Pith. sign in

REVIEW 5 major objections 4 minor 54 references

Inferring processes within dynamic forest models using hybrid modeling

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Replacing a forest gap model's growth equations with a jointly calibrated neural network improves forecasts and lets researchers read the learned growth form back out of the network.

desk verdict The abstract promises a hybrid forest model, but the full text is an unrelated diffusion-model paper, so the FINN claims are unassessable and the manuscript should be desk rejected with an invitation to resubmit the real one. read the letter →

arxiv 2508.01228 v2 pith:KTJN42R3 submitted 2025-08-02 q-bio.QM cs.LGstat.ML

Maximilian Pichler , Yannek Käber This is my paper
classification q-bio.QMcs.LGstat.ML MSC 68T0792D40
keywords hybridmodelingforestgapmodelsdynamicvegetationdeepneuralnetworksjointcalibrationfunctionalforminferenceexplainableAIBarroColoradoIsland
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

The paper introduces Forest Informed Neural Networks (FINN), a hybrid simulator that keeps the mechanistic skeleton of a forest gap model but substitutes a deep neural network for the growth process. The network and the remaining mechanistic components are calibrated together in one step, so the data can correct the model's functional form instead of being forced through a fixed equation. On the Barro Colorado Island 50-hectare plot, the authors report that this substitution improves predictive performance and succession trajectories relative to the fully mechanistic version of FINN, and that the trained network's growth response, extracted with explainable-AI methods, is an 'ecologically plausible, improved functional form.' If this holds, hybrid modeling would give ecosystem forecasters a way to identify and fix misspecified process equations in dynamic vegetation models rather than tuning them by hand.

What carries the argument

FINN, Forest Informed Neural Networks: a hybrid in which a deep neural network replaces one process (here, tree growth) inside an otherwise mechanistic forest gap model. The load-bearing design choice is joint calibration: DNN weights and the parameters of the remaining mechanistic processes are optimized in one unified step, so the network is trained against the full model's dynamics rather than against a precomputed target. The second mechanism is the explainable-AI step that reads the learned growth response out of the network, converting the fitted black box back into an inspectable functional form. Together they turn a simulator into a process-inference device.

What would settle it

Simulate forest trajectories with a known, deliberately misspecified growth equation while keeping other processes correct, fit FINN, extract the learned growth form, and compare it to the known generating function; if the extracted form converges to the true growth function, the identification claim survives, and if it tracks the misspecification or the other processes' errors instead, the claim fails. A complementary check compares the extracted growth response against independent tree-ring or census-based growth measurements from the same plot that were not used in calibration.

Watch

Extended reading notes

Core claim

The central claim is that a dynamic forest model need not choose between mechanistic transparency and empirical flexibility: the same model can have both if a DNN takes over one process and is fit jointly with the rest. The authors report that, for the Barro Colorado Island case, replacing the growth submodel of their gap model with a DNN outperforms a mechanistic FINN on both raw predictive accuracy and the community succession trajectory. They further report that the fitted network encodes a growth response that is ecologically plausible and is an improved functional form relative to the mechanistic growth equation, and that this form can be extracted with explainable AI. The paper's stated ambition is to make process inference from forest data a standard byproduct of DVM calibration, so that forecasts under novel climates are grounded in data-corrected mechanisms.

Load-bearing premise

The paper's conclusion depends on attributing the data's unexplained variance to a misspecified growth equation; if the other gap-model components, such as mortality, recruitment, and resource competition, are also wrong, the network will absorb their errors and the 'ecologically plausible' extracted form will be an artifact.

Editorial extensions

If this is right

  • Dynamic vegetation model forecasts can be improved without rewriting the whole model: only the suspect process is replaced, and data corrects its form.
  • Process-level functional forms become recoverable outputs of model calibration, not just fitted parameters inside fixed equations.
  • Joint calibration avoids the plug-in estimators the paper identifies as ineffective for inferring process structure from forest data.
  • The same scheme should transfer to other dynamic vegetation models and to other processes, such as mortality, recruitment, and resource competition, where functional-form assumptions are doubtful.

Reading between the lines

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

  • If the claim is right, a hierarchy of hybrid models becomes attractive: start fully mechanistic, use explainable AI to find which process the data wants to change, replace only that process, and iterate; the paper's argument implies this loop without spelling it out.
  • Transfer to novel climates is the open risk: a growth form learned from a single plot's observed dynamics is fit to that plot's realized climate and competition, so testing on independent plots or on synthetically shifted climates is a natural next experiment.
  • A strict test of the recovered form would be to simulate forests with a known synthetic growth function, let FINN learn from those simulated data, and check whether explainable-AI extraction recovers the generating function better than the mechanistic equation does.
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

5 major / 4 minor

Summary. The abstract of arXiv:2508.01228 announces Forest Informed Neural Networks (FINN), a hybrid forest gap model in which a deep neural network replaces the growth process and is calibrated jointly with the remaining mechanistic components. The abstract claims that, on the Barro Colorado Island 50-ha plot, FINN improves predictive performance and succession trajectories relative to a mechanistic version, and that explainable-AI extraction from the trained DNN reveals an ecologically plausible, improved growth functional form. However, the full text attached to this submission is an unrelated paper titled 'Point-wise Diffusion Models for Physical Systems with Shape Variations: Application to Spatio-temporal and Large-scale system', which concerns diffusion models for cylinder flow, OLED drop impact, and road-car aerodynamics. The submission therefore contains no equations, data, results, or methods that pertain to FINN, forest dynamics, or the Barro Colorado Island case study. As submitted, the central claims are unverifiable.

Significance. If the FINN claims were supported, the work could be a useful contribution to ecological forecasting: jointly calibrating neural and mechanistic components, demonstrating improved DVM predictions at a well-known tropical forest plot, and using explainable AI to recover a process-level growth functional form would be of interest to quantitative ecology and scientific machine learning. The manuscript, however, provides none of the supporting material. There are no equations defining FINN, no calibration procedure, no comparison to a mechanistic baseline, no error metrics or uncertainty quantification, and no description of the explainable-AI extraction or the ecological plausibility criterion. The body text is a different paper on diffusion-based surrogate models for physical systems, and its tables and figures concern fluid dynamics, solid mechanics, and aerodynamics. Consequently, the potential significance of the claimed FINN contribution cannot be assessed from this submission.

major comments (5)
  1. [Abstract vs. Full text (Sections 1–7)] The abstract describes Forest Informed Neural Networks and a Barro Colorado Island case study, but the full text is the unrelated paper 'Point-wise Diffusion Models for Physical Systems with Shape Variations'. None of the equations (e.g., Eqs. (1)–(11)), tables (Tables 2–10), or figures in the full text concern forest dynamics, growth processes, succession trajectories, or the BCI 50-ha plot. This is a load-bearing structural failure: the central claim of the abstract has no supporting text in the manuscript.
  2. [Abstract; no FINN definition in body] The abstract states that FINN 'replaces processes with DNNs' and calibrates them 'alongside the other mechanistic components in one unified step', but the manuscript does not define the forest gap model, its state variables, the growth process being replaced, the DNN architecture, or the joint calibration objective. Without these definitions, the claim cannot be checked or reproduced.
  3. [Abstract; no quantitative comparison] The abstract asserts improved predictive performance and succession trajectories compared to a mechanistic version of FINN, but reports no error metrics, no baseline definition, no train/test split, no validation protocol, and no uncertainty quantification. The full text's quantitative tables (Tables 6–10) compare point-wise diffusion models with DeepONet and Meshgraphnet on physical systems, not with mechanistic forest models.
  4. [Abstract; explainable-AI extraction claim] The abstract claims that 'the DNN learned an ecologically plausible, improved functional form of the growth process, which we extracted from the DNN using explainable AI', but the submission does not identify the XAI method, the extraction protocol, the functional form obtained, or the operational definition of ecological plausibility. In addition, since the DNN is calibrated jointly on the same BCI data, an extracted form evaluated against those same data risks being a restatement of the fit rather than an independent discovery; the manuscript provides no holdout, perturbation, or synthetic-data test to address this circularity concern.
  5. [Abstract; identifiability of the growth process] The design presumes that unexplained variance in the BCI data can be attributed to a misspecified growth function while mortality, recruitment, and resource competition are correctly specified. If these other components are also misspecified, the DNN can absorb their errors and the extracted 'growth' functional form would be an artifact. The submission contains no experiment, simulation test, or argument that isolates the growth process; this is a central methodological risk that is not addressed anywhere in the manuscript.
minor comments (4)
  1. [Title and subject classification] The q-bio.QM subject classification and the title in the abstract refer to forest dynamics, but the full-text title and content are a physics/ML paper on diffusion models; these are inconsistent and should be reconciled by the authors before any resubmission.
  2. [Abstract, 'plug-in estimators' claim] The statement that 'current approaches, such as plug-in estimators, have proven ineffective' is an unsupported claim without citation or quantitative evidence; it should be substantiated or removed.
  3. [Full text, Data availability] The full text states 'Data will be made available on request' for the diffusion-model datasets, but no data or code availability statement is provided for the Barro Colorado Island case study or for FINN, which would be essential for reproducibility.
  4. [Full text, terminology] The acronym FINN appears only in the abstract; even a minimal description of the model architecture or a pseudocode summary would be required in the body to make the claims meaningful.

Circularity Check

1 steps flagged · score 4.0 of 10

The FINN abstract's 'discovered improved growth form' is the jointly calibrated DNN read back out; no independent validation is described, so the central discovery partly restates the fit.

  1. fitted input called prediction [Abstract (FINN); the only FINN content in arXiv:2508.01228]
    "FINN replaces processes with DNNs, which are then calibrated alongside the other mechanistic components in one unified step. In a case study on the Barro Colorado Island 50-ha plot we demonstrate that replacing the growth process with a DNN improves predictive performance and succession trajectories compared to a mechanistic version of FINN. Furthermore, we discovered that the DNN learned an ecologically plausible, improved functional form of the growth process, which we extracted from the DNN using explainable AI."

    The DNN's weights are the fitted parameters, calibrated on the BCI case study. The 'improved functional form' is then obtained by extracting that same calibrated DNN via explainable AI. The abstract reports no held-out or otherwise independent evaluation of the extracted form; it presents the in-sample fitted mapping as a discovered process-level result. Thus the 'discovery' is a restatement of the fit by construction, and the improvement claim is forced rather than independently tested. The absence of the FINN methods text (the submitted full text is an unrelated diffusion-model paper) prevents any non-circular identification argument from being checked.

full rationale

The full text of arXiv:2508.01228 is a different manuscript on point-wise diffusion models, so the FINN claim has no equations, data splits, or XAI details available for verification. The only circularity that can be assessed is internal to the abstract: the extracted functional form is read out of the jointly calibrated DNN, and no external validation is claimed, so presenting it as an independently 'discovered improved' growth process partly reduces to the calibration itself. This is a single central step, not a chain, and there is no load-bearing self-citation issue. Score 4 reflects one 'prediction' that reduces by construction, with the caveat that the missing methods could in principle contain a held-out validation that would remove the circularity. The unrelated full text is a structural evidentiary failure, not itself a circular argument.

Assumptions & free parameters 2 free parameters · 4 assumptions · 1 invented entities

Ledger reconstructed from the abstract only, because the manuscript body is an unrelated diffusion-model paper. FINN's learned growth parameters are fitted to the BCI case-study data; the identification assumptions above are the load-bearing premises. No external benchmarks are described, so nothing in the ledger has independent support visible in the abstract.

free parameters (2)
  • Learned parameters of the DNN growth process (weights and biases) = not reported in abstract
    Calibrated jointly with the mechanistic components on BCI data; the central claim of an improved growth functional form depends entirely on this fit.
  • DNN architecture and training hyperparameters = not reported in abstract
    Layer counts, learning rate, regularization, and training schedule are chosen by hand and affect the learned growth form; none are stated in the abstract.
assumptions (4)
  • domain assumption Dynamic Vegetation Models represent ecological processes mechanistically, making the gap model a sound base for hybrid calibration
    Abstract frames DVMs as mechanistic and FINN as replacing only the growth process with a DNN; if the gap model skeleton is wrong, the hybrid result is unsound.
  • domain assumption The Barro Colorado Island 50-ha plot data contain enough signal to identify the growth functional form
    The abstract's case study rests on a single plot; identification of an 'improved' growth form requires the data to be informative about growth separately from other processes.
  • domain assumption All mechanistic components other than growth are correctly specified
    If mortality, recruitment, or competition are misspecified, the DNN absorbs their errors and the extracted growth form is not identifiable. This is the load-bearing identification assumption.
  • domain assumption Explainable AI attribution faithfully represents the functional form implemented by the DNN
    The abstract asserts the extracted form is ecologically plausible and improved; this requires the XAI method to be a faithful surrogate of the network's growth response.
invented entities (1)
  • FINN (Forest Informed Neural Networks)
    purpose: Hybrid architecture replacing the growth process of a forest gap model with a deep neural network, calibrated jointly with mechanistic components.
    The abstract gives no falsifiable handle outside the BCI case study; the entity's value rests on in-sample predictive improvement and the extracted growth form, which cannot be independently verified from the submission.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inferring processes within dynamic forest models using hybrid modeling." pith.science (2026). https://pith.science/paper/KTJN42R3

@misc{pith2026250801228,
  author       = {Pith},
  title        = {Pith review of: Inferring processes within dynamic forest models using hybrid modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KTJN42R3}},
  note         = {Machine review of arXiv:2508.01228}
}
read the original abstract

Modeling forest dynamics under novel climatic conditions requires a careful balance between process-based understanding and empirical flexibility. Dynamic Vegetation Models (DVM) represent ecological processes mechanistically, but their performance is prone to misspecified assumptions about functional forms. Inferring the structure of these processes and their functional forms correctly from data remains a major challenge because current approaches, such as plug-in estimators, have proven ineffective. We introduce Forest Informed Neural Networks (FINN), a hybrid modeling approach that combines a forest gap model with deep neural networks (DNN). FINN replaces processes with DNNs, which are then calibrated alongside the other mechanistic components in one unified step. In a case study on the Barro Colorado Island 50-ha plot we demonstrate that replacing the growth process with a DNN improves predictive performance and succession trajectories compared to a mechanistic version of FINN. Furthermore, we discovered that the DNN learned an ecologically plausible, improved functional form of the growth process, which we extracted from the DNN using explainable AI. In conclusion, our new hybrid modeling approach offers a versatile opportunity to infer forest dynamics from data and to improve forecasts of ecosystem trajectories under unprecedented environmental change.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 25 canonical work pages

  1. [1]

    Kang, Generative ai-driven design optimization: eight key application scenarios, JMST Advances (2025) 1–7

    N. Kang, Generative ai-driven design optimization: eight key application scenarios, JMST Advances (2025) 1–7

    work page 2025

  2. [2]

    S. Yang, R. Vinuesa, N. Kang, Long-term auto-regressive prediction using lightweight ai models: Adams-bashforth time integration with adaptive multi-step rollout, arXiv preprint arXiv:2412.05657 (2024)

    arXiv 2024

  3. [3]

    S. Yang, Y . Lee, N. Kang, Physics-guided multi-fidelity deeponet for data-e fficient flow field prediction, arXiv preprint arXiv:2503.17941 (2025)

    arXiv 2025

  4. [4]

    P. Du, M. H. Parikh, X. Fan, X.-Y . Liu, J.-X. Wang, Conditional neural field latent diffusion model for generating spatiotemporal turbulence, Nature Communications 15 (1) (2024) 10416

    work page 2024

  5. [5]

    X. Fan, D. Akhare, J.-X. Wang, Neural di fferentiable modeling with di ffusion-based super-resolution for two-dimensional spatiotemporal turbulence, Computer Methods in Applied Mechanics and Engineering 433 (2025) 117478

    work page 2025

  6. [6]

    Z. Li, S. Patil, F. Ogoke, D. Shu, W. Zhen, M. Schneier, J. R. Buchanan Jr, A. B. Farimani, Latent neural pde solver: A reduced-order modeling framework for partial differential equations, Journal of Computational Physics 524 (2025) 113705

    2025

  7. [7]

    A. Zhou, Z. Li, M. Schneier, J. R. Buchanan Jr, A. B. Farimani, Text2pde: Latent di ffusion models for accessible physics simulation, arXiv preprint arXiv:2410.01153 (2024)

    arXiv 2024

  8. [8]

    Jadhav, J

    Y . Jadhav, J. Berthel, C. Hu, R. Panat, J. Beuth, A. B. Farimani, Stressd: 2d stress estimation using denoising di ffusion model, Computer Methods in Applied Mechanics and Engineering 416 (2023) 116343

    work page 2023

Show all 54 references
  1. [9]

    J. Xie, J. Zhang, H. Zhou, Z. Li, Z. Li, Spatiotemporal modeling based on manifold learning for collision dynamic prediction of thin-walled structures under oblique load, Computer Methods in Applied Mechanics and Engineering 440 (2025) 117926

    2025

  2. [10]

    Shin, A.-h

    S. Shin, A.-h. Jin, S. Yoo, S. Lee, C. Kim, S. Heo, N. Kang, Wheel impact test by deep learning: prediction of location and magnitude of maximum stress, Structural and Multidisciplinary Optimization 66 (1) (2023) 24

    2023

  3. [11]

    Cheng, L

    J. Cheng, L. Wang, H. Jin, X. Qian, Attention-based multi-fidelity deep neural network for e fficient estimation of welding residual stresses in v-shaped butt-welded high strength steel plate, Expert Systems with Applications 266 (2025) 126137

    2025

  4. [12]

    R. Lam, A. Sanchez-Gonzalez, M. Willson, P. Wirnsberger, M. Fortunato, F. Alet, S. Ravuri, T. Ewalds, Z. Eaton-Rosen, W. Hu, et al., Learning skillful medium-range global weather forecasting, Science 382 (6677) (2023) 1416–1421

    2023

  5. [13]

    Kochkov, J

    D. Kochkov, J. Yuval, I. Langmore, P. Norgaard, J. Smith, G. Mooers, M. Kl ¨ower, J. Lottes, S. Rasp, P. D ¨uben, et al., Neural general circulation models for weather and climate, Nature 632 (8027) (2024) 1060–1066

    2024

  6. [14]

    F. Alet, I. Price, A. El-Kadi, D. Masters, S. Markou, T. R. Andersson, J. Stott, R. Lam, M. Willson, A. Sanchez-Gonzalez, et al., Skillful joint probabilistic weather forecasting from marginals, arXiv preprint arXiv:2506.10772 (2025)

    2025 arXiv

  7. [15]

    Ogoke, P

    F. Ogoke, P. Pak, A. Myers, G. Quirarte, J. Beuth, J. Malen, A. B. Farimani, Deep learning for melt pool depth contour prediction from surface thermal images via vision transformers, Additive Manufacturing Letters 11 (2024) 100243. 31

    2024

  8. [16]

    R ¨uhling Cachay, B

    S. R ¨uhling Cachay, B. Zhao, H. Joren, R. Yu, Dyffusion: A dynamics-informed diffusion model for spatiotemporal forecasting, Advances in neural information processing systems 36 (2023) 45259–45287

    2023

  9. [17]

    Kohl, L.-W

    G. Kohl, L.-W. Chen, N. Thuerey, Benchmarking autoregressive conditional di ffusion models for turbulent flow simulation, arXiv preprint arXiv:2309.01745 (2023)

    2023 arXiv

  10. [18]

    B. Song, C. Yuan, F. Permenter, N. Arechiga, F. Ahmed, Surrogate modeling of car drag coe fficient with depth and normal renderings, in: International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, V ol. 87301, American Societ...

    2023

  11. [19]

    Jiang, G

    J. Jiang, G. Li, Y . Jiang, L. Zhang, X. Deng, Transcfd: A transformer-based decoder for flow field prediction, Engineering Applications of Artificial Intelligence 123 (2023) 106340

    2023

  12. [20]

    Pfa ff, M

    T. Pfa ff, M. Fortunato, A. Sanchez-Gonzalez, P. Battaglia, Learning mesh-based simulation with graph networks, in: International conference on learning representations, 2020

    2020

  13. [21]

    Fortunato, T

    M. Fortunato, T. Pfa ff, P. Wirnsberger, A. Pritzel, P. Battaglia, Multiscale meshgraphnets, arXiv preprint arXiv:2210.00612 (2022)

    2022 arXiv

  14. [22]

    M. A. Nabian, C. Liu, R. Ranade, S. Choudhry, X-meshgraphnet: Scalable multi-scale graph neural networks for physics simulation, arXiv preprint arXiv:2411.17164 (2024)

    2024 arXiv

  15. [23]

    Y . Cao, M. Chai, M. Li, C. Jiang, E fficient learning of mesh-based physical simulation with bi-stride multi-scale graph neural network, in: International conference on machine learning, PMLR, 2023, pp. 3541–3558

    2023

  16. [24]

    J. Kim, J. Park, N. Kim, Y . Yu, K. Chang, C.-S. Woo, S. Yang, N. Kang, Physics-constrained graph neural networks for spatio-temporal prediction of drop impact on oled display panels, Expert Systems with Applications 274 (2025) 126907

    2025

  17. [25]

    X. Han, H. Gao, T. Pfa ff, J.-X. Wang, L.-P. Liu, Predicting physics in mesh-reduced space with temporal attention, arXiv preprint arXiv:2201.09113 (2022)

    2022 arXiv

  18. [26]

    Alkin, A

    B. Alkin, A. F ¨urst, S. Schmid, L. Gruber, M. Holzleitner, J. Brandstetter, Universal physics transformers: A framework for efficiently scaling neural operators, Advances in Neural Information Processing Systems 37 (2024) 25152–25194

    2024

  19. [27]

    Zhdanov, M

    M. Zhdanov, M. Welling, J.-W. van de Meent, Erwin: A tree-based hierarchical transformer for large-scale physical systems, arXiv preprint arXiv:2502.17019 (2025)

    2025 arXiv

  20. [28]

    Serrano, T

    L. Serrano, T. X. Wang, E. Le Naour, J.-N. Vittaut, P. Gallinari, Aroma: Preserving spatial structure for latent pde modeling with local neural fields, Advances in Neural Information Processing Systems 37 (2024) 13489–13521

    2024

  21. [29]

    L. Lu, P. Jin, G. Pang, Z. Zhang, G. E. Karniadakis, Learning nonlinear operators via deeponet based on the universal approximation theorem of operators, Nature machine intelligence 3 (3) (2021) 218–229

    2021

  22. [30]

    J. He, S. Koric, S. Kushwaha, J. Park, D. Abueidda, I. Jasiuk, Novel deeponet architecture to predict stresses in elastoplastic structures with variable complex geometries and loads, Computer Methods in Applied Mechanics and Engineering 415 (2023) 116277

    2023

  23. [31]

    J. He, S. Koric, D. Abueidda, A. Najafi, I. Jasiuk, Geom-deeponet: A point-cloud-based deep operator network for field predictions on 3d parameterized geometries, Computer Methods in Applied Mechanics and Engineering 429 (2024) 117130

    2024

  24. [32]

    S. Kim, M. Seo, N. Kang, Decoupled dynamics framework with neural fields for 3d spatio-temporal prediction of vehicle collisions, arXiv preprint arXiv:2503.19712 (2025)

    2025

  25. [33]

    Rahaman, A

    N. Rahaman, A. Baratin, D. Arpit, F. Draxler, M. Lin, F. Hamprecht, Y . Bengio, A. Courville, On the spectral bias of neural networks, in: International conference on machine learning, PMLR, 2019, pp. 5301–5310

    2019

  26. [34]

    Oommen, A

    V . Oommen, A. Bora, Z. Zhang, G. E. Karniadakis, Integrating neural operators with di ffusion models improves spectral representation in turbulence modelling, Proceedings of the Royal Society A 481 (2309) (2025) 20240819

    2025

  27. [35]

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, Y . Bengio, Generative adversarial nets, Advances in neural information processing systems 27 (2014)

    2014

  28. [36]

    D. P. Kingma, M. Welling, Auto-encoding variational bayes, arXiv preprint arXiv:1312.6114 (2013)

    2013 arXiv

  29. [37]

    Sohl-Dickstein, E

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, S. Ganguli, Deep unsupervised learning using nonequilibrium thermodynamics, in: Inter- national conference on machine learning, pmlr, 2015, pp. 2256–2265

    2015

  30. [38]

    J. Ho, A. Jain, P. Abbeel, Denoising diffusion probabilistic models, Advances in neural information processing systems 33 (2020) 6840–6851

    2020

  31. [39]

    J. Song, C. Meng, S. Ermon, Denoising di ffusion implicit models, arXiv preprint arXiv:2010.02502 (2020)

    2020 arXiv

  32. [40]

    J. Liu, F. Yu, T. Yan, B. He, C. G. Soares, Cfd-driven physics-informed generative adversarial networks for predicting auv hydrodynamic performance, Ocean Engineering 313 (2024) 119638

    2024

  33. [41]

    Jiang, Z

    H. Jiang, Z. Nie, R. Yeo, A. B. Farimani, L. B. Kara, Stressgan: A generative deep learning model for two-dimensional stress distribution prediction, Journal of Applied Mechanics 88 (5) (2021) 051005

    2021

  34. [42]

    Y .-E. Kang, S. Yang, K. Yee, Physics-aware reduced-order modeling of transonic flow via�-variational autoencoder, Physics of Fluids 34 (7) (2022)

    2022

  35. [43]

    D. Shu, Z. Li, A. B. Farimani, A physics-informed di ffusion model for high-fidelity flow field reconstruction, Journal of Computational Physics 478 (2023) 111972

    2023

  36. [44]

    Ogoke, Q

    F. Ogoke, Q. Liu, O. Ajenifujah, A. Myers, G. Quirarte, J. Malen, J. Beuth, A. B. Farimani, Inexpensive high fidelity melt pool models in additive manufacturing using generative deep diffusion, Materials & Design 245 (2024) 113181

    2024

  37. [45]

    Drygala, E

    C. Drygala, E. Ross, F. di Mare, H. Gottschalk, Comparison of generative learning methods for turbulence modeling, arXiv preprint arXiv:2411.16417 (2024)

    2024

  38. [46]

    Z. Nie, H. Jiang, L. B. Kara, Stress field prediction in cantilevered structures using convolutional neural networks, Journal of Computing and Information Science in Engineering 20 (1) (2020) 011002

    2020

  39. [47]

    H. Gao, S. Kaltenbach, P. Koumoutsakos, Generative learning of the solution of parametric partial differential equations using guided diffusion models and virtual observations, Computer Methods in Applied Mechanics and Engineering 435 (2025) 117654

    2025

  40. [48]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, A. Anandkumar, Fourier neural operator for parametric partial differential equations, arXiv preprint arXiv:2010.08895 (2020)

    2020 arXiv

  41. [49]

    Z. Li, N. Kovachki, C. Choy, B. Li, J. Kossaifi, S. Otta, M. A. Nabian, M. Stadler, C. Hundt, K. Azizzadenesheli, et al., Geometry-informed neural operator for large-scale 3d pdes, Advances in Neural Information Processing Systems 36 (2023) 35836–35854. 32

    2023

  42. [50]

    Peebles, S

    W. Peebles, S. Xie, Scalable di ffusion models with transformers, in: Proceedings of the IEEE /CVF international conference on computer vision, 2023, pp. 4195–4205

    2023

  43. [51]

    Mildenhall, P

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, R. Ng, Nerf: Representing scenes as neural radiance fields for view synthesis, Communications of the ACM 65 (1) (2021) 99–106

    2021

  44. [52]

    Ashton, C

    N. Ashton, C. Mockett, M. Fuchs, L. Fliessbach, H. Hetmann, T. Knacke, N. Schonwald, V . Skaperdas, G. Fotiadis, A. Walle, et al., Drivaerml: High-fidelity computational fluid dynamics dataset for road-car external aerodynamics, arXiv preprint arXiv:2408.11969 (2024)

    2024 arXiv

  45. [53]

    von Platen, S

    P. von Platen, S. Patil, A. Lozhkov, P. Cuenca, N. Lambert, K. Rasul, M. Davaadorj, D. Nair, S. Paul, W. Berman, Y . Xu, S. Liu, T. Wolf, Diffusers: State-of-the-art diffusion models, https://github.com/huggingface/diffusers (2022)

    2022

  46. [54]

    L. Lu, X. Meng, S. Cai, Z. Mao, S. Goswami, Z. Zhang, G. E. Karniadakis, A comprehensive and fair comparison of two neural operators (with practical extensions) based on fair data, Computer Methods in Applied Mechanics and Engineering 393 (2022) 114778. 33 Appendix A Data gene...

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.