layer and in-plane magnetized polarizer. Applied Physics Express6(10), 103003 (2013)

    2013

  50. [50]

    Taniguchi, T

    T. Taniguchi, T. Ito, S. Tsunegi, H. Kubota, Y. Utsumi, Relaxation time and critical slowing down of a spin-torque oscillator. Physical Review B96(2), 024406 (2017) 22

    2017