pith. sign in

arxiv: 2605.25826 · v1 · pith:ZGGHUGFLnew · submitted 2026-05-25 · 🧮 math.NA · cs.CE· cs.LG· cs.NA

Branched Signature Kernel Solvers for ODEs with rough Single-Trajectory signals

Munawar Ali , Qi Feng , Charlie Pyle , George Xu This is my paper

Pith reviewed 2026-06-29 20:27 UTC · model grok-4.3

classification 🧮 math.NA cs.CEcs.LGcs.NA
keywords branched signature kernelsingle-trajectory ODErough signalskernel collocationuniversal approximationcount-samplingHairer-Kelly morphismstreaming solver
0
0 comments X

The pith

Count-sampling turns a single rough trajectory into nested paths that let branched signature kernels solve ODEs.

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

The paper develops a solver for linear and nonlinear ordinary differential equations driven by a rough forcing signal observed only once. It introduces a count-sampling construction that converts the single observation into a hierarchical family of N+1 nested training paths, allowing branched signature kernels to operate in a setting originally designed for multiple realizations. A kernel-collocation framework places the ansatz on the highest derivative or the integrated solution, and a universal approximation theorem is proved by expressing branched signatures through geometric signatures of time-extended paths via the Hairer-Kelly morphism. The approach further supports a streaming Test/Train/Retrain protocol with closed-form or Newton-step updates. Readers in applied fields would care because the method respects the governing ODE without requiring an ensemble of trajectories.

Core claim

The branched signature kernel solver for single-trajectory ODEs rests on the count-sampling construction that produces N+1 nested training paths from one observation, a collocation scheme that recovers lower derivatives by integration when needed, and a universal approximation result that reduces branched signature evaluations to geometric signatures of time-extended paths through the Hairer-Kelly morphism.

What carries the argument

the count-sampling construction, which converts one observed trajectory into a hierarchical family of N+1 nested training paths on which the branched signature kernel is evaluated

If this is right

  • The solver applies to both linear and nonlinear ODEs with rough forcing.
  • Numerical accuracy and stability are obtained on benchmarks including earthquake displacement, the Solow model, fractional Brownian motion driven equations, Duffing oscillators, and Kuramoto systems.
  • The method extends from offline to streaming operation with closed-form updates in the linear case and scalar Newton steps in the nonlinear case.
  • Kernel collocation can be performed either on the highest-order derivative with integration recovery or directly on the solution after m-fold integration of the ODE.

Where Pith is reading between the lines

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

  • The same count-sampling idea might adapt to other kernel methods that currently require multiple realizations, such as in stochastic simulation.
  • Testing the streaming protocol on real-time data streams with changing roughness could reveal practical limits not addressed in the offline benchmarks.
  • The reduction via the Hairer-Kelly morphism suggests possible extensions to higher-order branched structures in rough path theory.
  • Applications in structural health monitoring could use the online retraining step to update predictions as new forcing segments arrive.

Load-bearing premise

A single observed trajectory contains enough structure to be converted into a hierarchical family of nested paths that the branched signature kernel can use to approximate the ODE solution.

What would settle it

If the approximation error on the El-Centro earthquake displacement benchmark fails to decrease as the number of nested paths N is increased while keeping the kernel fixed, the claim that count-sampling enables accurate single-trajectory solutions would be refuted.

Figures

Figures reproduced from arXiv: 2605.25826 by Charlie Pyle, George Xu, Munawar Ali, Qi Feng.

Figure 5.1
Figure 5.1. Figure 5.1: Neural network lift for branched training architecture of linear ODEs The parameters λshuffle and λmodel are tuned so that the model learns an extension which fits the data while still following the branched (non-geometric) signature properties. Often, a higher weight is placed on the model loss than the shuffle loss so that a meaningful extension that actually aids in learning the right representative o… view at source ↗
Figure 5.2
Figure 5.2. Figure 5.2: Neural network lift for branched training architecture of nonlinear ODEs 6. Numerical Experiments We test the branched signature kernel solver developed in Section 4 and Section 5 on a range of examples. In particular, we focus on the setting where only a single trajectory of the forcing term is observed, motivated by real-world scenarios in which the forcing signal is typically noisy. We compare the per… view at source ↗
Figure 6.1
Figure 6.1. Figure 6.1: El-Centro earthquake calibration. Left: forcing calibration-learned kernel rep￾resentation against −9.81 a(t). Right: displacement-predicted uˆ(t) vs. reference u(t). Next, training and testing were applied using the streaming protocol of Section 5 and the same conditions as the calibration experiment. The initial training window uses the first n0 = 200 ob￾servations, and the rolling retrain-and-predict … view at source ↗
Figure 6.2
Figure 6.2. Figure 6.2: El-Centro streaming prediction. Top row: forcing calibration; bottom row: solution reconstruction. Left: non-branched path (t, −9.81 a(t)). Right: branched path (t, tα, −9.81 a(t)). 6.2. Solow capital-stock model. We apply Method 2 of Section 4.3 to this first order ODE. For this example, we use a modified version of the Solow capital-stock growth model [Solow, 1956], where we use real GDP from FRED [Fed… view at source ↗
Figure 6.3
Figure 6.3. Figure 6.3: Solow ODE calibration. Left: GDP forcing match. Right: capital-stock solu￾tion match. Next, a testing and training experiment is done using the rolling retrain protocol of Section 5.3 with initial training window n0 = 50 and retraining cadence κ = 10 on the non-branched path(t, F(t)). This example is well captured through geometric signature kernels with low depth because the GDP data was not rough enoug… view at source ↗
Figure 6.4
Figure 6.4. Figure 6.4: Solow streaming prediction with the rolling retrain protocol [PITH_FULL_IMAGE:figures/full_fig_p030_6_4.png] view at source ↗
Figure 6.5
Figure 6.5. Figure 6.5: Predicted Forcing for the normal signature vs branched signature model for Problem 6.3 [PITH_FULL_IMAGE:figures/full_fig_p031_6_5.png] view at source ↗
Figure 6.6
Figure 6.6. Figure 6.6: Forcing and Solution calibration and Prediction for a linear ODE driven by FBM. 6.4. Modeling Nonlinear ODE driven by FBM (Duffing Model). The Duffing oscillator ex￾tends the previously mentioned second order SDOF system by adding a nonlinear cubic term. It has a breadth of applications among different fields. [Reynolds et al., 2014] uses the Duffing os￾cillator to extract information on nonlinear vibrat… view at source ↗
Figure 6.7
Figure 6.7. Figure 6.7: Branched Signature Kernel Model fit for the Duffing Equation on test￾ing and training interval ability to encode the branched information not captured by the classical signature. The results for the branched models testing and training fits are presented in figure 6.7 [PITH_FULL_IMAGE:figures/full_fig_p033_6_7.png] view at source ↗
Figure 6.8
Figure 6.8. Figure 6.8: plots of generated Forcing, Arias Intensity, and Stiffness Coefficient Dividing by m to be consistent with the third SDOF coefficient: kef f (t) m = (1 − D(t))k0 m = (1 − D(t))ω 2 0 . As the Arias intensity strongly correlates with damage and destructive potential [Cabañas et al., 1997], it is plausible to introduce a degradation rate δ and use the intensity as a damage variable, giving us the following … view at source ↗
Figure 6.9
Figure 6.9. Figure 6.9: Calibration results for the branched signature kernel and a point-wise (no signatures) RBF kernel on the Arias Intensity degraded SDOF equation 6.9. The RBF kernel follows the general forcing trend, but does not pick up on the roughness that the branched model does. The solution match is generally accurate for the sole RBF model, but fails to be as accurate as the branched model. We note that the RBF par… view at source ↗
Figure 6.10
Figure 6.10. Figure 6.10: Solution and forcing fits for the branched signature kernel model on the noisy Kuramoto oscillator. References [Abbasbandy et al., 2015] Abbasbandy, S., Azarnavid, B., and Alhuthali, M. S. (2015). A shooting reproducing kernel hilbert space method for multiple solutions of nonlinear boundary value problems. Journal of Computational and Applied Mathematics, 279:293–305. [Ali and Feng, 2025] Ali, M. and F… view at source ↗
read the original abstract

We develop a branched signature kernel solver for linear and nonlinear ordinary differential equations driven by a \emph{single observed trajectory} of a possibly rough forcing signal -- a setting that arises naturally in earthquake engineering, finance, biology, and structural health monitoring, where the forcing is observed exactly once and the solver must respect the underlying physical law without recourse to an ensemble of realizations. Two ingredients are new. First, a \emph{count-sampling} construction turns the single observation into a hierarchical family of $N+1$ nested training paths on which the branched signature kernel can be evaluated; this allows the signature kernel machinery, originally designed for multi-realization regression problems, to operate on a single-trajectory observation. Second, a kernel-collocation framework places the ansatz either on the highest-order derivative of the solution (with lower derivatives recovered by integrating the kernel) or on the solution itself (after $m$-fold integration of the ODE). We prove a universal approximation theorem for the branched signature kernel, leveraging the Hairer--Kelly morphism to express branched signature evaluations through geometric signatures of time-extended paths. The offline solver is extended to a streaming Test/Train/Retrain protocol with closed-form online updates in the linear case and scalar Newton steps in the nonlinear case. Numerical experiments on six benchmarks (El-Centro earthquake displacement, the Solow capital-stock model, an fBM-driven second-order ODE, a forced Duffing oscillator, a path-dependent Arias-intensity-degraded oscillator with variable coefficients, and a noisy Kuramoto phase-oscillator system) show that the branched signature-kernel solver delivers accurate, stable predictions across all regimes.

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

3 major / 2 minor

Summary. The manuscript develops a branched signature kernel solver for linear and nonlinear ODEs driven by a single observed rough trajectory. It introduces a count-sampling construction to generate a hierarchical family of N+1 nested training paths from one observation, enabling signature kernel techniques originally for multi-realization settings. A kernel-collocation framework is proposed (ansatz on highest derivative or integrated solution), a universal approximation theorem is proved via the Hairer-Kelly morphism relating branched signatures to geometric signatures of time-extended paths, the solver is extended to a streaming Test/Train/Retrain protocol with closed-form updates, and accuracy is shown on six benchmarks including El-Centro earthquake data, Solow model, fBM-driven ODE, Duffing oscillator, Arias-intensity oscillator, and noisy Kuramoto system.

Significance. If the count-sampling step rigorously preserves the required rough-path structure, the work offers a new approach to single-trajectory rough ODE solvers in applications where ensembles are unavailable, extending signature kernels beyond their standard multi-realization regime. The Hairer-Kelly-based universal approximation theorem and the streaming protocol with closed-form updates constitute clear technical strengths.

major comments (3)
  1. [§3] §3 (count-sampling construction): the claim that the N+1 nested paths allow the branched signature kernel to operate on a single-trajectory observation requires explicit verification that the sampling rule preserves p-variation bounds and the algebraic relations needed for the Hairer-Kelly reduction to geometric signatures; without this, the kernel-collocation ansatz does not necessarily approximate the true solution operator.
  2. [§4] §4 (universal approximation theorem): while the Hairer-Kelly morphism is correctly invoked to express branched evaluations through geometric signatures, the theorem statement must specify the precise dependence of the approximation error on the count-sampling parameter N and on the single-trajectory regularity assumptions, as these are load-bearing for the solver's validity.
  3. [§6] Numerical experiments (§6): the six benchmarks report accurate predictions, but the absence of error bars, multiple realizations, or explicit exclusion criteria for the fBM-driven and noisy Kuramoto cases weakens the robustness claim across regimes; this is secondary to the theoretical gap but still affects the empirical support for the central solver.
minor comments (2)
  1. [§2] Notation for the branched signature kernel and the time-extension map should be introduced with a short self-contained definition in §2 before the count-sampling construction.
  2. [Figures] Figure captions for the benchmark plots should explicitly label the observed forcing trajectory alongside the predicted solution for direct visual comparison.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thorough review and constructive comments on our manuscript. We address each major comment point by point below, indicating the revisions we will implement.

read point-by-point responses

  1. Referee: [§3] §3 (count-sampling construction): the claim that the N+1 nested paths allow the branched signature kernel to operate on a single-trajectory observation requires explicit verification that the sampling rule preserves p-variation bounds and the algebraic relations needed for the Hairer-Kelly reduction to geometric signatures; without this, the kernel-collocation ansatz does not necessarily approximate the true solution operator.

    Authors: We agree that an explicit verification is required. In the revised manuscript we will add a new lemma in §3 proving that the count-sampling rule preserves the necessary p-variation bounds and the algebraic structure required for the Hairer-Kelly morphism to apply, thereby justifying the reduction to geometric signatures of time-extended paths. revision: yes

  2. Referee: [§4] §4 (universal approximation theorem): while the Hairer-Kelly morphism is correctly invoked to express branched evaluations through geometric signatures, the theorem statement must specify the precise dependence of the approximation error on the count-sampling parameter N and on the single-trajectory regularity assumptions, as these are load-bearing for the solver's validity.

    Authors: We accept that the current theorem statement lacks explicit error dependence. We will revise the universal approximation theorem in §4 to state quantitative bounds showing how the approximation error scales with the count-sampling parameter N and with the Hölder or p-variation regularity of the single observed trajectory, and we will update the proof to make these dependencies transparent. revision: yes

  3. Referee: [§6] Numerical experiments (§6): the six benchmarks report accurate predictions, but the absence of error bars, multiple realizations, or explicit exclusion criteria for the fBM-driven and noisy Kuramoto cases weakens the robustness claim across regimes; this is secondary to the theoretical gap but still affects the empirical support for the central solver.

    Authors: For real single-trajectory data such as the El-Centro record, multiple independent realizations are unavailable by nature of the problem. For the synthetic benchmarks we will add error bars computed over multiple simulated paths and include a short paragraph stating the explicit selection criteria used for all six examples. These additions will be made in the revised §6. revision: partial

Circularity Check

0 steps flagged

No significant circularity; central claims rest on external Hairer-Kelly morphism and explicit constructions

full rationale

The paper's universal approximation theorem is explicitly derived from the external Hairer-Kelly morphism applied to time-extended paths, with no reduction to self-citation or fitted inputs. The count-sampling construction is presented as a direct transformation of the single trajectory into N+1 nested paths, without any claim that it is 'predicted' from data or fitted parameters. No load-bearing steps reduce by definition or by self-referential fitting; the kernel-collocation ansatz and streaming updates are algorithmic extensions of the signature kernel framework. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides insufficient detail to identify concrete free parameters, axioms, or invented entities; the method description implies choices such as sample count N and integration order m but does not quantify them.

pith-pipeline@v0.9.1-grok · 5843 in / 1142 out tokens · 23636 ms · 2026-06-29T20:27:58.372977+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

69 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    Abbasbandy, S., Azarnavid, B., and Alhuthali, M. S. (2015). A shooting reproducing kernel hilbert space method for multiple solutions of nonlinear boundary value problems. Journal of Computational and Applied Mathematics , 279:293--305

    2015

  2. [2]

    and Feng, Q

    Ali, M. and Feng, Q. (2025). Branched signature model

    2025

  3. [3]

    Alòs, E., Òscar Burés, de Santiago, R., and Vives, J. (2025). Volatility modeling with rough paths: A signature-based alternative to classical expansions

    2025

  4. [4]

    Arias, A. (1970). A measure of earthquake intensity. Seismic Design for Nuclear Power Plants, R.J.\ Hansen (ed.), MIT Press , pages 438--483

    1970

  5. [5]

    Bayer, C., Gogolashvili, D., and Pelizzari, L. (2025). Local regression on path spaces with signature metrics

    2025

  6. [6]

    Bayraktar, E., Feng, Q., and Zhang, Z. (2024). Deep signature algorithm for multidimensional path-dependent options. SIAM Journal on Financial Mathematics , 15(1):194--214

    2024

  7. [7]

    Boedihardjo, H., Geng, X., Lyons, T., and Yang, D. (2016). The signature of a rough path: Uniqueness. Advances in Mathematics , 293:720--737

    2016

  8. [8]

    Caba \ n as, L., Benito, B., and Herr \'a iz, M. (1997). An approach to the measurement of the potential structural damage of earthquake ground motions. Earthquake Engineering & Structural Dynamics , 26(1):79--92

    1997

  9. [9]

    Kwossek, and David J

    Ceylan, M., Kwossek, A. P., and Pr \"o mel, D. J. (2026). Universal approximation with signatures of non-geometric rough paths. arXiv preprint arXiv:2602.05898

    work page arXiv 2026

  10. [10]

    Chen, Y., Hosseini, B., Owhadi, H., and Stuart, A. M. (2021). Solving and learning nonlinear pdes with gaussian processes

    2021

  11. [11]

    and Kormilitzin, A

    Chevyrev, I. and Kormilitzin, A. (2025). A primer on the signature method in machine learning. In Signature Methods in Finance: An Introduction with Computational Applications , pages 3--64. Springer

    2025

  12. [12]

    and Oberhauser, H

    Chevyrev, I. and Oberhauser, H. (2022). Signature moments to characterize laws of stochastic processes

    2022

  13. [13]

    N., Lui, S., Malpass, W., Mantoan, G., Nesheim, L., Áureo de Paula, Reeves, A., Scott, C., Small, E., and Yang, L

    Cohen, S. N., Lui, S., Malpass, W., Mantoan, G., Nesheim, L., Áureo de Paula, Reeves, A., Scott, C., Small, E., and Yang, L. (2023). Nowcasting with signature methods

    2023

  14. [14]

    and Kreimer, D

    Connes, A. and Kreimer, D. (1999). Hopf algebras, renormalization and noncommutative geometry. In Quantum field theory: perspective and prospective , pages 59--109. Springer

    1999

  15. [15]

    Cox, S., Khedher, A., and Maessen, T. (2026). Universal approximation by signatures for infinite-dimensional rough paths. arXiv preprint arXiv:2603.03058

    work page arXiv 2026

  16. [16]

    Cuchiero, C., Gazzani, G., and Svaluto-Ferro, S. (2023). Signature-based models: theory and calibration. SIAM journal on financial mathematics , 14(3):910--957

    2023

  17. [17]

    Cuchiero, C., Primavera, F., and Svaluto-Ferro, S. (2025). Universal approximation theorems for continuous functions of c \`a dl \`a g paths and l \'e vy-type signature models. Finance and Stochastics , 29(2):289--342

    2025

  18. [18]

    Deng, X., Liu, H., and Long, T. (2012). A new complex duffing oscillator used in complex signal detection. Chinese Science Bulletin , 57

    2012

  19. [19]

    Fang, B., Ni, H., and Wu, Y. (2023). A neural rde-based model for solving path-dependent pdes

    2023

  20. [20]

    Louis (2024)

    Federal Reserve Bank of St. Louis (2024). Real Gross Domestic Product . https://fred.stlouisfed.org/series/A191RL1Q225SBEA. FRED, Federal Reserve Bank of St. Louis. Accessed: 2026-05-19

    2024

  21. [21]

    Feng, Q., Luo, M., and Zhang, Z. (2023). Deep signature fbsde algorithm

    2023

  22. [22]

    and Flyer, N

    Fornberg, B. and Flyer, N. (2015). Solving pdes with radial basis functions. Acta Numerica , 24:215–258

    2015

  23. [23]

    Friz, P. K. and Victoir, N. B. (2010). Multidimensional Stochastic Processes as Rough Paths: Theory and Applications . Cambridge Studies in Advanced Mathematics. Cambridge University Press

    2010

  24. [24]

    and Inzirillo, H

    Genet, R. and Inzirillo, H. (2025). Keras sig: Efficient path signature computation on gpu in keras 3

    2025

  25. [25]

    Green, P., Worden, K., Atallah, K., and Sims, N. (2012). The Benefits of Duffing-type Nonlinearities and Electrical Optimisation of a Randomly Excited Energy Harvester , volume 6, pages 657--667

    2012

  26. [26]

    Gubinelli, M. (2010). Ramification of rough paths. Journal of Differential Equations , 248(4):693--721

    2010

  27. [27]

    Extracting information from the signature of a financial data stream

    Gyurk \'o , L. G., Lyons, T., Kontkowski, M., and Field, J. (2013). Extracting information from the signature of a financial data stream. arXiv preprint arXiv:1307.7244

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  28. [28]

    and Kelly, D

    Hairer, M. and Kelly, D. (2015). Geometric versus non-geometric rough paths. In Annales de l'IHP Probabilit \'e s et statistiques , volume 51, pages 207--251

    2015

  29. [29]

    and Lyons, T

    Hambly, B. and Lyons, T. (2010). Uniqueness for the signature of a path of bounded variation and the reduced path group. Annals of Mathematics , pages 109--167

    2010

  30. [30]

    Hao, B., Braga-Neto, U., Liu, C., Wang, L., and Zhong, M. (2025). Stability in training pinns for stiff pdes: Why initial conditions matter

    2025

  31. [31]

    Horvath, B., Lemercier, M., Liu, C., Lyons, T., and Salvi, C. (2023). Optimal stopping via distribution regression: a higher rank signature approach

    2023

  32. [32]

    Issa, Z., Horvath, B., Lemercier, M., and Salvi, C. (2023). Non-adversarial training of neural sdes with signature kernel scores

    2023

  33. [33]

    Kansa, E. (1990a). Multiquadrics – a scattered data approximation scheme with applications to computational fluid-dynamics. ii: Solutions to parabolic, hyperbolic and elliptic partial differential equations. Computers & Mathematics with Applications , 19:147--161

  34. [34]

    Kansa, E. (1990b). Multiquadrics—a scattered data approximation scheme with applications to computational fluid-dynamics—i surface approximations and partial derivative estimates. Computers & Mathematics with Applications , 19(8):127--145

  35. [35]

    Kidger, P., Bonnier, P., Perez Arribas, I., Salvi, C., and Lyons, T. (2019). Deep signature transforms. In Advances in Neural Information Processing Systems , volume 32. Curran Associates, Inc

    2019

  36. [36]

    and Lyons, T

    Kidger, P. and Lyons, T. (2021). S ignatory: differentiable computations of the signature and logsignature transforms, on both CPU and GPU . In International Conference on Learning Representations . https://github.com/patrick-kidger/signatory

    2021

  37. [37]

    Kir \'a ly, F. J. and Oberhauser, H. (2019). Kernels for sequentially ordered data. Journal of Machine Learning Research , 20(31):1--45

    2019

  38. [38]

    Krishnapriyan, A., Gholami, A., Zhe, S., Kirby, R., and Mahoney, M. (2021). Characterizing possible failure modes in physics-informed neural networks. In Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P., and Vaughan, J. W., editors, Advances in Neural Information Processing Systems , volume 34, pages 26548--26560. Curran Associates, Inc

    2021

  39. [39]

    Kuramoto, Y. (1975). Self-entrainment of a population of coupled non-linear oscillators. In Araki, H., editor, International Symposium on Mathematical Problems in Theoretical Physics , pages 420--422, Berlin, Heidelberg. Springer Berlin Heidelberg

    1975

  40. [40]

    and Leng, Y.-G

    Lai, Z.-H. and Leng, Y.-G. (2015). Generalized parameter-adjusted stochastic resonance of duffing oscillator and its application to weak-signal detection. Sensors , 15(9):21327--21349

    2015

  41. [41]

    V., and Lyons, T

    Lemercier, M., Salvi, C., Damoulas, T., Bonilla, E. V., and Lyons, T. (2021). Distribution regression for sequential data

    2021

  42. [42]

    Levin, D., Lyons, T., and Ni, H. (2016). Learning from the past, predicting the statistics for the future, learning an evolving system

    2016

  43. [43]

    Li, Q. (2001). Free vibration of sdof systems with arbitrary time-varying coefficients. International Journal of Mechanical Sciences , 43(3):759--770

    2001

  44. [44]

    Li, Q., Fang, J., and Liu, D. (2000). Exact solutions for free vibration of single-degree-of-freedom systems with nonperiodically varying parameters. Journal of Vibration and Control , 6:449--462

    2000

  45. [45]

    Li, Q. S. (1999). A new exact approach for analyzing free vibration of sdof systems with nonperiodically time varying parameters. Journal of Vibration and Acoustics , 122(2):175--179

    1999

  46. [46]

    Lobo, D., Ritto, T., Castello, D., and Cataldo, E. (2019). Dynamics of a duffing oscillator with the stiffness modeled as a stochastic process. International Journal of Non-Linear Mechanics , 116:273--280

    2019

  47. [47]

    Lyons, T. (2014). Rough paths, signatures and the modelling of functions on streams

    2014

  48. [48]

    and McLeod, A

    Lyons, T. and McLeod, A. D. (2025). Signature methods in machine learning

    2025

  49. [49]

    Lyons, T., Nejad, S., and Perez Arribas, I. (2020). Non-parametric pricing and hedging of exotic derivatives. Applied Mathematical Finance , 27(6):457--494

    2020

  50. [50]

    J., Caruana, M., and L \'e vy, T

    Lyons, T. J., Caruana, M., and L \'e vy, T. (2007). Differential equations driven by rough paths: Ecole d'Et \'e de Probabilit \'e s de Saint-Flour XXXIV-2004 . Springer

    2007

  51. [51]

    and Sims, N

    Mann, B. and Sims, N. (2009). Energy harvesting from the nonlinear oscillations of magnetic levitation. Journal of Sound and Vibration , 319:515--530

    2009

  52. [52]

    Mohaddes, A., Iafrate, F., and Lederer, J. (2025). Regularized learning for fractional brownian motion via path signatures

    2025

  53. [53]

    and Salvi, C

    Pannier, A. and Salvi, C. (2024). A path-dependent pde solver based on signature kernels

    2024

  54. [54]

    Ramtani, S. (2013). Basic concepts and models in continuum damage mechanics. Technical report, HAL Open Archive. hal-00776729, https://hal.science/hal-00776729/document

    2013

  55. [55]

    and Graham, B

    Reizenstein, J. and Graham, B. (2020). Algorithm 1004: The iisignature library: Efficient calculation of iterated-integral signatures and log signatures. ACM Transactions on Mathematical Software (TOMS)

    2020

  56. [56]

    Reynolds, T., Harris, R., and Chang, W.-S. (2014). Nonlinear pre-yield modal properties of timber structures with large-diameter steel dowel connections. Engineering Structures , 76:235--244

    2014

  57. [57]

    Sabate-Vidales, M., Šiška, D., and Szpruch, L. (2020). Solving path dependent pdes with lstm networks and path signatures

    2020

  58. [58]

    Sakaguchi, H. (1988). Cooperative phenomena in coupled oscillator systems under external fields. Progress of theoretical physics , 79(1):39--46

    1988

  59. [59]

    Salvi, C., Cass, T., Foster, J., Lyons, T., and Yang, W. (2021). The signature kernel is the solution of a goursat pde. SIAM Journal on Mathematics of Data Science , 3(3):873–899

    2021

  60. [60]

    and Salvi, C

    Shmelev, D. and Salvi, C. (2025). pysiglib -- fast signature-based computations on cpu and gpu

    2025

  61. [61]

    Solow, R. M. (1956). A contribution to the theory of economic growth. The Quarterly Journal of Economics , 70(1):65--94

    1956

  62. [62]

    T \'o th, C., Cruz, D. J. D., and Oberhauser, H. (2025). A user's guide to KSig : Gpu-accelerated computation of the signature kernel. arXiv preprint arXiv:2501.07145

    work page arXiv 2025

  63. [63]

    Toth, C., Oberhauser, H., and Szabo, Z. (2024). Random fourier signature features

    2024

  64. [64]

    El centro earthquake

    Vibrationdata . El centro earthquake

  65. [65]

    Wang, S., Sankaran, S., and Perdikaris, P. (2024). Respecting causality for training physics-informed neural networks. Computer Methods in Applied Mechanics and Engineering , 421:116813

    2024

  66. [66]

    Wang, S., Teng, Y., and Perdikaris, P. (2021). Understanding and mitigating gradient flow pathologies in physics-informed neural networks. SIAM Journal on Scientific Computing , 43(5):A3055--A3081

    2021

  67. [67]

    Xiang, J., Guo, J., and Li, X. (2024). A two-stage duffing equation-based oscillator and stochastic resonance for mechanical fault diagnosis. Chaos, Solitons & Fractals , 182:114755

    2024

  68. [68]

    and Unal, G

    Yilmaz, A. and Unal, G. (2019). Stochastic duffing equation in modelling of financial time series. International Journal of Dynamics and Control , 7:1--22

    2019

  69. [69]

    and Jiang, S

    Zeng, P. and Jiang, S. (2025). Semi-parametric functional classification via path signatures logistic regression

    2025