pith. sign in

arxiv: 2604.25655 · v1 · submitted 2026-04-28 · 📊 stat.ML · cs.LG

Residual-loss Anomaly Analysis of Physics-Informed Neural Networks: An Inverse Method for Change-point Detection in Nonlinear Dynamical Systems with Regime Switching

Yuhe Bai , Chengli Tan , Jiaqi Li , Xiangjun Wang , Zhikun Zhang This is my paper

Pith reviewed 2026-05-07 14:31 UTC · model grok-4.3

classification 📊 stat.ML cs.LG
keywords physics-informed neural networkschange-point detectionnonlinear dynamical systemsregime switchingresidual lossparameter estimationinverse problems
0
0 comments X

The pith

Physics-informed neural networks jointly detect change points and recover piecewise parameters in regime-switching dynamical systems by flagging elevations in local residual losses.

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

The paper develops a unified method that uses physics-informed neural networks to handle dynamical systems whose parameters jump between regimes. Traditional approaches first find the transition times and then fit the parameters separately, but this paper shows the two tasks are coupled so tightly that separating them reduces accuracy. Instead, the method scans overlapping subintervals for spikes in how much the network violates the governing equations; intervals that straddle a true jump produce a distinctly higher residual with a positive lower bound when no noise is present. These candidate intervals are then fed into a single loss function that optimizes both the transition locations and the piecewise parameter values at once. Experiments on standard test systems confirm the joint procedure improves both localization of the jumps and the accuracy of the recovered parameters.

Core claim

The residual-loss anomaly analysis of physics-informed neural networks jointly infers piecewise parameters and transition points under a single set of constraints. It proceeds in two stages: overlapping subinterval decomposition first identifies potential transition intervals by the structural elevation that appears in the local physical residual whenever a subinterval contains a true parameter jump, an elevation that possesses a non-zero lower bound under noise-free conditions; the second stage folds the candidate locations and the piecewise parameters into one unified physical loss that is minimized simultaneously, thereby solving the coupled inverse problem without decoupling the tasks.

What carries the argument

Overlapping subinterval decomposition of the local physical residual inside a physics-informed neural network, where any subinterval that spans a true regime transition produces a structural elevation whose lower bound remains strictly positive in the absence of noise.

If this is right

  • Change-point localization and parameter estimation both become more accurate than in methods that treat the two tasks separately.
  • The same set of physical constraints suffices for simultaneous recovery of transition times and piecewise coefficients.
  • The approach applies directly to standard nonlinear benchmarks including logistic growth, Van der Pol, Lotka-Volterra, and Lorenz systems.
  • Because the elevation is guaranteed to have a positive lower bound in noise-free data, candidate intervals can be screened before the joint optimization step.

Where Pith is reading between the lines

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

  • The two-stage procedure might be iterated to locate multiple transitions by successively refining the subintervals around each detected jump.
  • The same residual-elevation signature could be monitored during training to decide when to split an interval and introduce an additional parameter piece.
  • If the lower-bound property holds approximately under moderate noise, the framework could be extended to experimental time series by adding a statistical threshold on the observed residuals.

Load-bearing premise

When a subinterval contains a true transition point, the local physical residual exhibits a distinct structural elevation that has a non-zero lower bound even without measurement noise.

What would settle it

On a noise-free simulation of the Lotka-Volterra model with a known jump at a chosen time, the residual computed over the subinterval that straddles the jump remains at the same level as residuals from intervals that lie entirely inside one regime.

Figures

Figures reproduced from arXiv: 2604.25655 by Chengli Tan, Jiaqi Li, Xiangjun Wang, Yuhe Bai, Zhikun Zhang.

Figure 1
Figure 1. Figure 1: Schematic illustration of the proposed RAA-PINNs framework for change-point detection and parameter view at source ↗
Figure 2
Figure 2. Figure 2: Malthus model. Top row: the output of Pmptq over r0, 100s (left) and the corresponding sample path (right). Bottom row: Stage I identifies the candidate change-point interval determined from the physical loss in the overlapping domain (left), and Stage II results for refined parameter estimation and change-point localization (right). Accordingly, the temporal evolution of rptq follows rptq : 1 Ñ 2. The pie… view at source ↗
Figure 3
Figure 3. Figure 3: Logistic model. Top row: the output of Plptq over r0, 100s (left) and the corresponding sample path (right). Bottom row: Stage I candidate change-point interval (left) and Stage II refined parameter estimation and change-point localization (right). single-regime ODE, xptq “ Plptq and θ “ rl . We consider a time-varying logistic equation with a change point. Introducing a discrete state variable rptq over r… view at source ↗
Figure 4
Figure 4. Figure 4: Van der Pol oscillator model. Top row: the output of view at source ↗
Figure 5
Figure 5. Figure 5: Lotka-Volterra model. First row: trajectories of view at source ↗
Figure 6
Figure 6. Figure 6: Lorenz model. Top row: trajectories of pUptq, V ptq, Wptqq over r0, 20s (left) and the corresponding sample path (right). Second row: Stage I candidate change-point intervals determined from the physical loss in the overlapping domain (left), and Stage II results for parameter estimation and change-point localization (right). Other rows: Stage II parameter estimation results. 32 view at source ↗
Figure 7
Figure 7. Figure 7: Learned parameter trajectories for the Lotka-Volterra model obtained by different methods, where the view at source ↗
Figure 8
Figure 8. Figure 8: First and second rows: learned parameter trajectories view at source ↗
read the original abstract

Nonlinear dynamical systems with regime transitions are typically described by ordinary differential equations with jumping parameters parameters. Traditional methods often treat change-point detection and parameter estimation as separate tasks, ignoring the inherent coupling between them. To address this, we propose residual-loss anomaly analysis of physics-informed neural networks, a unified framework that leverages dynamical consistency within the physics-informed learning paradigm. This approach jointly infers piecewise parameters and transition points under a single set of constraints. The method follows a two-stage strategy: First, local physical residuals are analyzed through overlapping subinterval decomposition. When a subinterval spans a true transition point, the residual exhibits a distinct structural elevation in noise-free conditions, which has a non-zero lower bound, enabling effective localization of potential transition intervals. Second, within our framework, change-point locations and piecewise parameters are integrated into a unified physical loss function for joint optimization, enabling simultaneous identification. Experiments on benchmark nonlinear dynamical systems, including Malthusian and logistic growth models, Van der Pol oscillator, Lotka-Volterra model and Lorenz system, demonstrate that the proposed method outperforms traditional decoupled approaches in both change-point localization and parameter estimation accuracy. This study provides an efficient, unified solution for structurally coupled inverse problems in nonlinear dynamical systems with regime switching.

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 proposes residual-loss anomaly analysis of physics-informed neural networks (PINNs) as a unified two-stage framework for jointly detecting change points and estimating piecewise parameters in nonlinear dynamical systems with regime switching. Stage 1 decomposes trajectories into overlapping subintervals and localizes candidate transitions by identifying structural elevations in the local physical residual, which the authors claim possesses a non-zero lower bound in noise-free conditions. Stage 2 folds the detected intervals and parameters into a single physics-informed loss for simultaneous optimization. Experiments on the Malthusian and logistic growth models, Van der Pol oscillator, Lotka-Volterra system, and Lorenz equations are reported to outperform traditional decoupled change-point and parameter-estimation pipelines.

Significance. If the non-zero lower-bound claim on straddling residuals can be rigorously established and the method demonstrably improves localization and estimation accuracy, the work would constitute a meaningful contribution to inverse problems for hybrid dynamical systems. It offers a principled way to exploit dynamical consistency rather than treating detection and identification as independent tasks, which could benefit applications in ecology, neuroscience, and fluid dynamics where regime switches are common. The use of PINNs for joint inference is a natural extension of existing physics-informed learning techniques.

major comments (2)
  1. [Abstract] Abstract (and the first-stage localization strategy): The central claim that any subinterval containing a true parameter jump produces a physical residual with a 'distinct structural elevation' and 'non-zero lower bound' in noise-free conditions is load-bearing for the entire pipeline, yet the manuscript provides no derivation, explicit lower bound, or analysis of its dependence on jump magnitude, window length, or the approximation capacity of the PINN. Without such a guarantee, it is unclear whether the elevation reliably exceeds optimizer error or can be driven arbitrarily close to zero for small jumps or edge cases.
  2. [Experiments] Experiments section: The abstract asserts superior performance in both change-point localization and parameter estimation on standard benchmarks, but the description supplies no quantitative metrics, error bars, ablation studies on window size or overlap, or direct comparisons against specific baseline algorithms. This absence prevents assessment of whether the joint-optimization stage actually improves upon the decoupled approaches it criticizes.
minor comments (1)
  1. [Abstract] The abstract would benefit from a concise statement of the precise conditions (e.g., noise level, jump size relative to window) under which the non-zero lower bound is asserted to hold.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and indicate the revisions planned for the next version of the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract (and the first-stage localization strategy): The central claim that any subinterval containing a true parameter jump produces a physical residual with a 'distinct structural elevation' and 'non-zero lower bound' in noise-free conditions is load-bearing for the entire pipeline, yet the manuscript provides no derivation, explicit lower bound, or analysis of its dependence on jump magnitude, window length, or the approximation capacity of the PINN. Without such a guarantee, it is unclear whether the elevation reliably exceeds optimizer error or can be driven arbitrarily close to zero for small jumps or edge cases.

    Authors: We acknowledge that the manuscript presents the non-zero lower bound as an observed property arising from the dynamical inconsistency when a single-regime PINN is applied across a true switch, but does not supply a formal derivation. In the revised version we will add a dedicated theoretical paragraph deriving an explicit lower bound on the integrated residual for straddling intervals. The bound will be expressed in terms of the squared parameter jump magnitude, the overlap length, and the Lipschitz constant of the vector field; we will also include a brief discussion of its sensitivity to small jumps and to the residual approximation error of the PINN. revision: yes

  2. Referee: [Experiments] Experiments section: The abstract asserts superior performance in both change-point localization and parameter estimation on standard benchmarks, but the description supplies no quantitative metrics, error bars, ablation studies on window size or overlap, or direct comparisons against specific baseline algorithms. This absence prevents assessment of whether the joint-optimization stage actually improves upon the decoupled approaches it criticizes.

    Authors: We agree that the experimental presentation is insufficiently quantitative. Although the manuscript contains visual results on the listed benchmark systems, it lacks tabulated error statistics and systematic comparisons. In the revision we will expand the Experiments section with tables reporting mean localization error and parameter RMSE together with standard deviations over repeated runs, ablation tables for window length and overlap ratio, and head-to-head numerical comparisons against concrete baselines (PELT + separate PINN fitting, Bayesian online change-point detection + least-squares, and a two-stage residual-threshold method). Error bars will be added to all figures. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper proposes a two-stage residual-loss anomaly analysis within the PINN framework for joint change-point detection and parameter estimation in regime-switching ODEs. The core claim—that straddling subintervals produce structurally elevated residuals with a non-zero lower bound—is presented as a direct consequence of dynamical inconsistency when a single parameter set is applied across a true transition, rather than being defined in terms of the subsequent optimization outputs or fitted values. The unified loss function integrates the localized intervals but does not equate the final piecewise parameters or transition points to quantities already determined by construction in the first stage. No self-citation load-bearing steps, ansatz smuggling, or renaming of known results appear in the provided text; the method adds independent content via the anomaly-based localization prior to joint inference. Experimental comparisons on benchmark systems further indicate external validation rather than tautological reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the domain assumption that dynamical systems obey ODEs with piecewise constant parameters and that PINN residuals can be made to exhibit detectable structural anomalies at true transitions.

axioms (2)
  • domain assumption Nonlinear dynamical systems with regime transitions are described by ordinary differential equations with jumping parameters.
    Stated in the opening of the abstract as the modeling premise.
  • ad hoc to paper Local physical residuals exhibit a distinct structural elevation with non-zero lower bound when a subinterval spans a true transition point in noise-free conditions.
    This property is invoked to enable localization of transition intervals.

pith-pipeline@v0.9.0 · 5543 in / 1399 out tokens · 77466 ms · 2026-05-07T14:31:52.664537+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

52 extracted references · 52 canonical work pages

  1. [1]

    J. Bian, T. Huang, X. Zhang, C. Wang, Y. Zhang, C. Zeng, Fluctuation-variable correlation as early warning signals of non-equilibrium critical transitions, Physica A: Statistical Mechanics and Its Applications 661 (2025) 130401

    work page 2025

  2. [2]

    Evers, D

    K. Evers, D. Borsboom, E. I. Fried, F. Hasselman, L. Waldorp, Early warning signals of complex critical transitions in deterministic dynamics, Nonlinear Dynamics 112 (21) (2024) 19071–19094

    work page 2024

  3. [3]

    Masuda, K

    N. Masuda, K. Aihara, N. G. MacLaren, Anticipating regime shifts by mixing early warning signals from different nodes, Nature Communications 15 (1) (2024) 1086

    work page 2024

  4. [4]

    Dmitriev, V

    A. Dmitriev, V. Kornilov, V. Dmitriev, N. Abbas, Early warning signals for critical transitions in sandpile cellular automata, Frontiers in Physics 10 (2022) 839383

    work page 2022

  5. [5]

    Dakos, C

    V. Dakos, C. A. Boulton, J. E. Buxton, J. F. Abrams, B. Arellano-Nava, D. I. Armstrong McKay, S. Bathiany, L. Blaschke, N. Boers, D. Dylewsky, et al., Tipping point detection and early warnings in climate, ecological, and human systems, Earth System Dynamics 15 (4) (2024) 1117–1135

    work page 2024

  6. [6]

    Southall, T

    E. Southall, T. S. Brett, M. J. Tildesley, L. Dyson, Early warning signals of infectious disease transitions: a review, Journal of the Royal Society Interface 18 (182) (2021) 20210555

    work page 2021

  7. [7]

    Zhang, Q

    Z. Zhang, Q. Shen, X. Wang, Parameter identification framework of nonlinear dynamical systems with markovian switching, Chaos: An Interdisciplinary Journal of Nonlinear Science 33 (12) (2023) 123117

    work page 2023

  8. [8]

    Bettini, M

    L. Bettini, M. Cenedese, G. Haller, Model reduction to spectral submanifolds in piecewise smooth dynamical systems, International Journal of Non-Linear Mechanics 163 (2024) 104753

    work page 2024

  9. [9]

    X. Pan, H. Shu, L. Wang, X.-S. Wang, J. Yu, On the periodic solutions of switching scalar dynamical systems, Journal of Differential Equations 415 (2025) 365–382

    work page 2025

  10. [10]

    Lemus, B

    J. Lemus, B. Herrmann, Multi-objective sindy for parameterized model discovery from single tran- sient trajectory data, Nonlinear Dynamics 113 (10) (2025) 10911–10927

    work page 2025

  11. [11]

    K. Egan, W. Li, R. Carvalho, Automatically discovering ordinary differential equations from data with sparse regression, Communications Physics 7 (1) (2024) 20

    work page 2024

  12. [12]

    Strebel, Preprocessing algorithms for the estimation of ordinary differential equation models with polynomial nonlinearities, Nonlinear Dynamics 111 (8) (2023) 7495–7510

    O. Strebel, Preprocessing algorithms for the estimation of ordinary differential equation models with polynomial nonlinearities, Nonlinear Dynamics 111 (8) (2023) 7495–7510

    work page 2023

  13. [13]

    W. Zhai, D. Tao, Y. Bao, Parameter estimation and modeling of nonlinear dynamical systems based on runge–kutta physics-informed neural network, Nonlinear Dynamics 111 (22) (2023) 21117–21130

    work page 2023

  14. [14]

    Peherstorfer, K

    B. Peherstorfer, K. Willcox, Data-driven operator inference for nonintrusive projection-based model reduction, Computer Methods in Applied Mechanics and Engineering 306 (2016) 196–215

    work page 2016

  15. [15]

    Panahi, Y.-C

    S. Panahi, Y.-C. Lai, Global phase-space approach to rate-induced tipping: A brief review, Chaos: An Interdisciplinary Journal of Nonlinear Science 35 (4) (2025) 043139. 42

    work page 2025

  16. [16]

    Z. Liu, X. Zhang, X. Ru, T.-T. Gao, J. M. Moore, G. Yan, Early predictor for the onset of critical transitions in networked dynamical systems, Physical Review X 14 (3) (2024) 031009

    work page 2024

  17. [17]

    S. L. Brunton, J. L. Proctor, J. N. Kutz, Discovering governing equations from data by sparse identification of nonlinear dynamical systems, Proceedings of the National Academy of Sciences 113 (15) (2016) 3932–3937

    work page 2016

  18. [18]

    S. H. Rudy, S. L. Brunton, J. L. Proctor, J. N. Kutz, Data-driven discovery of partial differential equations, Science Advances 3 (4) (2017) e1602614

    work page 2017

  19. [19]

    Truong, L

    C. Truong, L. Oudre, N. Vayatis, Selective review of offline change point detection methods, Signal Processing 167 (2020) 107299

    work page 2020

  20. [20]

    Fryzlewicz, Wild binary segmentation for multiple change-point detection, The Annals of Statistics 42 (6) (2014)

    P. Fryzlewicz, Wild binary segmentation for multiple change-point detection, The Annals of Statistics 42 (6) (2014)

    work page 2014

  21. [21]

    Dette, W

    H. Dette, W. Wu, Z. Zhou, Change point analysis of correlation in non-stationary time series, Statistica Sinica 29 (2) (2019) 611–643

    work page 2019

  22. [22]

    Baranowski, Y

    R. Baranowski, Y. Chen, P. Fryzlewicz, Narrowest-over-threshold detection of multiple change points and change-point-like features, Journal of the Royal Statistical Society Series B: Statistical Method- ology 81 (3) (2019) 649–672

    work page 2019

  23. [23]

    Dette, J

    H. Dette, J. G¨ osmann, A likelihood ratio approach to sequential change point detection for a general class of parameters, Journal of the American Statistical Association 115 (531) (2020) 1361–1377

    work page 2020

  24. [24]

    P. J. Green, D. I. Hastie, Reversible jump mcmc, Genetics 155 (3) (2009) 1391–1403

    work page 2009

  25. [25]

    J. Wang, E. Zivot, A bayesian time series model of multiple structural changes in level, trend, and variance, Journal of Business & Economic Statistics 18 (3) (2000) 374–386

    work page 2000

  26. [26]

    A. T. Levin, J. Piger, Bayesian model selection for structural break models, Available at SSRN 1132463 (2008)

    work page 2008

  27. [27]

    N. A. Heard, M. J. Turcotte, Adaptive sequential monte carlo for multiple changepoint analysis, Journal of Computational and Graphical Statistics 26 (2) (2017) 414–423

    work page 2017

  28. [28]

    X. Xiu, Y. Yang, L. Kong, W. Liu, Laplacian regularized robust principal component analysis for process monitoring, Journal of Process Control 92 (2020) 212–219

    work page 2020

  29. [29]

    Birg´ e, Model selection via testing: an alternative to (penalized) maximum likelihood estimators, Annales de l’IHP Probabilit´ es et Statistiques 42 (3) (2006) 273–325

    L. Birg´ e, Model selection via testing: an alternative to (penalized) maximum likelihood estimators, Annales de l’IHP Probabilit´ es et Statistiques 42 (3) (2006) 273–325

    work page 2006

  30. [30]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-informed machine learning, Nature Reviews Physics 3 (6) (2021) 422–440

    work page 2021

  31. [31]

    Kovachki, Z

    N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, A. Anandkumar, Neural operator: Learning maps between function spaces with applications to pdes, Journal of Machine Learning Research 24 (89) (2023) 1–97

    work page 2023

  32. [32]

    Cuomo, V

    S. Cuomo, V. S. Di Cola, F. Giampaolo, G. Rozza, M. Raissi, F. Piccialli, Scientific machine learning through physics–informed neural networks: Where we are and what’s next, Journal of Scientific Computing 92 (3) (2022) 88

    work page 2022

  33. [33]

    Kissas, Y

    G. Kissas, Y. Yang, E. Hwuang, W. R. Witschey, J. A. Detre, P. Perdikaris, Machine learning in cardiovascular flows modeling: Predicting arterial blood pressure from non-invasive 4d flow mri data using physics-informed neural networks, Computer Methods in Applied Mechanics and Engineering 358 (2020) 112623

    work page 2020

  34. [34]

    E. O. Oluwasakin, A. Q. Khaliq, Optimizing physics-informed neural network in dynamic system simulation and learning of parameters, Algorithms 16 (12) (2023) 547. 43

    work page 2023

  35. [35]

    Zhang, Y

    G. Zhang, Y. Duan, G. Pan, Q. Chen, H. Yang, Z. Zhang, Data-driven discovery of state-changes in underlying system from hidden change-points in partial differential equations with spatiotemporal varying coefficients, Journal of Computational and Applied Mathematics (2025) 116962

    work page 2025

  36. [36]

    Lux, Estimation of regime-switching diffusions via fourier transforms, Statistics and Computing 34 (2) (2024) 88

    T. Lux, Estimation of regime-switching diffusions via fourier transforms, Statistics and Computing 34 (2) (2024) 88

    work page 2024

  37. [37]

    Niknejad, H

    N. Niknejad, H. Modares, Physics-informed data-driven safe and optimal control design, IEEE Con- trol Systems Letters 8 (2023) 285–290

    work page 2023

  38. [38]

    Huang, C.-W

    Y.-J. Huang, C.-W. Chang, C.-h. Hsieh, Detecting shifts in nonlinear dynamics using empirical dy- namic modeling with nested-library analysis, PLOS Computational Biology 20 (1) (2024) e1011759

    work page 2024

  39. [39]

    Raissi, G

    M. Raissi, G. E. Karniadakis, Hidden physics models: Machine learning of nonlinear partial differ- ential equations, Journal of Computational Physics 357 (2018) 125–141

    work page 2018

  40. [40]

    Zhang, Y

    Y. Zhang, Y. Duan, X. Wang, Z. Zhang, Solving fokker–planck-kolmogorov equation by distribution self-adaptation normalized physics-informed neural networks, Physica A: Statistical Mechanics and Its Applications (2026) 131251

    work page 2026

  41. [41]

    Farid, Unsupervised data-driven response regime exploration and identification for dynamical systems, Chaos: An Interdisciplinary Journal of Nonlinear Science 34 (12) (2024)

    M. Farid, Unsupervised data-driven response regime exploration and identification for dynamical systems, Chaos: An Interdisciplinary Journal of Nonlinear Science 34 (12) (2024)

    work page 2024

  42. [42]

    Stiasny, G

    J. Stiasny, G. S. Misyris, S. Chatzivasileiadis, Physics-informed neural networks for non-linear system identification for power system dynamics, in: 2021 IEEE Madrid PowerTech, IEEE, 2021, pp. 1–6

    work page 2021

  43. [43]

    Mishra, R

    S. Mishra, R. Molinaro, Estimates on the generalization error of physics-informed neural networks for approximating pdes, IMA Journal of Numerical Analysis 43 (1) (2023) 1–43

    work page 2023

  44. [44]

    Z. Lin, Y. Li, F. Yin, J. Maro˜ nas, A. H. Thi´ ery, Efficient transformed gaussian process state- space models for non-stationary high-dimensional dynamical systems, IEEE Transactions on Signal Processing 73 (2025) 5229–5243

    work page 2025

  45. [45]

    Teschl, Ordinary differential equations and dynamical systems, Vol

    G. Teschl, Ordinary differential equations and dynamical systems, Vol. 140, American Mathematical Soc., 2012

    work page 2012

  46. [46]

    J. J. Huerta y Munive, G. Struth, Predicate transformer semantics for hybrid systems: Verification components for isabelle/hol, Journal of Automated Reasoning 66 (1) (2022) 93–139

    work page 2022

  47. [47]

    Quarteroni, R

    A. Quarteroni, R. Sacco, F. Saleri, Numerical mathematics, Vol. 37, Springer Science & Business Media, 2006

    work page 2006

  48. [48]

    Raissi, P

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational Physics 378 (2019) 686–707

    work page 2019

  49. [49]

    Mishra, R

    S. Mishra, R. Molinaro, Estimates on the generalization error of physics-informed neural networks for approximating a class of inverse problems for pdes, IMA Journal of Numerical Analysis 42 (2) (2022) 981–1022

    work page 2022

  50. [50]

    Y. Qian, Y. Zhang, Y. Huang, S. Dong, Physics-informed neural networks for approximating dynamic (hyperbolic) pdes of second order in time: Error analysis and algorithms, Journal of Computational Physics 495 (2023) 112527

    work page 2023

  51. [51]

    Dondelinger, D

    F. Dondelinger, D. Husmeier, S. Rogers, M. Filippone, Ode parameter inference using adaptive gradient matching with gaussian processes, in: Artificial Intelligence and Statistics, PMLR, 2013, pp. 216–228

    work page 2013

  52. [52]

    Killick, P

    R. Killick, P. Fearnhead, I. A. Eckley, Optimal detection of changepoints with a linear computational cost, Journal of the American Statistical Association 107 (500) (2012) 1590–1598. 44

    work page 2012