pith. sign in

arxiv: 2606.22752 · v1 · pith:2RQIPPH2new · submitted 2026-06-22 · 💻 cs.LG · cs.CE· physics.comp-ph

One-Step Flow Matching for Generative Modeling of Path-Dependent Physical Fields

Yijing Zhou , Jasmin Jelovica This is my paper

Pith reviewed 2026-06-26 09:37 UTC · model grok-4.3

classification 💻 cs.LG cs.CEphysics.comp-ph
keywords flow matchinggenerative modelingpath-dependent fieldsstress fieldsvariational autoencodertransformerplasticityfinite element analysis
0
0 comments X

The pith

A transformer-based flow matching model generates accurate high-resolution path-dependent stress fields in one step from limited data.

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

The paper proposes a flow matching model using a transformer backbone inside the latent space of a variational autoencoder to generate path-dependent physical fields such as plastic stress under stochastic loading-unloading paths and varying geometries. It treats the simulation as a video synthesis task and introduces a non-Gaussian source distribution so that the model can produce usable samples in a single step without distillation or repeated sampling. Token-level loading embeddings plus two auxiliary networks further improve fidelity for path-dependent constitutive behavior. A sympathetic reader would care because traditional finite-element methods become prohibitively slow for intricate geometries, while this approach claims to retain accuracy yet deliver 6-7 times speedup on CPUs and roughly two orders of magnitude on consumer GPUs even with modest training sets.

Core claim

The authors develop a flow matching model based on a transformer backbone that operates in the latent space of a variational autoencoder, formulating plastic field simulation as a video synthesis task. By designing a non-Gaussian source distribution, the model reduces crossings among conditional transport paths, enabling satisfactory one-step samples. Token-level loading embeddings and two auxiliary networks enhance performance for path-dependent constitutive behavior with stochastic paths and geometry.

What carries the argument

Flow matching with non-Gaussian source distribution inside VAE latent space, using a transformer backbone plus token-level loading embeddings and auxiliary networks

If this is right

  • Accurate generation of high-resolution path-dependent stress fields is possible even with a limited training dataset.
  • The model produces full time sequences of stress fields directly, without requiring hundreds of sampling steps.
  • Computational cost is reduced by a factor of 6 to 7 relative to finite-element analysis on CPUs.
  • Speedup reaches approximately two orders of magnitude on consumer-grade GPUs while preserving accuracy for path-dependent behavior.

Where Pith is reading between the lines

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

  • The one-step property could enable interactive design loops where engineers query stress fields for many candidate geometries in real time.
  • The same latent-space video formulation might transfer to other history-dependent physical quantities such as damage evolution or thermal history.
  • Because the method avoids iterative sampling, it could be paired with uncertainty quantification techniques that require many independent realizations.

Load-bearing premise

The non-Gaussian source distribution combined with token-level loading embeddings and auxiliary networks is sufficient to produce accurate one-step samples for path-dependent constitutive behavior without post-hoc tuning or multiple sampling steps.

What would settle it

Quantitative comparison of one-step generated stress fields against independent finite-element solutions on a held-out collection of unseen stochastic loading paths and geometries, using field-wise error norms such as mean absolute stress deviation.

Figures

Figures reproduced from arXiv: 2606.22752 by Jasmin Jelovica, Yijing Zhou.

Figure 1
Figure 1. Figure 1: Overall schematic. (a) Spatiotemporal DiT backbone for learning the velocity field in the sample (latent) space; (b) Scientific [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Spatiotemporal DiT backbone. After being compressed into the latent space, each frame is patchified, and each patch is projected [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Boundary conditions and loading. The right side boundary is displaced in the direction that depends on the sign of the current [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance of one-step generation for the single hole case. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance of one-step generation for the 3 holes case. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Performance of one-step generation for the 6 holes case. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
read the original abstract

Physical simulations for intricate geometries with path-dependent constitutive models face difficulties due to the enormous computational cost they require. Recently, the emergence of generative AI models, which succeed in image and video synthesis tasks, has provided a promise to further improve simulations. Although U-Net-based denoising diffusion probabilistic models (DDPMs) have been adopted for elastic stress field generation, they typically require hundreds of sampling steps, and applications of generative models to path-dependent, e.g. plastic, stress fields remain very limited. In this work, we propose a novel flow matching (FM) model based on a transformer backbone for high-resolution path-dependent stress field generation with stochastic loading-unloading paths and geometry. The proposed model operates within the latent space of a variational autoencoder (VAE) and formulates the simulation of plastic fields as a video synthesis task, directly generating the stress fields across all time steps. Meanwhile, we design a non-Gaussian source distribution for flow matching, such that crossings among conditional transport paths are reduced during training. This enables our model to generate satisfactory samples in one step without relying on distillation. In addition, we introduce token-level loading embeddings and two auxiliary networks to further enhance the model performance in path-dependent simulation. The results demonstrate that, even with a limited training dataset, our model can accurately generate high-resolution path-dependent fields. It is much more computationally efficient than finite element analysis, providing a speedup of 6 to 7 times over FEM on CPUs and approximately two orders of magnitude speedup on consumer-grade GPUs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript proposes a flow matching model with a transformer backbone operating in the latent space of a variational autoencoder (VAE) to generate high-resolution path-dependent (e.g., plastic) stress fields under stochastic loading-unloading paths and varying geometries. It formulates the task as video synthesis across time steps, introduces a non-Gaussian source distribution to reduce crossings in conditional transport paths for one-step sampling without distillation, and adds token-level loading embeddings plus two auxiliary networks. The work claims that the resulting model produces accurate outputs even with limited training data and delivers speedups of 6-7x over FEM on CPUs and roughly two orders of magnitude on consumer GPUs.

Significance. If the central claims are substantiated, the approach could meaningfully accelerate high-fidelity simulations of path-dependent constitutive behavior, offering a practical alternative to expensive finite-element computations for engineering geometries. The framing of time-dependent plastic fields as a generative video task and the attempt at one-step flow matching without post-training distillation represent technically interesting directions at the intersection of generative modeling and computational mechanics.

major comments (2)
  1. [Abstract (model design paragraph)] Abstract (paragraph on model design): The claim that the non-Gaussian source distribution reduces crossings among conditional transport paths sufficiently to enable accurate one-step samples rests on an unverified mechanism; no quantitative measure of crossing reduction, no ablation against a Gaussian baseline, and no comparison of one-step versus multi-step sampling error on the trained model are supplied to support that this design choice is load-bearing for the reported fidelity.
  2. [Abstract (results paragraph)] Abstract (results paragraph): The assertions of accurate high-resolution generation with a limited dataset and concrete speedups (6-7x on CPUs, ~100x on GPUs) versus FEM are presented without accompanying error metrics (e.g., relative L2 norms, plastic strain error), dataset cardinality, geometry resolution, hardware specifications, or timing breakdowns that would allow independent verification of the performance claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment point by point below and indicate where revisions to the manuscript will be made.

read point-by-point responses

  1. Referee: [Abstract (model design paragraph)] Abstract (paragraph on model design): The claim that the non-Gaussian source distribution reduces crossings among conditional transport paths sufficiently to enable accurate one-step samples rests on an unverified mechanism; no quantitative measure of crossing reduction, no ablation against a Gaussian baseline, and no comparison of one-step versus multi-step sampling error on the trained model are supplied to support that this design choice is load-bearing for the reported fidelity.

    Authors: We agree that the abstract presents the non-Gaussian source as enabling one-step sampling without providing quantitative support for the claimed reduction in path crossings. The manuscript motivates the choice in the methods and shows qualitative path visualizations, but does not include the requested ablations or metrics. In the revised manuscript we will add an ablation study (Gaussian vs. non-Gaussian source) with quantitative measures of crossings and one-step versus multi-step error, and we will update the abstract to reference these results. revision: yes

  2. Referee: [Abstract (results paragraph)] Abstract (results paragraph): The assertions of accurate high-resolution generation with a limited dataset and concrete speedups (6-7x on CPUs, ~100x on GPUs) versus FEM are presented without accompanying error metrics (e.g., relative L2 norms, plastic strain error), dataset cardinality, geometry resolution, hardware specifications, or timing breakdowns that would allow independent verification of the performance claims.

    Authors: We acknowledge that the abstract states the accuracy and speedup claims without the supporting numerical details requested. The experimental section of the manuscript reports relative L2 errors, dataset sizes, resolutions, and timing results, but these are not summarized in the abstract. We will revise the abstract to include the key quantitative values (error norms, dataset cardinality, resolution, and hardware/timing specifications) so that the performance claims can be verified directly from the abstract. revision: yes

Circularity Check

0 steps flagged

No significant circularity; design choices and empirical results are independent

full rationale

The paper introduces a flow-matching model with a transformer backbone, VAE latent space, non-Gaussian source, token-level embeddings, and auxiliary networks as explicit design decisions to enable one-step generation of path-dependent fields. These choices are motivated by reducing path crossings during training but are not shown to be equivalent to the target performance metrics by construction; results are reported from training on a limited dataset and direct comparison to FEM runtimes. No equations, self-citations, or fitted parameters are described that rename or force the claimed accuracy or speedups. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no explicit free parameters, axioms, or invented entities are detailed beyond high-level design choices such as the non-Gaussian source distribution.

pith-pipeline@v0.9.1-grok · 5809 in / 1181 out tokens · 23560 ms · 2026-06-26T09:37:26.138575+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

79 extracted references · 50 canonical work pages · 2 internal anchors

  1. [1]

    Bishara, Y

    D. Bishara, Y . Xie, W. K. Liu, S. Li, A state-of-the-art review on machine learning-based multiscale modeling, simulation, homogenization and design of materials, Archives of computational methods in engineering 30 (1) (2023) 191–222.doi:https://doi.org/10.1007/s11831-022-09795-8

    work page doi:10.1007/s11831-022-09795-8 2023

  2. [2]

    F. E. Bock, R. C. Aydin, C. J. Cyron, N. Huber, S. R. Kalidindi, B. Klusemann, A review of the application of machine learning and data mining approaches in continuum materials mechanics, Frontiers in Materials 6 (2019) 110.doi:https://doi.org/10.3389/fmats.2019.00110

    work page doi:10.3389/fmats.2019.00110 2019

  3. [3]

    X. Liu, S. Tian, F. Tao, W. Yu, A review of artificial neural networks in the constitutive modeling of composite ma- terials, Composites Part B: Engineering 224 (2021) 109152.doi:https://doi.org/10.1016/j.compositesb. 2021.109152

    work page doi:10.1016/j.compositesb 2021

  4. [4]

    J. Fish, G. J. Wagner, S. Keten, Mesoscopic and multiscale modelling in materials, Nature materials 20 (6) (2021) 774–786.doi:https://doi.org/10.1038/s41563-020-00913-0

    work page doi:10.1038/s41563-020-00913-0 2021

  5. [5]

    W. K. Liu, S. Li, H. S. Park, Eighty years of the finite element method: Birth, evolution, and future, Archives of Computational Methods in Engineering 29 (6) (2022) 4431–4453.doi:https://doi.org/10.1007/ s11831-022-09740-9

    2022

  6. [6]

    Mozaffar, R

    M. Mozaffar, R. Bostanabad, W. Chen, K. Ehmann, J. Cao, M. Bessa, Deep learning predicts path-dependent plastic- ity, Proceedings of the National Academy of Sciences 116 (52) (2019) 26414–26420.doi:https://doi.org/10. 1073/pnas.1911815116

    2019

  7. [7]

    Bonatti, D

    C. Bonatti, D. Mohr, On the importance of self-consistency in recurrent neural network models representing elasto- plastic solids, Journal of the Mechanics and Physics of Solids 158 (2022) 104697.doi:https://doi.org/10. 1016/j.jmps.2021.104697

    arXiv 2022

  8. [8]

    Zhang, S.-L

    N. Zhang, S.-L. Shen, A. Zhou, Y .-F. Jin, Application of lstm approach for modelling stress–strain behaviour of soil, Applied Soft Computing 100 (2021) 106959.doi:https://doi.org/10.1016/j.asoc.2020.106959

    work page doi:10.1016/j.asoc.2020.106959 2021

  9. [9]

    S. Im, J. Lee, M. Cho, Surrogate modeling of elasto-plastic problems via long short-term memory neural networks and proper orthogonal decomposition, Computer Methods in Applied Mechanics and Engineering 385 (2021) 114030. doi:https://doi.org/10.1016/j.cma.2021.114030

    work page doi:10.1016/j.cma.2021.114030 2021

  10. [10]

    J. N. Heidenreich, D. Mohr, Recurrent neural network plasticity models: Unveiling their common core through multi-task learning, Computer Methods in Applied Mechanics and Engineering 426 (2024) 116991.doi:https: //doi.org/10.1016/j.cma.2024.116991

    work page doi:10.1016/j.cma.2024.116991 2024

  11. [11]

    S. Guan, X. Zhang, S. Ranftl, T. Qu, A neural network-based material cell for elastoplasticity and its performance in fe analyses of boundary value problems, International Journal of Plasticity 171 (2023) 103811.doi:https: //doi.org/10.1016/j.ijplas.2023.103811

    work page doi:10.1016/j.ijplas.2023.103811 2023

  12. [12]

    Tancogne-Dejean, M

    T. Tancogne-Dejean, M. B. Gorji, J. Zhu, D. Mohr, Recurrent neural network modeling of the large deformation of lithium-ion battery cells, International Journal of Plasticity 146 (2021) 103072.doi:https://doi.org/10.1016/ j.ijplas.2021.103072

    arXiv 2021

  13. [13]

    W. G. Dettmer, E. J. Muttio, R. Alhayki, D. Peri ´c, A framework for neural network based constitutive modelling of inelastic materials, Computer Methods in Applied Mechanics and Engineering 420 (2024) 116672.doi:https: //doi.org/10.1016/j.cma.2023.116672

    work page doi:10.1016/j.cma.2023.116672 2024

  14. [14]

    Chen, Recurrent neural networks (rnns) learn the constitutive law of viscoelasticity, Computational Mechanics 67 (3) (2021) 1009–1019.doi:https://doi.org/10.1007/s00466-021-01981-y

    G. Chen, Recurrent neural networks (rnns) learn the constitutive law of viscoelasticity, Computational Mechanics 67 (3) (2021) 1009–1019.doi:https://doi.org/10.1007/s00466-021-01981-y

    work page doi:10.1007/s00466-021-01981-y 2021

  15. [15]

    Y . Zhou, S. J. Semnani, A machine learning based multi-scale finite element framework for nonlinear com- posite materials, Engineering with Computers 41 (5) (2025) 2795–2831.doi:https://doi.org/10.1007/ s00366-025-02121-3

    2025

  16. [16]

    L. Wu, V . D. Nguyen, N. G. Kilingar, L. Noels, A recurrent neural network-accelerated multi-scale model for elasto- plastic heterogeneous materials subjected to random cyclic and non-proportional loading paths, Computer Meth- ods in Applied Mechanics and Engineering 369 (2020) 113234.doi:https://doi.org/10.1016/j.cma.2020. 113234

    work page doi:10.1016/j.cma.2020 2020

  17. [17]

    Ghavamian, A

    F. Ghavamian, A. Simone, Accelerating multiscale finite element simulations of history-dependent materials using a recurrent neural network, Computer Methods in Applied Mechanics and Engineering 357 (2019) 112594.doi: https://doi.org/10.1016/j.cma.2019.112594

    work page doi:10.1016/j.cma.2019.112594 2019

  18. [18]

    H. J. Logarzo, G. Capuano, J. J. Rimoli, Smart constitutive laws: Inelastic homogenization through machine learning, Computer methods in applied mechanics and engineering 373 (2021) 113482.doi:https://doi.org/10.1016/ j.cma.2020.113482

    arXiv 2021

  19. [19]

    K. Wang, W. Sun, A multiscale multi-permeability poroplasticity model linked by recursive homogenizations and deep learning, Computer Methods in Applied Mechanics and Engineering 334 (2018) 337–380.doi:https:// doi.org/10.1016/j.cma.2018.01.036

    work page doi:10.1016/j.cma.2018.01.036 2018

  20. [20]

    E. M. Haghighi, S. Na, A single long short-term memory network for enhancing the prediction of path-dependent plasticity with material heterogeneity and anisotropy, arXiv preprint arXiv:2204.01466 (2022).doi:https://doi. org/10.48550/arXiv.2204.01466

    work page doi:10.48550/arxiv.2204.01466 2022

  21. [21]

    J. M. Calleja Vazquez, V . D. Nguyen, L. Wu, S. M. Mustafa, L. Noels, Development of history-dependent surro- gate models in the context of stochastic multi-scale simulations for elasto-plastic composites, in: 5th International 22 Conference on Uncertainty Quantification in Computational Science and Engineering (UNCECOMP2023), 2023. doi:https://doi.org/10...

    work page doi:10.24435/materials.20230710 2023

  22. [22]

    L. Wu, L. Noels, Self-consistency reinforced minimal gated recurrent unit for surrogate modeling of history- dependent non-linear problems: Application to history-dependent homogenized response of heterogeneous materials, Computer Methods in Applied Mechanics and Engineering 424 (2024) 116881.doi:https://doi.org/10.1016/ j.cma.2024.116881

    arXiv 2024

  23. [23]

    H. Kim, I. Jeong, H. Cho, M. Cho, Surrogate model based on data-driven model reduction for inelastic behavior of composite microstructure, International Journal of Aeronautical and Space Sciences 24 (3) (2023) 732–752.doi: https://doi.org/10.1007/s42405-022-00547-3

    work page doi:10.1007/s42405-022-00547-3 2023

  24. [24]

    Vijayaraghavan, L

    S. Vijayaraghavan, L. Wu, L. Noels, S. Bordas, S. Natarajan, L. A. Beex, A data-driven reduced-order surrogate model for entire elastoplastic simulations applied to representative volume elements, Scientific Reports 13 (1) (2023) 12781.doi:https://doi.org/10.1038/s41598-023-38104-x

    work page doi:10.1038/s41598-023-38104-x 2023

  25. [25]

    M. B. Gorji, M. Mozaffar, J. N. Heidenreich, J. Cao, D. Mohr, On the potential of recurrent neural networks for modeling path dependent plasticity, Journal of the Mechanics and Physics of Solids 143 (2020) 103972.doi: https://doi.org/10.1016/j.jmps.2020.103972

    work page doi:10.1016/j.jmps.2020.103972 2020

  26. [26]

    Friemann, B

    J. Friemann, B. Dashtbozorg, M. Fagerström, S. Mirkhalaf, A micromechanics-based recurrent neural networks model for path-dependent cyclic deformation of short fiber composites, International Journal for Numerical Methods in Engineering 124 (10) (2023) 2292–2314.doi:https://doi.org/10.1002/nme.7211

    work page doi:10.1002/nme.7211 2023

  27. [27]

    S. B. Tandale, F. Bamer, B. Markert, M. Stoffel, Physics-based self-learning recurrent neural network enhanced time integration scheme for computing viscoplastic structural finite element response, Computer Methods in Applied Mechanics and Engineering 401 (2022) 115668.doi:https://doi.org/10.1016/j.cma.2022.115668

    work page doi:10.1016/j.cma.2022.115668 2022

  28. [28]

    M. Maia, I. Rocha, P. Kerfriden, F. van der Meer, Physically recurrent neural networks for path-dependent het- erogeneous materials: Embedding constitutive models in a data-driven surrogate, Computer Methods in Applied Mechanics and Engineering 407 (2023) 115934.doi:https://doi.org/10.1016/j.cma.2023.115934

    work page doi:10.1016/j.cma.2023.115934 2023

  29. [29]

    S. Deng, S. Hosseinmardi, L. Wang, D. Apelian, R. Bostanabad, Data-driven physics-constrained recurrent neural networks for multiscale damage modeling of metallic alloys with process-induced porosity, Computational Mechan- ics (2024) 1–31doi:https://doi.org/10.1007/s00466-023-02429-1

    work page doi:10.1007/s00466-023-02429-1 2024

  30. [30]

    Borkowski, C

    L. Borkowski, C. Sorini, A. Chattopadhyay, Recurrent neural network-based multiaxial plasticity model with regu- larization for physics-informed constraints, Computers & Structures 258 (2022) 106678.doi:https://doi.org/ 10.1016/j.compstruc.2021.106678

    work page doi:10.1016/j.compstruc.2021.106678 2022

  31. [31]

    T. Qu, S. Di, Y . Feng, M. Wang, T. Zhao, Towards data-driven constitutive modelling for granular materials via micromechanics-informed deep learning, International Journal of Plasticity 144 (2021) 103046.doi:https:// doi.org/10.1016/j.ijplas.2021.103046

    work page doi:10.1016/j.ijplas.2021.103046 2021

  32. [32]

    M. Maia, I. B. Rocha, D. Kovaˇcevi´c, F. van der Meer, Physically recurrent neural network for rate and path-dependent heterogeneous materials in a finite strain framework, arXiv preprint arXiv:2404.17583 (2024).doi:https://doi. org/10.48550/arXiv.2404.17583

    work page doi:10.48550/arxiv.2404.17583 2024

  33. [33]

    J. Fish, Y . Yu, Data-physics driven reduced order homogenization, International Journal for Numerical Methods in Engineering 124 (7) (2023) 1620–1645.doi:https://doi.org/10.1002/nme.7178

    work page doi:10.1002/nme.7178 2023

  34. [34]

    J. N. Heidenreich, C. Bonatti, D. Mohr, Transfer learning of recurrent neural network-based plasticity models, Inter- national Journal for Numerical Methods in Engineering 125 (1) (2024) e7357.doi:https://doi.org/10.1002/ nme.7357

    2024

  35. [35]

    T. Qu, J. Zhao, S. Guan, Y . Feng, Data-driven multiscale modelling of granular materials via knowledge transfer and sharing, International Journal of Plasticity 171 (2023) 103786.doi:https://doi.org/10.1016/j.ijplas. 2023.103786

    work page doi:10.1016/j.ijplas 2023

  36. [36]

    S. B. Tandale, M. Stoffel, Recurrent and convolutional neural networks in structural dynamics: a modified atten- tion steered encoder–decoder architecture versus lstm versus gru versus tcn topologies to predict the response of shock wave-loaded plates, Computational Mechanics 72 (4) (2023) 765–786.doi:https://doi.org/10.1007/ s00466-023-02317-8

    2023

  37. [37]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, I. Polosukhin, Attention is all you need, Advances in neural information processing systems 30 (2017). URLhttps://proceedings.neurips.cc/paper_files/paper/2017/file/ 3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf

    2017

  38. [38]

    Y . Zhou, S. J. Semnani, Vit-transformer: Self-attention mechanism based constitutive modeling for nonlinear het- erogeneous materials, arXiv preprint arXiv:2510.16575 (2025).doi:https://doi.org/10.48550/arXiv.2510. 16575

    work page doi:10.48550/arxiv.2510 2025

  39. [39]

    Zhongbo, P

    Y . Zhongbo, P. L. Hien, Pre-trained transformer model as a surrogate in multiscale computational homogenization framework for elastoplastic composite materials subjected to generic loading paths, Computer Methods in Applied Mechanics and Engineering 421 (2024) 116745.doi:https://doi.org/10.1016/j.cma.2024.116745

    work page doi:10.1016/j.cma.2024.116745 2024

  40. [41]

    E. L. Buehler, M. J. Buehler, End-to-end prediction of multimaterial stress fields and fracture patterns using cycle- 23 consistent adversarial and transformer neural networks, Biomedical Engineering Advances 4 (2022) 100038.doi: https://doi.org/10.1016/j.bea.2022.100038

    work page doi:10.1016/j.bea.2022.100038 2022

  41. [42]

    Z. Yang, M. J. Buehler, Words to matter: De novo architected materials design using transformer neural networks, Frontiers in Materials 8 (2021) 740754.doi:https://doi.org/10.3389/fmats.2021.740754

    work page doi:10.3389/fmats.2021.740754 2021

  42. [44]

    Gupta, A

    A. Gupta, A. Bhaduri, L. Graham-Brady, Accelerated multiscale mechanics modeling in a deep learning framework, Mechanics of Materials (2023) 104709doi:https://doi.org/10.1016/j.mechmat.2023.104709

    work page doi:10.1016/j.mechmat.2023.104709 2023

  43. [45]

    I. Saha, A. Gupta, L. Graham-Brady, Prediction of local elasto-plastic stress and strain fields in a two-phase composite microstructure using a deep convolutional neural network, Computer Methods in Applied Mechanics and Engineering 421 (2024) 116816.doi:https://doi.org/10.1016/j.cma.2024.116816

    work page doi:10.1016/j.cma.2024.116816 2024

  44. [46]

    M. S. Khorrami, J. R. Mianroodi, N. H. Siboni, P. Goyal, B. Svendsen, P. Benner, D. Raabe, An artificial neural network for surrogate modeling of stress fields in viscoplastic polycrystalline materials, npj Computational Materials 9 (1) (2023) 37.doi:https://doi.org/10.1038/s41524-023-00991-z

    work page doi:10.1038/s41524-023-00991-z 2023

  45. [47]

    C. Yang, Y . Kim, S. Ryu, G. X. Gu, Using convolutional neural networks to predict composite properties beyond the elastic limit, MRS Communications 9 (2) (2019) 609–617.doi:https://doi.org/10.1557/mrc.2019.49

    work page doi:10.1557/mrc.2019.49 2019

  46. [48]

    H. Su, T. Guan, Y . Liu, A three-dimensional prediction method of stiffness properties of composites based on deep learning, Computational Mechanics 71 (3) (2023) 583–597.doi:https://doi.org/10.1007/ s00466-022-02253-z

    2023

  47. [49]

    Chang, J.-H

    H.-S. Chang, J.-H. Huang, J.-L. Tsai, Predicting mechanical properties of unidirectional composites using ma- chine learning, Multiscale Science and Engineering 4 (4) (2022) 202–210.doi:https://doi.org/10.1007/ s42493-022-00087-8

    2022

  48. [50]

    M. A. Mahdi, C. Crick, W. Zhao, Lattice structure design using machine learning and homogenization approach, in: ASME Aerospace Structures, Structural Dynamics, and Materials Conference, V ol. 87745, American Society of Mechanical Engineers, 2024, p. V001T01A015.doi:https://doi.org/10.1115/SSDM2024-121612

    work page doi:10.1115/ssdm2024-121612 2024

  49. [51]

    J. R. Mianroodi, S. Rezaei, N. H. Siboni, B.-X. Xu, D. Raabe, Lossless multi-scale constitutive elastic relations with artificial intelligence, npj Computational Materials 8 (1) (2022) 67.doi:https://doi.org/10.1038/ s41524-022-00753-3

    2022

  50. [52]

    C. Yang, Y . Kim, S. Ryu, G. X. Gu, Prediction of composite microstructure stress-strain curves using convolutional neural networks, Materials & Design 189 (2020) 108509.doi:https://doi.org/10.1016/j.matdes.2020. 108509

    work page doi:10.1016/j.matdes.2020 2020

  51. [53]

    C. Rao, Y . Liu, Three-dimensional convolutional neural network (3d-cnn) for heterogeneous material homogeniza- tion, Computational Materials Science 184 (2020) 109850.doi:https://doi.org/10.1016/j.commatsci. 2020.109850

    work page doi:10.1016/j.commatsci 2020

  52. [54]

    H. Peng, A. Liu, J. Huang, L. Cao, J. Liu, L. Lu, Ph-net: Parallelepiped microstructure homogenization via 3d convolutional neural networks, Additive Manufacturing 60 (2022) 103237.doi:https://doi.org/10.1016/j. addma.2022.103237

    work page doi:10.1016/j 2022

  53. [55]

    S. Yan, J. Jelovica, Kato: Neural-reparameterized topology optimization using convolutional kolmogorov-arnold network for stress minimization, International Journal for Numerical Methods in Engineering 126 (8) (2025) e70034. doi:https://doi.org/10.1002/nme.70034

    work page doi:10.1002/nme.70034 2025

  54. [56]

    S. Yan, J. Jelovica, Katosuper: Surrogate-accelerated neural topology optimization with sensitivity-consistent fourier neural operators, Available at SSRN 6184689 (2026)

    2026

  55. [57]

    Gulakala, B

    R. Gulakala, B. Markert, M. Stoffel, Graph neural network enhanced finite element modelling, PAMM 22 (1) (2023) e202200306.doi:https://doi.org/10.1002/pamm.202200306

    work page doi:10.1002/pamm.202200306 2023

  56. [58]

    Jiang, N.-Z

    C. Jiang, N.-Z. Chen, Graph neural networks (gnns) based accelerated numerical simulation, Engineering Applica- tions of Artificial Intelligence 123 (2023) 106370.doi:https://doi.org/10.1016/j.engappai.2023.106370

    work page doi:10.1016/j.engappai.2023.106370 2023

  57. [59]

    Maurizi, C

    M. Maurizi, C. Gao, F. Berto, Predicting stress, strain and deformation fields in materials and structures with graph neural networks, Scientific reports 12 (1) (2022) 21834.doi:https://doi.org/10.1038/ s41598-022-26424-3

    2022

  58. [60]

    Y . Cai, J. Jelovica, Efficient graph representation in graph neural networks for stress predictions in stiffened panels, Thin-Walled Structures 203 (2024) 112157.doi:https://doi.org/10.1016/j.tws.2024.112157

    work page doi:10.1016/j.tws.2024.112157 2024

  59. [61]

    Y . Cai, J. Jelovica, Heterogeneous graph representation of stiffened panels with non-uniform boundary conditions and loads, Engineering with Computers 42 (1) (2026) 5.doi:https://doi.org/10.1007/s00366-025-02261-6

    work page doi:10.1007/s00366-025-02261-6 2026

  60. [62]

    Y . Cai, J. Jelovica, A hybrid global–local computational framework for ship hull structural analysis using homoge- nized model and graph neural network, Marine Structures 109 (2026) 104077.doi:https://doi.org/10.1016/ j.marstruc.2026.104077

    arXiv 2026

  61. [63]

    M. Dai, M. F. Demirel, Y . Liang, J.-M. Hu, Graph neural networks for an accurate and interpretable prediction of the properties of polycrystalline materials, npj Computational Materials 7 (1) (2021) 103.doi:https://doi.org/10. 1038/s41524-021-00574-w

    2021

  62. [64]

    Perera, D

    R. Perera, D. Guzzetti, V . Agrawal, Graph neural networks for simulating crack coalescence and propagation in brittle materials, Computer Methods in Applied Mechanics and Engineering 395 (2022) 115021.doi:https://doi.org/ 24 10.1016/j.cma.2022.115021

    work page doi:10.1016/j.cma.2022.115021 2022

  63. [65]

    Karapiperis, D

    K. Karapiperis, D. M. Kochmann, Prediction and control of fracture paths in disordered architected materials us- ing graph neural networks, Communications Engineering 2 (1) (2023) 32.doi:https://doi.org/10.1038/ s44172-023-00085-0

    2023

  64. [66]

    Thomas, A

    A. Thomas, A. R. Durmaz, M. Alam, P. Gumbsch, H. Sack, C. Eberl, Materials fatigue prediction using graph neural networks on microstructure representations, Scientific Reports 13 (1) (2023) 12562.doi:https://doi.org/10. 1038/s41598-023-39400-2

    2023

  65. [67]

    J. Ho, A. Jain, P. Abbeel, Denoising diffusion probabilistic models, in: H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, H. Lin (Eds.), Advances in Neural Information Processing Systems, V ol. 33, Curran Associates, Inc., 2020, pp. 6840–6851. URLhttps://proceedings.neurips.cc/paper_files/paper/2020/file/ 4c5bcfec8584af0d967f1ab10179ca4b-Paper.pdf

    2020

  66. [68]

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, B. Poole, Score-based generative modeling through stochastic differential equations, in: International Conference on Learning Representations, 2021. URLhttps://openreview.net/forum?id=PxTIG12RRHS

    2021

  67. [69]

    Lipman, R

    Y . Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, M. Le, Flow matching for generative modeling, in: The Eleventh International Conference on Learning Representations, 2023. URLhttps://openreview.net/forum?id=PqvMRDCJT9t

    2023

  68. [70]

    Lipman, M

    Y . Lipman, M. Havasi, P. Holderrieth, N. Shaul, M. Le, B. Karrer, R. T. Chen, D. Lopez-Paz, H. Ben-Hamu, I. Gat, Flow matching guide and code, arXiv preprint arXiv:2412.06264 (2024).doi:https://doi.org/10.48550/ arXiv.2412.06264

    Pith/arXiv arXiv 2024

  69. [71]

    Albergo, N

    M. Albergo, N. M. Boffi, E. Vanden-Eijnden, Stochastic interpolants: A unifying framework for flows and diffusions, Journal of Machine Learning Research 26 (209) (2025) 1–80. URLhttp://jmlr.org/papers/v26/23-1605.html

    2025

  70. [72]

    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.doi:https://doi.org/ 10.1016/j.cma.2023.116343

    work page doi:10.1016/j.cma.2023.116343 2023

  71. [73]

    P. V . Kota, M. M. Rashid, S. Goswami, L. Graham-Brady, A hybrid conditional diffusion-deeponet framework for high-fidelity stress prediction in hyperelastic materials, arXiv preprint arXiv:2603.18225 (2026).doi:https:// doi.org/10.48550/arXiv.2603.18225

    work page doi:10.48550/arxiv.2603.18225 2026

  72. [74]

    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.doi:https://doi.org/10.1016/j.jcp.2023.111972

    work page doi:10.1016/j.jcp.2023.111972 2023

  73. [75]

    A. Kashefi, Flow matching and diffusion models via pointnet for generating fluid fields on irregular geometries, Computer Methods in Applied Mechanics and Engineering 458 (2026) 119037.doi:https://doi.org/10.1016/ j.cma.2026.119037

    arXiv 2026

  74. [76]

    Scalable Diffusion Models with Transformers

    W. Peebles, S. Xie, Scalable diffusion models with transformers, in: Proceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4195–4205.doi:https://doi.org/10.48550/arXiv.2212.09748

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2212.09748 2023

  75. [77]

    A. Tong, K. FATRAS, N. Malkin, G. Huguet, Y . Zhang, J. Rector-Brooks, G. Wolf, Y . Bengio, Improving and generalizing flow-based generative models with minibatch optimal transport, Transactions on Machine Learning ResearchExpert Certification (2024). URLhttps://openreview.net/forum?id=CD9Snc73AW

    2024

  76. [78]

    X. Liu, C. Gong, qiang liu, Flow straight and fast: Learning to generate and transfer data with rectified flow, in: The Eleventh International Conference on Learning Representations, 2023. URLhttps://openreview.net/forum?id=XVjTT1nw5z

    2023

  77. [79]

    Z. Geng, M. Deng, X. Bai, J. Z. Kolter, K. He, Mean flows for one-step generative modeling, in: The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2026. URLhttps://openreview.net/forum?id=uWj4s7rMnR

    2026

  78. [80]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, et al., An image is worth 16x16 words: Transformers for image recognition at scale, Interna- tional Conference on Learning Representations (2020). URLhttps://openreview.net/forum?id=YicbFdNTTy

    2020

  79. [81]

    High-Resolution Image Synthesis with Latent Diffusion Models

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, B. Ommer, High-resolution image synthesis with latent diffusion models, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 10684–10695.doi:https://doi.org/10.48550/arXiv.2112.10752. 25

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2112.10752 2022