Pith. sign in

REVIEW 4 major objections 4 minor 57 references

Point-wise Diffusion Models for Physical Systems with Shape Variations: Application to Spatio-temporal and Large-scale system

T0 review · 4 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 A point-wise diffusion model that runs forward and backward diffusion on each spatio-temporal point independently, using a point-wise diffusion transformer, achieves faster training, fewer parameters, and greater accuracy than image-based…

desk verdict A plausible point-wise diffusion formulation for shape-varying physics, with real potential, but the abstract alone cannot support the headline accuracy claims. read the letter →

arxiv 2508.01230 v1 pith:CG3LITRZ submitted 2025-08-02 physics.comp-ph cs.CV

Jiyong Kim , Sunwoong Yang , Namwoo Kang This is my paper
classification physics.comp-phcs.CV
keywords point-wisediffusionspatio-temporalpredictionphysicalsimulationshapevariationtransformerDDIMpointcloudsurrogatemodel
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 proposes a diffusion model that treats every spatio-temporal point of a physical field as an independent diffusion target, paired with a point-wise transformer that denoises each point without relying on image grids, meshes, or graph connectivity. The authors argue that this point-wise formulation removes the structural constraints of image-based diffusion models while preserving geometric fidelity for arbitrary mesh and point-cloud data. They report that this approach trains 94.4% faster, uses 89.0% fewer parameters, improves prediction accuracy by over 28%, and achieves 100 to 200 times faster inference with 5 to 10 DDIM steps compared to a 1000-step image-based baseline. The same model also beats DeepONet and Meshgraphnet consistently on 2D cylinder flow, OLED drop impact, and 3D road-car aerodynamics. If correct, this points toward a parameter-light, geometry-agnostic surrogate for real-time prediction in physical systems with shape variations.

What carries the argument

The key machinery is point-wise diffusion: each spatio-temporal point is independently corrupted by forward diffusion and then denoised by a point-wise diffusion transformer, without constructing image-like tensors or graph adjacency. This removes the need for structured grids and makes the model directly applicable to meshes and point clouds while preserving geometric fidelity. Deterministic DDIM sampling reduces the required denoising steps from 1000 to 5-10, which is the primary source of the reported 100-200x inference speedup.

What would settle it

Run the same point-wise diffusion model on a strongly coupled benchmark, for example a turbulent wake behind a bluff body or a two-phase flow with sharp interfaces, and compare against a graph-based or image-based surrogate that explicitly encodes local neighborhoods; if accuracy degrades or training time does not remain substantially lower, the claim that point-wise independence preserves necessary correlations fails.

Watch

Extended reading notes

Core claim

The central claim is that independent per-point forward and backward diffusion, with a point-wise diffusion transformer as the denoiser, is sufficient to model complex spatio-temporal physical systems with shape variations at higher accuracy and far lower computational cost than structured diffusion models. The paper demonstrates this on three systems: cylinder fluid flow, OLED drop impact, and 3D automotive aerodynamics. It shows that deterministic DDIM sampling with 5-10 steps reproduces accurate predictions, delivering 100-200x inference speedup over the standard 1000-step process. Compared to image-based diffusion, the point-wise model reduces training time by 94.4%, parameter count by 89.0%, and improves accuracy by more than 28%. Against data-flexible baselines DeepONet and Meshgraphnet, the point-wise model reports consistent superiority across all three test systems. The authors also investigate two design choices: predicting final physical states versus incremental changes, and the effect of subsampling ratios from 10% to 100% on computational efficiency.

Load-bearing premise

The load-bearing premise is that applying diffusion independently to each spatio-temporal point, without any explicit mesh connectivity, neighborhood structure, or geometric inductive bias, preserves the spatial and temporal correlations needed for accurate physics.

Editorial extensions

If this is right

  • If the point-wise model is correct, real-time prediction of physical fields with shape variations becomes practical because inference needs only 5-10 DDIM steps, a 100-200x reduction in sampling cost.
  • The 89% parameter reduction and 94.4% training-time reduction suggest that large image-based diffusion surrogates are over-parameterized for point-wise physical prediction, opening the way to much lighter surrogate models.
  • Direct applicability to meshes and point clouds means no geometry re-meshing or image-like discretization is required, simplifying the pipeline for engineering design tasks with changing shapes.
  • The consistent superiority over DeepONet and Meshgraphnet implies that point-wise diffusion could become a standard surrogate for spatio-temporal physics where data are irregular or include topological variation.

Reading between the lines

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

  • An implicit but untested extension is that point-wise independence will generalize to strongly coupled 3D systems such as turbulent wake flows or multi-phase problems; the paper's 3D example is external aerodynamics, which is smoother and may not stress local coupling as hard as turbulence.
  • The reported accuracy gains came from comparing to image-based diffusion and two baselines; a natural next test is against graph-based diffusion that explicitly encodes connectivity, which could reveal whether the point-wise model's advantage holds when correlations are highly nonlocal.
  • The final-state versus incremental-change investigation suggests that the model's residual learning on changes rather than absolute states is a design lever; future work could test whether incremental prediction further stabilizes long-horizon rollout in time-dependent systems.
  • Since DDIM with 5-10 steps works, one could push further to 1-2 steps or use consistency distillation, potentially bringing real-time control or digital twin applications within reach.
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 / 4 minor

Summary. The manuscript proposes a point-wise diffusion framework for predicting physical systems with shape variations. Rather than operating on image-like grids, the method applies forward and backward diffusion independently at each spatio-temporal point and uses a point-wise diffusion transformer for denoising. The authors claim substantial computational benefits over image-based diffusion models (94.4% training time reduction, 89.0% fewer parameters, 100–200× faster inference with 5–10 DDIM steps) and over 28% improvement in prediction accuracy, with consistent superiority over DeepONet and Meshgraphnet across three physical systems (2D cylinder flow, OLED drop impact, and 3D road-car aerodynamics). The abstract also mentions an investigation of final-state versus incremental-change prediction and of subsampling ratios from 10% to 100%.

Significance. If substantiated, the point-wise formulation offers a compelling alternative for physical prediction on unstructured meshes and point clouds, addressing a genuine limitation of image-based diffusion models. The specific quantitative claims are falsifiable, which is a strength, and the three diverse test cases give the work broad scope. However, because the abstract provides no equations, experimental protocols, or uncertainty measures, the reproducibility and validity of the headline numbers cannot currently be assessed. The central mechanistic premise, that independent per-point diffusion preserves the spatial and temporal correlations needed for accurate physics, remains untested; this is the key risk to the paper's contribution.

major comments (4)
  1. [Abstract, third sentence] The load-bearing claim is that applying forward and backward diffusion independently at each spatio-temporal point preserves the spatial and temporal correlations needed for physical accuracy, yet the abstract provides no theoretical justification, ablation, or diagnostic evidence for this premise. If per-point independence discards small-scale coherence, the reported accuracy gains might reflect benchmark choices rather than the method's intrinsic merit. Please add an analysis that directly tests correlation recovery, for example by comparing against a variant that injects explicit local neighborhood features, or by reporting error as a function of point density and neighborhood radius.
  2. [Abstract, quantitative results] The headline numbers ('over 28% improvement', '94.4% training time reduction', '89.0% fewer parameters') are reported without any uncertainty measures, number of random seeds, or description of train/test splits. For such precise claims, the absence of variance and protocol details prevents verification and leaves open the possibility that the results are specific to a single favorable run or split. Please report mean ± standard deviation over multiple independent runs and specify the exact evaluation protocol, including data partitioning and metric definitions.
  3. [Abstract, last sentence] The abstract states that subsampling ratios from 10% to 100% are investigated, but it does not report how accuracy or inference speed depend on the subsampling ratio. If accuracy degrades substantially at low subsampling, the 'large-scale 3D' claim is restricted to dense point sets and the computational speedup from subsampling would be less meaningful. Please include the outcome of this investigation and discuss its implications for the scalability claim.
  4. [Abstract, comparisons to baselines] The claims of 'consistent superiority' over DeepONet and Meshgraphnet and 'over 28% improvement' over image-based diffusion are made without information about baseline model sizes, training budgets, hyperparameter tuning, or the exact metrics used. Without this context, the fairness of the comparisons cannot be judged, and the reported improvements could be an artifact of undertrained baselines. Please provide a detailed experimental setup for each baseline and report per-system results with the chosen accuracy metric (e.g., normalized RMSE, relative L2 error).
minor comments (4)
  1. [Abstract, first sentence] The term 'point-wise diffusion transformer' is introduced without a description; please provide architectural details or a citation to a related architecture so the reader can understand what is new.
  2. [Abstract, accuracy claims] The abstract does not define the accuracy metric used to compute the reported improvements; please specify whether it is RMSE, relative L2 error, or another measure.
  3. [Abstract, inference speedup] The '100 to 200 times faster inference' claim should clarify the exact conditions, such as the hardware, batch size, and whether the comparison uses the same number of evaluation samples for both methods.
  4. [Abstract, final-state vs incremental prediction] The abstract says the paper compares final physical state prediction with incremental change prediction, but it does not state which approach performed better; please include the outcome in the abstract or at least in the conclusions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity identified from the abstract; performance claims rest on external baselines.

full rationale

The available text (abstract only) contains no equations, no fitted-parameter-then-predicted quantity, and no self-citation chain that would make a claimed result equivalent to its inputs by construction. The central claims are empirical comparisons against external baselines (image-based diffusion, DeepONet, Meshgraphnet) on three physical systems. While the point-wise independence premise is unvalidated and the train/test split cannot be audited from the abstract, these are standard empirical-risk concerns, not circularity. There is no quoted reduction by which the output is defined from the input or by which a fitted parameter is renamed as a prediction. Per the hard rules, circularity is only flagged with quoted evidence of a specific reduction; none exists in the supplied text. Therefore the honest finding is no significant circularity (score 0).

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claims rest on the training of a learned denoising network on task-specific data, plus hand-chosen DDIM step counts and subsampling ratios. The abstract introduces an independence assumption for per-point diffusion without stated limitations, and treats three baselines as fair comparisons without reporting tuning protocols. No new physical entities are introduced.

free parameters (3)
  • learned denoising network parameters = not disclosed in abstract
    The central performance claim depends on training the point-wise diffusion transformer to fit the training data; only the relative parameter reduction (89.0%) is reported, not the absolute counts.
  • DDIM inference step count = 5 to 10
    The speedup claim depends on this user-chosen number of sampling steps; it is not derived from the data or from a convergence guarantee.
  • subsampling ratio = 10% to 100%
    The abstract evaluates computational efficiency across subsampling ratios, implying results vary with this hand-chosen parameter.
assumptions (4)
  • ad hoc to paper Spatio-temporal fields can be modeled by independent per-point diffusion processes without explicit neighborhood or connectivity information.
    This is introduced in the abstract as the core methodological premise; no derivation or ablation is available in the abstract to support it.
  • domain assumption Existing image-based diffusion models, DeepONet, and Meshgraphnet are representative and fairly tuned baselines.
    The superiority claim is relative to these baselines; the abstract does not report whether hyperparameters were optimized equally for all methods.
  • domain assumption DDIM sampling with 5 to 10 steps preserves accuracy relative to 1000-step noising and denoising.
    The speedup claim assumes DDIM's deterministic sampling does not degrade accuracy; this is an empirical property that the abstract asserts but does not demonstrate with error analysis.
  • domain assumption Training and test data for the three physical domains are representative and correctly split.
    No dataset sizes, splits, or preprocessing details are given; without this, the reported accuracy gains cannot be interpreted.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Point-wise Diffusion Models for Physical Systems with Shape Variations: Application to Spatio-temporal and Large-scale system." pith.science (2026). https://pith.science/paper/CG3LITRZ

@misc{pith2026250801230,
  author       = {Pith},
  title        = {Pith review of: Point-wise Diffusion Models for Physical Systems with Shape Variations: Application to Spatio-temporal and Large-scale system},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CG3LITRZ}},
  note         = {Machine review of arXiv:2508.01230}
}
read the original abstract

This study introduces a novel point-wise diffusion model that processes spatio-temporal points independently to efficiently predict complex physical systems with shape variations. This methodological contribution lies in applying forward and backward diffusion processes at individual spatio-temporal points, coupled with a point-wise diffusion transformer architecture for denoising. Unlike conventional image-based diffusion models that operate on structured data representations, this framework enables direct processing of any data formats including meshes and point clouds while preserving geometric fidelity. We validate our approach across three distinct physical domains with complex geometric configurations: 2D spatio-temporal systems including cylinder fluid flow and OLED drop impact test, and 3D large-scale system for road-car external aerodynamics. To justify the necessity of our point-wise approach for real-time prediction applications, we employ denoising diffusion implicit models (DDIM) for efficient deterministic sampling, requiring only 5-10 steps compared to traditional 1000-step and providing computational speedup of 100 to 200 times during inference without compromising accuracy. In addition, our proposed model achieves superior performance compared to image-based diffusion model: reducing training time by 94.4% and requiring 89.0% fewer parameters while achieving over 28% improvement in prediction accuracy. Comprehensive comparisons against data-flexible surrogate models including DeepONet and Meshgraphnet demonstrate consistent superiority of our approach across all three physical systems. To further refine the proposed model, we investigate two key aspects: 1) comparison of final physical states prediction or incremental change prediction, and 2) computational efficiency evaluation across varying subsampling ratios (10%-100%).

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 37 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  3. [3]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  4. [4]

    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

  5. [5]

    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)

    work page arXiv 2024

  6. [6]

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

    work page Pith review arXiv 2025

  7. [7]

    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

  8. [8]

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

    work page 2025

Show all 57 references
  1. [9]

    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

  2. [10]

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

    2024 arXiv

  3. [11]

    Jadhav, J

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

    2023

  4. [12]

    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

  5. [13]

    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

  6. [14]

    Cheng, L

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

    2025

  7. [15]

    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

  8. [16]

    o wer, J. Lottes, S. Rasp, P. D \

    D. Kochkov, J. Yuval, I. Langmore, P. Norgaard, J. Smith, G. Mooers, M. Kl \"o wer, J. Lottes, S. Rasp, P. D \"u ben, et al., Neural general circulation models for weather and climate, Nature 632 (8027) (2024) 1060--1066

    2024

  9. [17]

    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

  10. [18]

    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

    2024

  11. [19]

    R \"u hling Cachay, B

    S. R \"u hling 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

  12. [20]

    Kohl, L.-W

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

    2023 arXiv

  13. [21]

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

    2023

  14. [22]

    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

  15. [23]

    Pfaff, M

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

    2020

  16. [24]

    Fortunato, T

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

    2022 arXiv

  17. [25]

    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

  18. [26]

    Y. Cao, M. Chai, M. Li, C. Jiang, Efficient 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

  19. [27]

    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

  20. [28]

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

    2022 arXiv

  21. [29]

    Alkin, A

    B. Alkin, A. F \"u rst, 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

  22. [30]

    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

  23. [31]

    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

  24. [32]

    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

  25. [33]

    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

  26. [34]

    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

  27. [35]

    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

  28. [36]

    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

  29. [37]

    Oommen, A

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

    2025

  30. [38]

    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

  31. [39]

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

    2013 arXiv

  32. [40]

    Sohl-Dickstein, E

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

    2015

  33. [41]

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

    2020

  34. [42]

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

    2020 arXiv

  35. [43]

    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

  36. [44]

    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

  37. [45]

    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

  38. [46]

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

    2023

  39. [47]

    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

  40. [48]

    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

  41. [49]

    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

  42. [50]

    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

  43. [51]

    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

  44. [52]

    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

    2023

  45. [53]

    Peebles, S

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

    2023

  46. [54]

    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

  47. [55]

    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

  48. [56]

    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

  49. [57]

    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

    2022

Pith tools

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