pith. sign in

arxiv: 2606.22075 · v1 · pith:TFVEFDJ6new · submitted 2026-06-20 · 💻 cs.LG · math.DS

Frequency-Domain Neural ODEs for Modeling Non-Linear Dynamical Systems

Mohammed Ashraf , Ayman A. El-Badawy This is my paper

Pith reviewed 2026-06-26 12:26 UTC · model grok-4.3

classification 💻 cs.LG math.DS
keywords neural ODEfrequency domaindynamical systemsgeneralizationFFTnonlinear dynamicscontinuous-depth modelsensemble learning
0
0 comments X

The pith

Frequency-domain projection via FFT lets neural ODEs generalize better to highly nonlinear dynamical systems than standard continuous or discrete models.

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

The paper introduces the Frequency-domain Neural ODE (FNODE) that applies the Fast Fourier Transform to shift the dynamics of a neural ODE into the frequency domain. This step is presented as the fix for the poor performance of ordinary NODEs on strongly nonlinear systems. The model is tested on the Lotka-Volterra equations, forced Duffing oscillator, Van der Pol oscillator, and Lorenz system, using curriculum learning and model ensembles to measure both accuracy and convergence stability. Results are compared against GRUs, LSTMs, and Augmented Neural ODEs. If the frequency-domain step is responsible for the reported gains, it supplies a concrete architectural change that makes continuous-depth models more usable for physical simulation tasks.

Core claim

The FNODE architecture projects continuous temporal dynamics into the frequency domain using the Fast Fourier Transform. By operating in the frequency domain, the model provides better generalization to the dynamical system. Empirical evaluation on four systems shows that FNODE achieves better generalization while exhibiting remarkable convergence stability compared with discrete recurrent models and other continuous-depth variants.

What carries the argument

The FNODE architecture that projects the neural ODE vector field into the frequency domain after an FFT projection.

If this is right

  • FNODE outperforms GRUs, LSTMs, and ANODE on the Lotka-Volterra, Duffing, Van der Pol, and Lorenz systems.
  • Curriculum training combined with ensemble evaluation produces stable convergence whose confidence intervals can be estimated directly.
  • The frequency-domain step addresses the documented difficulty standard NODEs have with highly nonlinear dynamics.
  • The same architecture yields measurable robustness gains across both discrete and continuous baseline families.

Where Pith is reading between the lines

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

  • If the FFT projection is the decisive ingredient, analogous gains could appear when the same step is added to other continuous-depth architectures such as neural SDEs.
  • The stability observed under ensemble evaluation suggests the method may be useful for safety-critical forecasting where uncertainty bounds matter.
  • Further experiments could check whether the frequency representation can be used directly for control without an inverse transform step.
  • The approach invites direct comparison with classical spectral methods already used in numerical integration of nonlinear ODEs.

Load-bearing premise

That operating the neural ODE in the frequency domain after an FFT projection inherently improves generalization on highly nonlinear dynamical systems.

What would settle it

A side-by-side run on the same four systems in which FNODE produces higher test error or wider ensemble confidence intervals than ANODE on held-out trajectories would falsify the claimed generalization benefit.

read the original abstract

Standard continuous-depth models, such as Neural Ordinary Differential Equations (NODEs), offer significant advantages in modeling physical systems by learning continuous vector fields rather than discrete temporal steps. However, when applied to complex dynamical systems, standard NODEs frequently struggle with highly nonlinear dynamics. This paper investigates the Frequency-domain Neural ODE (FNODE), an architecture that projects continuous temporal dynamics into the frequency domain using the Fast Fourier Transform (FFT). By operating in the frequency domain, the model provides better generalization to the dynamical system. The architecture is empirically evaluated against discrete models, specifically Gated Recurrent Units (GRUs) and Long Short-Term Memory (LSTMs), and other continuous-depth variants, including Augmented Neural ODE (ANODE), across four distinct dynamical systems: the Lotka-Volterra model, the forced Duffing oscillator, the Van der Pol oscillator, and the Lorenz system. To rigorously assess generalization and robustness, curriculum and ensemble learning are used to evaluate the model's convergence by estimating confidence intervals across different ensemble models. The empirical results demonstrate that the FNODE architecture achieves better generalization while exhibiting remarkable convergence stability.

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 Frequency-Domain Neural ODEs (FNODE), which apply an FFT projection to operate Neural ODEs in the frequency domain. It evaluates FNODE against GRUs, LSTMs, and ANODE baselines on the Lotka-Volterra, forced Duffing, Van der Pol, and Lorenz systems, using curriculum learning and ensemble methods to report improved generalization and convergence stability for the proposed architecture.

Significance. If the frequency-domain representation can be shown to drive the reported gains independently of training protocol, the work would supply a concrete architectural alternative for stiff or chaotic continuous-depth modeling; the ensemble-based confidence intervals are a methodological strength that supports more rigorous claims about stability.

major comments (2)
  1. [Abstract] Abstract and Experiments section: the central claim that the FFT projection yields better generalization is not isolated from the shared curriculum and ensemble protocol. No matched time-domain NODE (or ANODE) trained under identical curriculum schedules and ensemble confidence-interval estimation is reported, so any observed gap could be driven by the training procedure rather than the frequency-domain representation.
  2. [Abstract] Abstract: the assertion of 'superior generalization' and 'remarkable convergence stability' is stated without quantitative metrics, error bars, or a description of the generalization measure (e.g., test-set rollout error, horizon length, or distribution shift). This prevents direct evaluation of the empirical claim even before considering controls.
minor comments (1)
  1. [Methods] Notation for the frequency-domain vector field and the precise form of the inverse FFT reconstruction step should be stated explicitly in the methods section to allow reproduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these detailed comments, which highlight important issues regarding experimental controls and the presentation of results. We agree that stronger isolation of the frequency-domain contribution and more quantitative support in the abstract are needed, and we will revise the manuscript to address both points directly.

read point-by-point responses

  1. Referee: [Abstract] Abstract and Experiments section: the central claim that the FFT projection yields better generalization is not isolated from the shared curriculum and ensemble protocol. No matched time-domain NODE (or ANODE) trained under identical curriculum schedules and ensemble confidence-interval estimation is reported, so any observed gap could be driven by the training procedure rather than the frequency-domain representation.

    Authors: We agree that the current experiments do not fully isolate the FFT projection because the ANODE baseline was not trained under the identical curriculum schedule and ensemble protocol used for FNODE. To address this, we will add a matched ANODE control trained with the same curriculum learning and ensemble-based confidence interval estimation. The revised Experiments section will report these new results to allow direct attribution of any performance differences to the frequency-domain representation. revision: yes

  2. Referee: [Abstract] Abstract: the assertion of 'superior generalization' and 'remarkable convergence stability' is stated without quantitative metrics, error bars, or a description of the generalization measure (e.g., test-set rollout error, horizon length, or distribution shift). This prevents direct evaluation of the empirical claim even before considering controls.

    Authors: We acknowledge that the abstract currently uses qualitative phrasing without supporting numbers. In the revision we will update the abstract to include concrete quantitative details: specifically, the test-set rollout error (with ensemble standard deviations), the rollout horizon length, and the precise error metric and distribution-shift protocol used. These quantities are already defined and reported in the Experiments section; the abstract will now summarize them explicitly. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical architecture comparison with no derivation chain

full rationale

The manuscript reports an empirical comparison of FNODE against GRUs, LSTMs, ANODE and other baselines on four dynamical systems, with curriculum learning and ensemble confidence intervals applied uniformly. No equations, uniqueness theorems, fitted parameters renamed as predictions, or self-citation chains are present in the provided text that would reduce any claimed result to its inputs by construction. The generalization and stability claims are presented strictly as measured experimental outcomes rather than derived identities, rendering the work self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no information on free parameters, background axioms, or newly postulated entities; the contribution is described as an architectural modification of existing Neural ODEs.

pith-pipeline@v0.9.1-grok · 5724 in / 1105 out tokens · 24926 ms · 2026-06-26T12:26:29.437727+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

55 extracted references · 3 canonical work pages

  1. [1]

    Neural Networks in System Identification,

    Sjöberg, J., Hjalmarsson, H., and Ljung, L., 1994, “Neural Networks in System Identification,” IFAC Proceedings Volumes,27(8), pp. 359–382, IFAC Sympo- sium on System Identification (SYSID’94), Copenhagen, Denmark, 4-6 July

    1994

  2. [2]

    Multilayer feedforward networks are universal approximators,

    Hornik, K., Stinchcombe, M., and White, H., 1989, “Multilayer feedforward networks are universal approximators,” Neural Networks,2(5), pp. 359–366

    1989

  3. [3]

    AComprehensiveReviewofDeepLearn- ing: Architectures, Recent Advances, and Applications,

    Mienye, I.D.andSwart, T.G., 2024, “AComprehensiveReviewofDeepLearn- ing: Architectures, Recent Advances, and Applications,” Information,15(12)

    2024

  4. [4]

    DeepLearningand System Identification,

    Ljung, L., Andersson, C., Tiels, K., andSchön, T.B., 2020, “DeepLearningand System Identification,” IFAC-PapersOnLine,53(2), pp. 1175–1181, 21st IFAC World Congress

    2020

  5. [5]

    Deep networks for system identification: A survey,

    Pillonetto, G., Aravkin, A., Gedon, D., Ljung, L., Ribeiro, A. H., and Schön, T. B., 2025, “Deep networks for system identification: A survey,” Automatica, 171, p. 111907

    2025

  6. [6]

    DeepXDE: A Deep Learning Library for Solving Differential Equations,

    Lu, L., Meng, X., Mao, Z., and Karniadakis, G. E., 2019, “DeepXDE: A Deep Learning Library for Solving Differential Equations,” ArXiv,abs/1907.04502

    arXiv 2019

  7. [7]

    A Proposal on Machine Learning via Dynamical Systems,

    E, W., 2017, “A Proposal on Machine Learning via Dynamical Systems,” Com- munications in Mathematics and Statistics,5, pp. 1 – 11

    2017

  8. [8]

    Subspace State-Space Iden- tification of Nonlinear Dynamical System Using Deep Neural Network with a Bottleneck,

    Yamada, K., Maruta, I., and Fujimoto, K., 2023, “Subspace State-Space Iden- tification of Nonlinear Dynamical System Using Deep Neural Network with a Bottleneck,” IFAC-PapersOnLine,56(1), pp. 102–107, 12th IFAC Symposium on Nonlinear Control Systems NOLCOS 2022

    2023

  9. [9]

    Nonlinear Systems Identification Using Deep Dynamic Neural Networks,

    Ogunmolu, O., Gu, X., Jiang, S., and Gans, N., 2016, “Nonlinear Systems Identification Using Deep Dynamic Neural Networks,”

    2016

  10. [10]

    Scientific Machine Learning Through Physics–Informed Neural Net- works: Where we are and What’s Next,

    Cuomo, S., Di Cola, V. S., Giampaolo, F., Rozza, G., Raissi, M., and Piccialli, F., 2022, “Scientific Machine Learning Through Physics–Informed Neural Net- works: Where we are and What’s Next,” Journal of Scientific Computing,92(3), p. 88

    2022

  11. [11]

    Multilayer perceptron and neural networks,

    Popescu, M., Balas, V. E., Perescu-Popescu, L., and Mastorakis, N. E., 2009, “Multilayer perceptron and neural networks,” WSEAS Transactions on Circuits and Systems archive,8, pp. 579–588

    2009

  12. [12]

    Deep Learning,

    LeCun, Y. and Hinton, G., 2015, “Deep Learning,” Nature,521, pp. 436–44

    2015

  13. [13]

    Why and when can deep-but not shallow-networks avoid the curse of dimensionality: A review,

    Poggio, T., Mhaskar, H., Rosasco, L., Miranda, B., and Liao, Q., 2017, “Why and when can deep-but not shallow-networks avoid the curse of dimensionality: A review,” International Journal of Automation and Computing,14(5), pp. 503– 519

    2017

  14. [14]

    Recurrent Neural Net- works: A Comprehensive Review of Architectures, Variants, and Applications,

    Mienye, I. D., Swart, T. G., and Obaido, G., 2024, “Recurrent Neural Net- works: A Comprehensive Review of Architectures, Variants, and Applications,” Information,15(9)

    2024

  15. [15]

    Continual learning for recurrent neural networks: An empirical evaluation,

    Cossu, A., Carta, A., Lomonaco, V., and Bacciu, D., 2021, “Continual learning for recurrent neural networks: An empirical evaluation,” Neural Networks,143, pp. 607–627

    2021

  16. [16]

    Long Short-Term Memory,

    Hochreiter, S. and Schmidhuber, J., 1997, “Long Short-Term Memory,” Neural Computation,9, pp. 1735–1780

    1997

  17. [17]

    A surveyonlongshort-termmemorynetworksfortimeseriesprediction,

    Lindemann, B., Müller, T., Vietz, H., Jazdi, N., and Weyrich, M., 2021, “A surveyonlongshort-termmemorynetworksfortimeseriesprediction,”Procedia CIRP,99, pp. 650–655, 14th CIRP Conference on Intelligent Computation in Manufacturing Engineering, 15-17 July 2020

    2021

  18. [18]

    RNN-LSTM: From applications to modeling techniques and beyond—Systematic review,

    Al-Selwi, S. M., Hassan, M. F., Abdulkadir, S. J., Muneer, A., Sumiea, E. H., Alqushaibi, A., and Ragab, M. G., 2024, “RNN-LSTM: From applications to modeling techniques and beyond—Systematic review,” Journal of King Saud University - Computer and Information Sciences,36(5), p. 102068

    2024

  19. [19]

    GatedRecurrentUnitsViewedThrough the Lens of Continuous Time Dynamical Systems,

    Jordan, I., Sokół, P., andPark, I., 2021, “GatedRecurrentUnitsViewedThrough the Lens of Continuous Time Dynamical Systems,” Frontiers in Computational Neuroscience,15, p. 678158

    2021

  20. [20]

    Towards Under- standing the Spectral Bias of Deep Learning,

    Cao, Y., Fang, Z., Wu, Y., Zhou, D.-X., and Gu, Q., 2021, “Towards Under- standing the Spectral Bias of Deep Learning,” , pp. 2205–2211

    2021

  21. [21]

    On understanding and overcoming spectral biases of deep neural network learning methods for solving PDEs,

    Xu, Z.-Q. J., Zhang, L., and Cai, W., 2025, “On understanding and overcoming spectral biases of deep neural network learning methods for solving PDEs,” Journal of Computational Physics,530, p. 113905

    2025

  22. [22]

    Closed-form continuous-time neural networks,

    Hasani, R., Lechner, M., Amini, A., Liebenwein, L., Ray, A., Tschaikowski, M., Teschl, G., and Rus, D., 2022, “Closed-form continuous-time neural networks,” Nature Machine Intelligence,4(11), pp. 992–1003

    2022

  23. [23]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems,

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

    2016

  24. [24]

    Multidimensional Ap- proximation of Nonlinear Dynamical Systems,

    Gelß, P., Klus, S., Eisert, J., and Schütte, C., 2019, “Multidimensional Ap- proximation of Nonlinear Dynamical Systems,” Journal of Computational and Nonlinear Dynamics,14(6), p. 061006

    2019

  25. [25]

    Physics-informed machine learning,

    Karniadakis, G., Kevrekidis, Y., Lu, L., Perdikaris, P., Wang, S., and Yang, L., 2021, “Physics-informed machine learning,” Nature Reviews Physics, pp. 1–19

    2021

  26. [26]

    Physics guided neural networks for modelling of non-linear dynamics,

    Robinson, H., Pawar, S., Rasheed, A., and San, O., 2022, “Physics guided neural networks for modelling of non-linear dynamics,” Neural Networks,154, pp. 333–345

    2022

  27. [27]

    When and why PINNs fail to train: A neural tangent kernel perspective,

    Wang, S., Yu, X., and Perdikaris, P., 2022, “When and why PINNs fail to train: A neural tangent kernel perspective,” Journal of Computational Physics,449, p. 110768

    2022

  28. [28]

    Neural Ordinary Differential Equations,

    Chen, T. Q., Rubanova, Y., Bettencourt, J., and Duvenaud, D. K., 2018, “Neural Ordinary Differential Equations,” ArXiv,abs/1806.07366

    Pith/arXiv arXiv 2018

  29. [29]

    2016, in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 1, doi: 10.1109/CVPR.2016.90

    He, K., Zhang, X., Ren, S., and Sun, J., 2016, “Deep Residual Learning for Image Recognition,”2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, doi: 10.1109/CVPR.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  30. [30]

    On neural differential equations.arXiv preprint arXiv:2202.02435, 2022

    Kidger, P., 2022, “On Neural Differential Equations,” doi: 10.48550/arXiv.2202.02435

    work page doi:10.48550/arxiv.2202.02435 2022

  31. [31]

    Dissect- ing Neural ODEs,

    Massaroli, S., Poli, M., Park, J., Yamashita, A., and Asama, H., 2020, “Dissect- ing Neural ODEs,” ArXiv,abs/2002.08071

    arXiv 2020

  32. [32]

    XNODE: A XAI Suite to Understand Neural Ordinary Differential Equations,

    Coelho, C., da Costa, M. F. P., and Ferrás, L. L., 2025, “XNODE: A XAI Suite to Understand Neural Ordinary Differential Equations,” AI,6(5)

    2025

  33. [33]

    Auto- matic differentiation in machine learning: a survey,

    Baydin, A. G., Pearlmutter, B. A., Radul, A., and Siskind, J. M., 2015, “Auto- matic differentiation in machine learning: a survey,” J. Mach. Learn. Res.,18, pp. 153:1–153:43

    2015

  34. [34]

    Enhanc- ing predictive capabilities in data-driven dynamical modeling with automatic differentiation: Koopman and neural ODE approaches,

    RicardoConstante-Amores,C.,Linot,A.J.,andGraham,M.D.,2024,“Enhanc- ing predictive capabilities in data-driven dynamical modeling with automatic differentiation: Koopman and neural ODE approaches,” Chaos: An Interdisci- plinary Journal of Nonlinear Science,34(4), p. 043119

    2024

  35. [35]

    A guide to neural ordinary differen- tial equations: Machine learning for data-driven digital engineering,

    Worsham, J. M. and Kalita, J. K., 2025, “A guide to neural ordinary differen- tial equations: Machine learning for data-driven digital engineering,” Digital Engineering,6, p. 100060

    2025

  36. [36]

    Parameterized neural ordinary differential equa- tions: applications to computational physics problems,

    Lee, K. and Parish, E., 2021, “Parameterized neural ordinary differential equa- tions: applications to computational physics problems,” Proceedings of the Royal Society A,477, p. 20210162

    2021

  37. [37]

    Neural ordinary differential equations for predicting the temporal dynamics of a ZnO solid electrolyte FET,

    Gaurav, A., Song, X., Manhas, S. K., and De Souza, M. M., 2025, “Neural ordinary differential equations for predicting the temporal dynamics of a ZnO solid electrolyte FET,” Journal of Materials Chemistry C,13(8), pp. 2804–2813

    2025

  38. [38]

    LyaNet: A Lyapunov Frame- work for Training Neural ODEs,

    Rodriguez, I. D. J., Ames, A., and Yue, Y., 2022, “LyaNet: A Lyapunov Frame- work for Training Neural ODEs,”International Conference on Machine Learn- ing, https://api.semanticscholar.org/CorpusID:246634091

    2022

  39. [39]

    Neural Ordinary Differential Equa- tions for Model Order Reduction of Stiff Systems,

    Caldana, M. and Hesthaven, J. S., 2025, “Neural Ordinary Differential Equa- tions for Model Order Reduction of Stiff Systems,” International Journal for Numerical Methods in Engineering,126(12), p. e70060

    2025

  40. [40]

    Augmented Neural ODEs,

    Dupont, E., Doucet, A., and Teh, Y. W., 2019, “Augmented Neural ODEs,” ArXiv,abs/1904.01681

    arXiv 2019

  41. [41]

    SNODE: Spectral Discretization of Neural ODEs for System Identification,

    Quaglino, A., Gallieri, M., Masci, J., and Koutn’ik, J., 2019, “SNODE: Spectral Discretization of Neural ODEs for System Identification,” . 6 /

    2019

  42. [42]

    Mono- tonic Neural Ordinary Differential Equation: Time-series Forecasting for Cu- mulative Data,

    Chen, Z., Ding, L., Chu, Z., Qi, Y., Huang, J.-B., and Wang, H., 2023, “Mono- tonic Neural Ordinary Differential Equation: Time-series Forecasting for Cu- mulative Data,” Proceedings of the 32nd ACM International Conference on Information and Knowledge Management

    2023

  43. [43]

    Stabilized Neural Ordinary Differential Equations for Long- Time Forecasting of Dynamical Systems,

    Linot, A. J., Burby, J. W., Tang, Q., Balaprakash, P., Graham, M. D., and Maulik, R., 2022, “Stabilized Neural Ordinary Differential Equations for Long- Time Forecasting of Dynamical Systems,” ArXiv,abs/2203.15706

    arXiv 2022

  44. [44]

    Neural ordinary differential equations with irregular and noisy data,

    Goyal, P. and Benner, P., 2023, “Neural ordinary differential equations with irregular and noisy data,” Royal Society Open Science,10(7), p. 221475

    2023

  45. [45]

    Multiple Shooting for Training Neural Differential Equations on Time Series,

    Turan, E. M. and Jäschke, J., 2021, “Multiple Shooting for Training Neural Differential Equations on Time Series,” IEEE Control Systems Letters,6, pp. 1897–1902

    2021

  46. [46]

    Neural ODE-Based Fre- quency Stability Assessment and Control of Energy Storage Systems,

    Gao, S., Liu, E., Wu, Z., Li, J., and Zhang, M., 2025, “Neural ODE-Based Fre- quency Stability Assessment and Control of Energy Storage Systems,” Applied Sciences,15(22)

    2025

  47. [47]

    Warm, comforting recollection

    Dang, T., Dimitriadis, A., Wu, J., Sethu, V., and Ambikairajah, E., 2023, “Constrained Dynamical Neural ODE for Time Series Mod- elling: A Case Study on Continuous Emotion Prediction,” pp. 1–5, doi: 10.1109/ICASSP49357.2023.10095778

    work page doi:10.1109/icassp49357.2023.10095778 2023

  48. [48]

    STEER : Simple Temporal Regularization For Neural ODEs,

    Ghosh, A., Behl, H. S., Dupont, E., Torr, P. H. S., and Namboodiri, V., 2020, “STEER : Simple Temporal Regularization For Neural ODEs,” ArXiv, abs/2006.10711

    arXiv 2020

  49. [49]

    Autoencoders reloaded,

    Bourlard, H. and Kabil, S., 2022, “Autoencoders reloaded,” Biological Cyber- netics,116

    2022

  50. [50]

    2007,van der Pol Oscillator, Springer New York, pp. 505–508

    2007

  51. [51]

    Korsch, H. J. and Jodl, H.-J., 1994,The Duffing Oscillator, Springer Berlin Heidelberg, pp. 157–180

    1994

  52. [52]

    Bacaër, N., 2011,Lotka, Volterra and the predator–prey system (1920–1926), Springer London, pp. 71–76

    2011

  53. [53]

    Origin and structure of the Lorenz attractor,

    Afraimovich, V., Bykov, V., and Shilnikov, L., 1977, “Origin and structure of the Lorenz attractor,” Akademiia Nauk SSSR Doklady,234, pp. 336–339

    1977

  54. [54]

    Discrete Fourier transform techniques for noise reduction and digital enhancement of analytical signals,

    Wahab, M. F., Gritti, F., and O’Haver, T. C., 2021, “Discrete Fourier transform techniques for noise reduction and digital enhancement of analytical signals,” TrAC Trends in Analytical Chemistry,143, p. 116354

    2021

  55. [55]

    Adaptive Asynchronous Control Using Meta-Learned Neural Ordinary Differential Equations,

    Salehi, A., Rühl, S., and Doncieux, S., 2024, “Adaptive Asynchronous Control Using Meta-Learned Neural Ordinary Differential Equations,” IEEE Transac- tions on Robotics,40, pp. 403–420. / 7 Residual Network 0 1 2 3 4 5 Depth −5 0 5 Input/Hidden/Output ODE Network 0 1 2 3 4 5 Depth −5 0 5 Input/Hidden/Output Overcomplete Encoder-Decoder 𝑋 𝜙𝑒𝑛𝑐 𝑧 𝜙𝑑𝑒𝑐 ˆ𝑋 Lo...

    2024