Pith. sign in

REVIEW 4 major objections 5 minor 37 references

Modelling Chemical Reaction Networks using Neural Ordinary Differential Equations

T0 review · 4 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read Adding a neural correction term to a chemical ODE model yields better predictions of oscillation periods at settings the model has never seen.

desk verdict Genuine transfer test for a UDE correction term on a chemical oscillator, but the headline period claim is only three numbers without error bars and an unexplained full-vs-half inconsistency. read the letter →

arxiv 2502.19397 v1 pith:AIEBGSVN submitted 2025-02-11 q-bio.MN cs.LG

classification q-bio.MNcs.LG
keywords neuralordinarydifferentialequationschemicalreactionnetworksmass-actionkineticsmodelmisspecificationoscillatoryreactionsuniversalflowreactortransferlearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Chemical reaction networks are usually modelled by systems of ODEs derived from mass-action kinetics, but these models are incomplete when reactions are hidden or the mixture is not ideal. The paper adds a neural-network correction term to the theoretical ODE system, forming a neural ODE, and trains the correction on measurements of a small-molecule chemical oscillator. The central claim is that the corrected model predicts the period of sustained oscillations at experimental settings never seen in training more accurately than the theory alone: across the three tested settings the nODE period is closer to the observed 4.3, 2.7 and 3.6 h than the ODE-only predictions. The paper also shows that the per-species correction compensates for a deliberately omitted reaction and thus can flag shortcomings of the theoretical model. If true, this makes nODEs a practical tool for designing oscillatory reaction networks without re-fitting the model at every new condition.

What carries the argument

The central object is the universal differential equation $\frac{dy}{dt} = h_\kappa(t, y) + f_\theta(t, y)$, where $h_\kappa$ is the mass-action ODE system inferred from the known reaction network and $f_\theta$ is an LSTM-based neural network that learns the residual between theory and data. The addition is evaluated at every solver step, and its per-species output is read as the neural contribution that tells where the theoretical model goes wrong. This object carries the paper's argument because the same learned correction is frozen and transferred between experiments.

What would settle it

Train the nODE at the 100 mM Fmoc-piperidine / 1.8 M phenyl acetate setting, then measure the period at a new flow setting such as 90 mM Fmoc-piperidine / 2.0 M phenyl acetate with careful records of reactor start-up time; if the nODE period is not closer to the observed period than the ODE-only prediction, the transfer claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that a neural ordinary differential equation formed by adding a trained neural correction to the mass-action ODE system predicts the oscillation period of a chemical oscillator at unseen flow settings more accurately than the theoretical ODE alone. In Table 1, observed periods are 4.3, 2.7 and 3.6 h; the ODE-only predictions are 3.52, 2.37 and 2.33 h, while the full-time-series nODE predicts 3.70, 2.45 and 2.40 h and the half-time-series nODE predicts 4.00, 2.55 and 2.52 h. The same architecture compensates for a deliberately removed inhibition reaction by increasing the neural contribution to the affected species, which the authors read as evidence that the correction term captures model misspecification. The authors also report that the nODE does not improve classification of sustained versus damped oscillations, and that its transfer advantage rests on the assumption that residual noise is independent and identically distributed across experimental settings.

Load-bearing premise

The load-bearing premise is that the differences between the theoretical model and the measurements are mostly the same kind of model error at every experimental setting, with only ordinary random noise varying between settings.

Editorial extensions

If this is right

  • One trained nODE can replace per-setting model calibration for period estimates, since the learned correction transfers between experimental settings without retraining.
  • The per-species neural contribution works as a diagnostic: a large, structured contribution flags a missing interaction, as demonstrated when the slow inhibition pathway was removed from the model.
  • nODEs do not improve the classification of sustained versus damped oscillations; their improvement is limited to the period of the oscillations.
  • Training on only the first half of the time series already recovers most of the period improvement, suggesting the learned correction is not simply memorizing the training interval.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implication the authors leave implicit is that the transfer protocol could be used as a screening tool: train once at a well-characterized setting, then cheaply sweep the predicted period over a grid of flow concentrations before committing to reactor experiments.
  • A testable extension would be to use the per-species neural contributions as input to symbolic regression, turning the residual-capturing network into a proposal for which reaction terms are missing; the paper mentions this direction but does not carry it out.
  • Because the paper's strongest quantitative evidence rests on three settings, the claim of general transferability would need evaluation at more settings and on other oscillatory networks, such as inorganic or enzymatic oscillators, before it can be treated as a general property of nODEs.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes to augment the theoretical ODE model of a chemical reaction network with a neural-network correction term, implemented as a neural ODE (nODE), and applies it to the small-molecule oscillator data of ter Harmsel et al. The authors first show improved in-sample fits for single-pulse and oscillating experiments, then test the trained nODE at new inflow concentrations. Their headline claim is that nODEs are more accurate predictors of the oscillation period at unseen experimental settings than the theoretical ODE alone, supported by Table 1. The paper also reports that nODEs do not improve regime classification (stable vs. damped oscillations), and it discusses how the neural contributions can identify model misspecification. The treatment is honest about the regime-classification failure, and the code and data availability statements are useful.

Significance. If the period-transfer claim is robust, the paper would provide a practical method for using limited experimental data to correct mechanistic models and guide experiments on oscillatory chemical networks. The paper's strengths include: testing on real experimental data rather than only synthetic examples; explicitly separating the neural-network contribution from the mechanistic term; reproducing the honest negative result on regime classification; and sharing source code. The central quantitative claim, however, rests on three un-replicated scalar comparisons with no uncertainty quantification, and the unexplained superiority of the half-time-series model over the full-time-series model is a red flag. The methodology is therefore promising but the evidence as presented is not yet sufficient for the strength of the stated conclusion.

major comments (4)
  1. [Table 1 and 'Predicting the Oscillation Space'] The central claim that nODEs are more accurate predictors of the period at unseen settings is supported only by three single scalar comparisons, with no error bars or seed-averaged statistics. The paper generates N=1000 synthetic training samples, but Table 1 reports one period per model per setting. It is therefore unclear whether the reported differences (e.g., 0.78 h error vs. 0.30 h error at [Fmoc]=0.08, [PhAc]=1.8) are reproducible across random seeds, training draws, or solver tolerances. Please report mean and standard deviation over repeated training runs, or otherwise provide a distribution over predictions, and state clearly whether the reported periods are single runs or averages.
  2. [Predicting the Oscillation Space; Figure 5; Table 1] The nODE trained on the first half of the time series is closer to the observed period than the nODE trained on the full time series at all three settings. This systematic degradation with more training data contradicts the transfer assumption stated in 'Interpreting the Neural Network Contribution', under which more data should improve or at least not degrade transfer. The paper does not explain this result. If the correction term were transferable, the full-data model should not be consistently worse. Please analyze and discuss this pattern; for example, report per-setting neural contributions, residual correlations, and whether the half/full difference persists with seed-averaged comparisons. This is load-bearing because it casts doubt on whether the learned correction is capturing reproducible model misspecification or setting-specific transients.
  3. [Results: Table 1 and Methods] The procedure for estimating the oscillation period from a predicted time series is not described. Table 1 reports observed and predicted periods, and Figures 4 and 5 color regions by period, but there is no equation or algorithm defining how the period is extracted (e.g., peak detection, zero crossings, FFT), nor how the observed experimental periods in Table 1 were determined. Since the quantitative claim is entirely about period values, this missing information prevents reproducibility and makes the comparison hard to audit.
  4. [Interpreting the Neural Network Contribution] The paper's transfer assumption is that residuals are reproducible model misspecification plus i.i.d. noise across settings, and the violation of this assumption is invoked to explain the 120 mM misclassification. However, no evidence is presented that the successful settings satisfy the assumption. A concrete test would be to compare the neural-network contributions and residual autocorrelations across the three successful settings and the failing setting, or to perform leave-one-setting-out training-and-transfer experiments. Without such a check, the success at the three settings could be driven by the same per-setting artifacts that are blamed for the failure at 120 mM.
minor comments (5)
  1. [Equation (2)] The time indices in Equation (2) are inconsistent: the left side is dy/dt at t+δt while the right side is evaluated at t. This should be written as dy/dt(t) = hκ(t,y(t)) + fθ(t,y(t)), or as a discrete update y(t+δt) = y(t) + δt·(...).
  2. [Equation (3)] The equation for d[p-nitrophenyl acetate (3)]/dt reads sv([3]_in - [2]), which should presumably be sv([3]_in - [3]). Please correct the typo.
  3. [Equation (4)] Equation (4) writes dyfθ/dt(t_{n+1}) = fθ(t_n, y(t_n)); the output time index on the left should match the input time index on the right, e.g., dyfθ/dt(t_n) = fθ(t_n, y(t_n)).
  4. [Figure 5] The color scale in panel (b) shows a period difference plot with values in hours, but the caption and text do not explain why the difference is asymmetric (negative values only) or what the units of the colorbar represent in each panel.
  5. [Methods: Neural ODE Training and Inference] The paper says N=1000 train samples are generated by drawing from a normal distribution, but it does not specify how the validation and test sets are constructed or how predictions at new inflow concentrations use these samples. Please clarify whether Table 1 entries are single deterministic solves or aggregates over synthetic draws.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: nODE period predictions are genuine transfers to unseen settings, not refits; minor self-citations are background only.

full rationale

The derivation chain is self-contained. The nODE correction fθ is trained once on data from one flow setting (100 mM Fmoc-piperidine, 1.8 M phenyl acetate), and the paper explicitly states: 'It is important to note that we do not train a nODE for each experimental setting. Instead, we train nODE once and transfer the learned dynamics between experiments.' The Table 1 periods at the other settings are obtained by running that fixed model at those inflow concentrations, not by refitting at those settings. The baseline hκ uses rate constants 'retrieved from ter Harmsel et al.' and is likewise not refit to the test settings, so the nODE-versus-ODE comparison is not a fitted-value comparison. The neural-network 'contribution' analysis is descriptive residual inspection, not a derivation that presupposes its conclusion. The iid-noise transfer assumption in 'Interpreting the Neural Network Contribution' is an explicit modeling assumption (the paper calls it 'a strong assumption') and the failure at 120 mM is attributed to its violation; that is a correctness/robustness concern, not circularity. The only self-citations (refs 3 and 13, with Kachman as co-author; ref 22 shares Huck) are background references or external experimental data; no load-bearing argument reduces to a self-cited uniqueness theorem or ansatz. The unexplained ordering of half- versus full-training periods is a statistical weakness (n=3, no error bars) but does not amount to a definitional equivalence between input and output.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

This paper introduces no new physical entity; the neural network correction is a diagnostic artifact, not a postulated mediator or species. The free parameters are the neural network weights, which are fitted to data derived from two experimental runs, plus training hyperparameters chosen without reported sensitivity analysis. The load-bearing assumptions are domain assumptions: mass-action kinetics with literature rate constants, transferable residuals, well-mixed CSTR conditions. Two of these are explicitly named by the authors as assumptions, which is a point in their favor. The rate constants themselves are inputs from ter Harmsel et al., not fitted here, but neither are they refit, so the base model inherits any calibration error from the source paper.

free parameters (3)
  • Neural network parameters theta (LSTM hidden state 32, linear layer 32) = Learned during training; no single value reported
    The correction term f_theta is fitted by minimizing MSE against 1000 synthetic samples drawn from the mean and standard deviation of two experimental runs. All predictive content of the augmentation is in these weights.
  • Training hyperparameters = Learning rate 6e-3 (AdaBelief); N=1000 train samples; hidden dim 32
    Chosen without a reported sensitivity study; the period results already differ noticeably between the full- and half-time-series training variants, so the results are sensitive to training configuration.
  • Oscillatory-training warmup fraction = First 10% of each time series
    Ad hoc schedule used to prevent getting stuck in a local minimum; it affects the learned correction and is not justified by an ablation.
assumptions (5)
  • domain assumption Mass-action kinetics with rate constants from ter Harmsel et al. correctly describes the base dynamics up to the neural correction term
    Invoked in Methods, Eq. 3; the constants ktr, kac, kinh1, kinh2 and sv are taken from ref 22, not refit here.
  • domain assumption The residuals between theory and data are reproducible model misspecification plus iid noise that transfers across experimental settings
    Stated in 'Interpreting the Neural Network Contribution': 'we assume that the noise around the observations is independent and identically distributed (iid.) across experimental settings'; the authors call this a strong assumption.
  • domain assumption The CSTR is well-mixed and initial conditions are comparable across runs
    The authors note 'the chemicals are not instantaneously well-mixed in the continuous stirred-tank reactor (CSTR), which is an important assumption underlying all work in such reactors'; invoked to explain the fifth-setting failure.
  • standard math Numerical integration error of the Kvaerno 5/4 solver is negligible relative to the correction term
    Implicit in the Methods choice of solver for stiff systems; no tolerances are reported.
  • ad hoc to paper An LSTM with a 32-dimensional hidden state can represent the missing dynamics over the measured time horizon
    Architecture and size are chosen without an approximation argument or sensitivity study; the paper notes the LSTM outperformed a fully connected net on synthetic data (SI).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Modelling Chemical Reaction Networks using Neural Ordinary Differential Equations." pith.science (2026). https://pith.science/paper/AIEBGSVN

@misc{pith2026250219397,
  author       = {Pith},
  title        = {Pith review of: Modelling Chemical Reaction Networks using Neural Ordinary Differential Equations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AIEBGSVN}},
  note         = {Machine review of arXiv:2502.19397}
}
read the original abstract

In chemical reaction network theory, ordinary differential equations are used to model the temporal change of chemical species concentration. As the functional form of these ordinary differential equations systems is derived from an empirical model of the reaction network, it may be incomplete. Our approach aims to elucidate these hidden insights in the reaction network by combining dynamic modelling with deep learning in the form of neural ordinary differential equations. Our contributions not only help to identify the shortcomings of existing empirical models but also assist the design of future reaction networks.

Figures

Figures reproduced from arXiv: 2502.19397 by the authors.

Figure 1
Figure 1. The predictive performance of the nODE on the single-pulse data. (a), the experi [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. The predictive performance of the nODE for which the reaction for N-acetyl [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. The predictive performance of the nODE on the oscillating open system for diben [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: The oscillation spaces predicted by (a) the theoretical model, (b) the nODE trained [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: The difference between the predicted oscillation spaces. (a), the predictions of the [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: The predictive performance of the nODE on the single-pulse data. (a), the experi [PITH_FULL_IMAGE:figures/full_fig_p025_6.png]
Figure 7
Figure 7. Figure 7: The predictive performance of the nODE on the single-pulse data in the open [PITH_FULL_IMAGE:figures/full_fig_p026_7.png]
Figure 8
Figure 8. Figure 8: The predictive performance of the nODE with an incomplete vector field on the [PITH_FULL_IMAGE:figures/full_fig_p027_8.png]
Figure 9
Figure 9. Figure 9: The predictive performance of the nODE on the oscillating data. (a), the ex [PITH_FULL_IMAGE:figures/full_fig_p028_9.png]
Figure 10
Figure 10. Figure 10: The predictive performance of the nODE on the oscillating data in the open [PITH_FULL_IMAGE:figures/full_fig_p029_10.png]
Figure 11
Figure 11. Figure 11: The predictive performance of the nODE on the oscillating data of the missing [PITH_FULL_IMAGE:figures/full_fig_p030_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 31 canonical work pages

  1. [2]

    Mathematical models of chemical reactions: theory and applications of deterministic and stochastic models; Manchester University Press, 1989

    \'E rdi, P.; T \'o th, J. Mathematical models of chemical reactions: theory and applications of deterministic and stochastic models; Manchester University Press, 1989

  2. [3]

    A.; England, J

    Kachman, T.; Owen, J. A.; England, J. L. Self-Organized Resonance During Search of a Diverse Chemical Space . Physical review letters 2017, 119, 038001

  3. [1]

    M.; Waage, P

    Guldberg, C. M.; Waage, P. Ueber die chemische Affinität. § 1. Einleitung. Journal für Praktische Chemie 1879, 19, 69--114

  4. [4]

    F.; Pathirana, D.; Fröhlich, F.; Hasenauer, J.; Banga, J

    Villaverde, A. F.; Pathirana, D.; Fröhlich, F.; Hasenauer, J.; Banga, J. R. A protocol for dynamic model calibration . Briefings in Bioinformatics 2021, 23, bbab387

  5. [5]

    T.; Weindl, D.; Hasenauer, J

    Lakrisenko, P.; Stapor, P.; Grein, S.; Paszkowski, .; Pathirana, D.; Fr \"o hlich, F.; Lines, G. T.; Weindl, D.; Hasenauer, J. Efficient computation of adjoint sensitivities at steady-state in ODE models of biochemical reaction networks. PLOS Computational Biology 2023, 19, e1010783

  6. [6]

    Linking data to models: Data regression

    Jaqaman, K.; Danuser, G. Linking data to models: Data regression. Nature reviews. Molecular cell biology 2006, 7, 813--9

  7. [7]

    Competition for catalytic resources alters biological network dynamics

    Rondelez, Y. Competition for catalytic resources alters biological network dynamics. Physical review letters 2012, 108, 018102

  8. [8]

    Wen, M.; Spotte-Smith, E. W. C.; Blau, S. M.; McDermott, M. J.; Krishnapriyan, A. S.; Persson, K. A. Chemical reaction networks and opportunities for machine learning. Nature Computational Science 2023, 3, 12--24

Show all 37 references
  1. [9]

    T.; Rubanova, Y.; Bettencourt, J.; Duvenaud, D

    Chen, R. T.; Rubanova, Y.; Bettencourt, J.; Duvenaud, D. K. Neural ordinary differential equations. Advances in neural information processing systems 2018, 31

  2. [10]

    Deep learning

    LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. nature 2015, 521, 436--444

  3. [11]

    V.; Azevedo, P.; Cardoso, V

    Badue, C.; Guidolini, R.; Carneiro, R. V.; Azevedo, P.; Cardoso, V. B.; Forechi, A.; Jesus, L.; Berriel, R.; Paixao, T. M.; Mutz, F.; others Self-driving cars: A survey. Expert systems with applications 2021, 165, 113816

  4. [12]

    D.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; others Language models are few-shot learners

    Brown, T.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J. D.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; others Language models are few-shot learners. Advances in neural information processing systems 2020, 33, 1877--1901

  5. [13]

    E.; Bachrach, Y.; Huck, W

    Hödl, S.; Robinson, W. E.; Bachrach, Y.; Huck, W. T. S.; Kachman, T. Explainability Techniques for Chemical Language Models . 2023,

  6. [14]

    nature 2021, 596, 583--589

    Jumper, J.; Evans, R.; Pritzel, A.; Green, T.; Figurnov, M.; Ronneberger, O.; Tunyasuvunakool, K.; Bates, R.; Z \' dek, A.; Potapenko, A.; others Highly accurate protein structure prediction with AlphaFold. nature 2021, 596, 583--589

  7. [15]

    Multilayer feedforward networks are universal approximators

    Hornik, K.; Stinchcombe, M.; White, H. Multilayer feedforward networks are universal approximators. Neural Networks 1989, 2, 359--366

  8. [16]

    Deep Residual Learning for Image Recognition

    He, K.; Zhang, X.; Ren, S.; Sun, J. Deep Residual Learning for Image Recognition. 2015

  9. [17]

    Density estimation using real nvp

    Dinh, L.; Sohl-Dickstein, J.; Bengio, S. Density estimation using real nvp. arXiv preprint arXiv:1605.08803 2016,

  10. [18]

    Neural Controlled Differential Equations for Irregular Time Series

    Kidger, P.; Morrill, J.; Foster, J.; Lyons, T. Neural Controlled Differential Equations for Irregular Time Series. 2020

  11. [19]

    Autonomous discovery of unknown reaction pathways from data by chemical reaction neural network

    Ji, W.; Deng, S. Autonomous discovery of unknown reaction pathways from data by chemical reaction neural network. The Journal of Physical Chemistry A 2021, 125, 1082--1092

  12. [20]

    ChemNODE: A neural ordinary differential equations framework for efficient chemical kinetic solvers

    Owoyele, O.; Pal, P. ChemNODE: A neural ordinary differential equations framework for efficient chemical kinetic solvers. Energy and AI 2022, 7, 100118

  13. [21]

    Universal differential equations for scientific machine learning

    Rackauckas, C.; Ma, Y.; Martensen, J.; Warner, C.; Zubov, K.; Supekar, R.; Skinner, D.; Ramadhan, A.; Edelman, A. Universal differential equations for scientific machine learning. arXiv preprint arXiv:2001.04385 2020,

  14. [22]

    R.; Runikhina, S

    ter Harmsel, M.; Maguire, O. R.; Runikhina, S. A.; Wong, A. S. Y.; Huck, W. T. S.; Harutyunyan, S. R. A catalytically active oscillator made from small organic molecules. Nature 2023, 621, 87–93

  15. [23]

    A.; Essex, C

    Adams, R. A.; Essex, C. Calculus: a complete course, 8th ed.; Pearson: Ontario, 2013

  16. [24]

    Singly diagonally implicit Runge--Kutta methods with an explicit first stage

    Kv rn , A. Singly diagonally implicit Runge--Kutta methods with an explicit first stage. BIT Numerical Mathematics 2004, 44, 489--502

  17. [25]

    Babuschkin, I. et al. The D eep M ind JAX E cosystem. 2020; http://github.com/deepmind

  18. [26]

    E quinox: neural networks in JAX via callable P y T rees and filtered transformations

    Kidger, P.; Garcia, C. E quinox: neural networks in JAX via callable P y T rees and filtered transformations. Differentiable Programming workshop at Neural Information Processing Systems 2021 2021,

  19. [27]

    Zhuang, J.; Tang, T.; Ding, Y.; Tatikonda, S.; Dvornek, N.; Papademetris, X.; Duncan, J. S. AdaBelief Optimizer: Adapting Stepsizes by the Belief in Observed Gradients . 2020

  20. [28]

    Distilling Free-Form Natural Laws from Experimental Data

    Schmidt, M.; Lipson, H. Distilling Free-Form Natural Laws from Experimental Data. Science 2009, 324, 81--85

  21. [29]

    Bishop, C. M. Pattern recognition and machine learning; Springer New York, NY, 2006

  22. [30]

    Long short-term memory

    Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural computation 1997, 9, 1735--1780

  23. [31]

    A Tutorial on Chemical Reaction Network Dynamics

    Angeli, D. A Tutorial on Chemical Reaction Network Dynamics . European Journal of Control - EUR J CONTROL 2009, 15, 398--406

  24. [32]

    Upadhyay, S. K. Chemical Kinetics and Reaction Dynamics; Springer: New York, 2006

  25. [33]

    Levine, R. D. Molecular reaction dynamics; Cambridge University Press: Cambridge, UK, 2009

  26. [34]

    Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations

    Lu, Y.; Zhong, A.; Li, Q.; Dong, B. Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations. International Conference on Machine Learning. 2018; pp 3276--3285

  27. [35]

    O n N eural D ifferential E quations

    Kidger, P. O n N eural D ifferential E quations. Ph.D.\ thesis, University of Oxford, 2021

  28. [36]

    Aro, C. J. CHEMSODE: a stiff ODE solver for the equations of chemical kinetics . Computer Physics Communications 1996, 97, 304--314

  29. [37]

    A protocol for dynamic model calibration

    Alexander, R. Diagonally implicit Runge--Kutta methods for stiff ODE’s. SIAM Journal on Numerical Analysis 1977, 14, 1006--1021 mcitethebibliography main.bib0000664000000000000000000005013114752617565011202 0ustar rootroot =1 @article wen2023chemical, title= Chemical reaction ...

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.