Pith. sign in

REVIEW 3 major objections 85 references

Data-efficient continuous conditional denoising diffusion model for microstructure generation

T0 review · 3 major / 0 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read A continuous-condition diffusion model generates steel microstructures from sparse process data by treating nearby conditions as neighbors in training.

desk verdict Solid applied continuous-conditional diffusion for steel microstructures; the real contribution is the integrated pipeline and ablations, not a new algorithm, and the open-boundary / simulator-only limits are real but not fatal. read the letter →

arxiv 2607.10429 v1 pith:QGXPGHDG submitted 2026-07-11 cs.CE cond-mat.mtrl-sci

Tarakram Ramgopal , Gowtham Nimmal Haribabu , Hussein Farahani , Cornelis Bos , Siddhant Kumar This is my paper · ORCID
classification cs.CEcond-mat.mtrl-sci
keywords denoisingdiffusioncontinuousconditioningvicinallossmicrostructuregenerationlow-carbonsteelprocess-structuremapsclassifier-freeguidanceDDIM
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

Physics-based microstructure simulators are too slow for high-throughput process design, and standard generative models need huge datasets when the conditioning variables are continuous numbers rather than discrete labels. This paper shows that a denoising diffusion model can be trained on a compact set of process–microstructure pairs and still produce statistically representative low-carbon steel images for continuous manganese concentration. The key training trick is a vicinal loss that lets every image teach the model about nearby, previously unseen manganese values; classifier-free guidance and faster implicit sampling are then used to keep quality high and generation quick. On held-out continuous conditions the generated images recover phase morphology, grain-size and grain-count distributions, banding, phase fractions and interfacial area fractions that match cellular-automata ground truth. The result is a practical surrogate for process-structure maps that can be queried in seconds rather than hours.

What carries the argument

Vicinal loss (Eqs. 22–25): the indicator that normally requires an exact match between query condition y and a training condition yi is relaxed so that any yi within distance κ contributes; sampling from a Gaussian kernel around existing conditions then fills the continuous condition space during training.

What would settle it

Generate 100 microstructures at a continuous manganese value that lies between two training points but near the observed banding transition; if the grain-size KDE, band-formation index and phase fractions deviate beyond the ground-truth variance of the cellular-automata simulator, the vicinal-loss claim fails.

Watch

Extended reading notes

Core claim

Vicinal-loss training, which associates a continuous process condition with all dataset microstructures whose conditions lie within a fixed distance κ, combined with classifier-free guidance and DDIM sampling, yields data-efficient continuous conditional generation of representative low-carbon steel microstructures that match phase and grain morphology, grain-size distribution, phase fraction and interfacial-area statistics on unseen manganese values.

Load-bearing premise

Microstructures change smoothly enough with manganese concentration that treating nearby conditions as interchangeable does not systematically bias the learned distribution, especially near the sharp onset of austenite banding.

Editorial extensions

If this is right

  • Process-structure maps for continuous alloy or thermal parameters can be queried in seconds instead of hours of cellular-automata or phase-field runtime.
  • The same three-component recipe (vicinal loss + CFG + DDIM) can be reused for other continuous process variables and for multi-channel or three-dimensional microstructures.
  • Generated ensembles can be fed directly into crystal-plasticity or property models, closing a fast process–structure–property loop for real-time optimization.
  • Sparse experimental datasets become usable for continuous conditioning once the same vicinal training is applied.

Reading between the lines

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

  • Open grain-boundary artifacts remain the main quality bottleneck; a topology-aware regularizer or multi-scale loss would be a natural next architectural step.
  • The method should transfer to other continuous-condition materials problems (composition, cooling rate, pressure) where physics simulators are expensive and data are sparse.
  • Because DDIM already cuts sampling cost by roughly 4 imes, further distillation or consistency-model acceleration could make on-line industrial control feasible.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 0 minor

Summary. The paper proposes a continuous conditional denoising diffusion model for generating low-carbon steel microstructures conditioned on continuous manganese concentration. Building on a standard DDPM reverse process with a U-Net, the authors introduce (i) a vicinal-loss training objective that softens the indicator over process conditions within a radius κ (Eqs. 22–25, §2.5), (ii) classifier-free guidance, and (iii) DDIM sampling. Training data are CASIPT cellular-automata simulations (N=40k pairs after float16 quantization). On 10 held-out χ_avg_Mn values the model is shown to reproduce grain morphology, phase maps, autocorrelation, grain-size/count KDEs, band-formation index, IPB area fractions and ferrite phase fraction, with component-wise ablations and limited data-sparsity studies.

Significance. If the central claim holds, the work supplies a practical, data-efficient surrogate for continuous process–structure maps that classical categorical-conditional diffusion models do not address. The combination of vicinal loss, CFG and DDIM is a concrete engineering contribution for materials applications where process parameters are continuous and simulation budgets are limited. Strengths include systematic ablations (DDIM, CFG, vicinal, all-off), quantitative multi-descriptor evaluation on held-out continuous conditions, and explicit discussion of open-boundary limitations. The framework is data-agnostic in principle and could accelerate high-throughput process design once the smoothness assumption is better stress-tested.

major comments (3)
  1. §2.5, Eqs. (22)–(25) and §3.5: The load-bearing claim that vicinal loss enables genuine continuous conditioning rests on the assumption that p(X|y) varies smoothly enough that associating a query y with all Xi whose yi lie inside κ=0.00024 does not bias the conditional. Banding onset is a relatively sharp morphological transition (autocorrelation and band-formation index change rapidly between ~0.0065–0.0109; Figs. 6, 8, 14–15). Ablations that disable vicinal loss already degrade precisely these descriptors in that window, yet the paper reports neither a controlled κ-sweep nor a leave-one-transition-out experiment. Without such a test it remains possible that residual agreement partly reflects local averaging rather than true interpolation; this must be addressed before the data-efficiency claim can be accepted at face value.
  2. Appendix A and §2.2: Dataset construction uses biased rejection sampling that densifies high-χ_Mn points and float16 quantization that collapses 1000 unique conditions to 746. Both choices interact with the vicinal kernel (Eq. 23) and the vicinity radius. The manuscript should quantify how many unique conditions fall inside a typical κ-ball near the banding transition and whether the reported statistics remain stable under uniform (unbiased) sampling of χ_avg_Mn; otherwise the claimed data efficiency is conditioned on a sampling scheme that already concentrates data where morphology changes most.
  3. §3.7 and quantitative panels of Figs. 8, 10, 13, 15, 17: Open/disconnected grain boundaries systematically skew grain-size/count KDEs and IPB fractions. The paper correctly flags the issue but still presents those descriptors as primary evidence of statistical fidelity. Either a boundary-aware post-processing correction or an explicit sensitivity analysis (e.g., metrics recomputed after morphological closing) is needed so that readers can judge how much of the reported agreement survives the known failure mode.

Circularity Check

0 steps flagged · score 1.0 of 10

Standard simulator-trained generative surrogate with empirical held-out validation; no derivation reduces by construction to its inputs.

full rationale

The paper trains a continuous conditional diffusion model (U-Net reverse process with vicinal loss, CFG, DDIM) on CASIPT process–microstructure pairs and evaluates statistical fidelity (grain size/count KDEs, band-formation index, IPB fractions, phase fraction, autocorrelation) on held-out continuous χ_avg_Mn values never seen in training. That is ordinary likelihood-style surrogate modeling: the generated samples are not forced to equal training labels by identity, nor is any reported statistic a fitted constant renamed as a prediction. Vicinal loss (Eqs. 22–25) relaxes the indicator 1{|y−yi|≤κ} to improve data efficiency under continuous conditioning; it softens associations but does not make p̂(X|y) equal the training set by construction. Ablations that disable vicinal loss, CFG or DDIM degrade the same metrics, confirming the components contribute rather than tautologically force the result. Self-citations (Bos et al. for the CASIPT data generator; Ding et al. for the external continuous-conditioning/vicinal-loss technique) supply the simulator and a prior method, respectively; neither is a uniqueness theorem or load-bearing ansatz that collapses the central claim. No equation equates a predicted quantity to a fitted input, and no self-citation chain replaces independent empirical checks. The only residual self-reference is the inherent use of the same physics-based simulator for both training and ground-truth statistics—an expected feature of any surrogate study, not circularity. Score 1 reflects that minor, non-load-bearing self-reference only.

Assumptions & free parameters 6 free parameters · 4 assumptions · 2 invented entities

The central claim rests on standard diffusion modeling assumptions, a smoothness prior encoded as vicinal loss with hand-chosen κ/σ_δ, simulation-as-ground-truth from CASIPT, and several sampling/guidance hyperparameters. No new physical entities are postulated; algorithmic constructs (single-channel phase/boundary encoding, covariance condition embedding) are engineering choices.

free parameters (6)
  • vicinity radius κ = 0.00024
    Controls which nearby process conditions share microstructure labels in the vicinal loss; set to 0.00024 (Table E.1). Central to the data-efficiency claim.
  • vicinal kernel std σ_δ = 0.00069
    Variance of Gaussian kernel density for sampling continuous y near dataset conditions (Eq. 23); set to 0.00069.
  • CFG scale Γ and dropout p_drop = Γ=1.5, p_drop=0.1
    Trade conditional fidelity vs diversity; Γ=1.5, p_drop=0.1 in main model; ablated to Γ=1 and p_drop≈0.
  • diffusion steps T and DDIM steps T' = T=1000, T'=250
    T=1000 training noising steps; T'=250 sampling steps for speed claim.
  • cosine noise schedule offset s = 0.008
    Controls β_t schedule (Appendix B); s=0.008.
  • embedding dimensions d, d_Ξ, d_t and learning rate = d=128, d_Ξ=65536, lr=1e-5
    Architecture/optimizer choices that affect capacity and convergence (d=128, d_Ξ=65536, d_t=64, lr=1e-5, 100k epochs).
assumptions (4)
  • standard math Forward/reverse diffusion as Gaussian Markov chain with shared covariance form (standard DDPM assumptions, §2.3).
    Underpins the KL training objective and sampling equations; taken from Ho et al. / Luo.
  • domain assumption CASIPT mixed-mode cellular automata with fixed cooling schedule and sinusoidal Mn banding produce statistically valid process-structure pairs for low-carbon steel (§2.2).
    All training and evaluation labels come from this simulator; real experimental microstructures are not used.
  • ad hoc to paper Conditional microstructure distribution varies smoothly enough that fuzzy association of y with Xi for |y−yi|≤κ is a valid training surrogate (§2.5).
    Load-bearing for the data-efficiency claim; justified by citation to Ding et al. and ablations, not derived from materials theory.
  • domain assumption Single-channel discrete pixel encoding {−1,−0.25,0.25,1} plus post-hoc thresholding fully represents phases and grain boundaries for generative learning (§2.1, Eq. 30).
    Simplifies multi-phase structure; open-boundary failures in §3.7 show the encoding/generation of closed boundaries is fragile.
invented entities (2)
  • Covariance condition embedding H_y via ResNet-MLP cascade
    purpose: Inject continuous process condition into the noise covariance of forward/reverse diffusion (§2.4).
    Engineering construct adapted from continuous conditional generation literature; no independent physical meaning.
  • Single-channel multi-value microstructure image representation
    purpose: Encode austenite/ferrite interiors and boundaries in one channel for efficient generation (§2.1).
    Paper-specific encoding choice; validated only by successful training on CASIPT images.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-efficient continuous conditional denoising diffusion model for microstructure generation." pith.science (2026). https://pith.science/paper/QGXPGHDG

@misc{pith2026260710429,
  author       = {Pith},
  title        = {Pith review of: Data-efficient continuous conditional denoising diffusion model for microstructure generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QGXPGHDG}},
  note         = {Machine review of arXiv:2607.10429}
}
read the original abstract

Traditional computational models, such as cellular automata and phase-field methods, are effective for simulating microstructural evolution but often face computational bottlenecks, limiting their application in high-throughput and on-demand process optimization. Generative machine learning approaches, such as denoising diffusion models, have emerged as powerful tools for surrogate modeling of process-structure maps, specifically producing representative microstructures conditioned on process parameters. However, they often require large amounts of data for training, particularly when process conditions are continuous rather than discrete categorical variables. To address this, we present a continuous conditional denoising diffusion model for generating microstructures conditioned on processing parameters. Trained on a compact dataset of process-microstructure pairs, this framework first adds noise to microstructure images and then trains a neural network to progressively remove the noise, learning the underlying statistical patterns of the microstructure. To address data inefficiencies associated with continuously valued process conditions, we propose a vicinal-loss training strategy that associates process conditions in data-sparse regions with nearby conditions in the dataset. Combined with classifier-free guidance and denoising diffusion implicit sampling, this approach enables data-efficient continuous conditional generation of microstructures compared to classical denoising diffusion models. The model successfully generates representative microstructures for low-carbon steel conditioned on manganese composition, matching key physical features such as phase and grain morphology, grain size distribution, phase fraction, and interfacial area distribution. More generally, this approach opens avenues for efficient process design and optimization of materials and their microstructures.

Figures

Figures reproduced from arXiv: 2607.10429 by the authors.

Figure 1
Figure 1. (a) Schematic of the image representation used for two-dimensional low-carbon steel microstructures, with insets showcasing phase [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Schematic of the denoising diffusion model, where the forward diffusion process is performed using q(X t | X t−1 , y) for t ∈ {1, . . . , T}, and the reverse diffusion process is parameterized using a NN: pθ(X t−1 | X t , y) with the set of trainable weights θ. Reverse diffusion process: We aim to acquire X 0 (y) from a y-dependent Gaussian noise X T ∼ N(0, Hy) for given process condition y. However, estimating the … view at source ↗
Figure 3
Figure 3. Schematic of training scheme for high-dimensional latent embeddings [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (15 more)
Figure 4
Figure 4. Figure 4: Schematic of vicinal loss-based training. (a) In non-vicinal training, a queried process condition [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: U-net network architecture of the denoising di [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison of microstructures between ground truth and di [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Evolution of the t-SNE map of ground-truth and generated microstructures at various training epochs. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Quantitative comparison of microstructure statistics between ground truth (gray dashed line with circles) and di [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: DDIM ablation: Qualitative comparison of microstructures between ground truth and DDIM-ablation model at specific χ avg Mn values (indicated above the top panel) not present in the training dataset. Top panel: grain morphology maps. Middle panel: phase maps. Bottom pan…
Figure 10
Figure 10. Figure 10: DDIM ablation: Quantitative comparison of microstructure statistics between ground truth (gray dashed line with circles) and DDIM￾ablation model (red line with filled stars) at specific χ avg Mn not present in the training dataset. The χ avg Mn values for the top and …
Figure 11
Figure 11. Figure 11: Comparison of sampling speed (averaged across 1 [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]
Figure 12
Figure 12. Figure 12: CFG ablation: Qualitative comparison of microstructures between ground truth and CFG-ablation model at specific χ avg Mn values (indicated above the top panel) not present in the training dataset. Top panel: grain morphology maps. Middle panel: phase maps. Bottom pane…
Figure 13
Figure 13. Figure 13: CFG ablation: Quantitative comparison of microstructure statistics between ground truth (gray dashed line with circles) and CFG￾ablation model (red line with filled stars) at specific χ avg Mn not present in the training dataset. The χ avg Mn values for the top and mi…
Figure 14
Figure 14. Figure 14: Vicinal loss ablation: Qualitative comparison of microstructures between ground truth and vicinal-loss-ablation model at specific χ avg Mn values (indicated above the top panel) not present in the training dataset. Top panel: grain morphology maps. Middle panel: phase…
Figure 15
Figure 15. Figure 15: Vicinal loss ablation: Quantitative comparison of microstructure statistics between ground truth (gray dashed line with circles) and vicinal-loss-ablation model (red line with filled stars) at specific χ avg Mn not present in the training dataset. The χ avg Mn values …
Figure 16
Figure 16. Figure 16: DDIM, CFG, and vicinal loss ablation: Qualitative comparison of microstructures between ground truth and ablation model at specific χ avg Mn values (indicated above the top panel) not present in the training dataset. Top panel: grain morphology maps. Middle panel: pha…
Figure 17
Figure 17. Figure 17: DDIM, CFG, and vicinal loss ablation: Quantitative comparison of microstructure statistics between ground truth (gray dashed line with circles) and ablation model (red line with filled stars) at specific χ avg Mn not present in the training dataset. The χ avg Mn value…
Figure 18
Figure 18. Figure 18: Representative example of discrepancies (highlighted by red circles) associated with open grain boundaries in microstructures generated [PITH_FULL_IMAGE:figures/full_fig_p029_18.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

85 extracted references · 3 canonical work pages

  1. [1]

    Journal of Machine Learning Research , year =

    Laurens van der Maaten and Geoffrey Hinton , title =. Journal of Machine Learning Research , year =

  2. [2]

    Journal of Machine Learning Research , year =

    Yingfan Wang and Haiyang Huang and Cynthia Rudin and Yaron Shaposhnik , title =. Journal of Machine Learning Research , year =

  3. [3]

    Phase-field models for microstructure evolution , volume =

    Long Qing Chen , doi =. Phase-field models for microstructure evolution , volume =. Annual Review of Materials Science , keywords =

  4. [4]

    Accelerating phase-field simulation of three-dimensional microstructure evolution in laser powder bed fusion with composable machine learning predictions , volume =

    Jin Young Choi and Tianju Xue and Shuheng Liao and Jian Cao , doi =. Accelerating phase-field simulation of three-dimensional microstructure evolution in laser powder bed fusion with composable machine learning predictions , volume =. Additive Manufacturing , keywords =

  5. [5]

    H. K.D.H. Bhadeshia , doi =. Computational design of advanced steels , volume =. Scripta Materialia , keywords =

  6. [6]

    Siboni and Ghasem Alahyarizadeh and Reza Ghaderi and Bob Svendsen and Dierk Raabe and Jaber Rezaei Mianroodi , doi =

    Iman Peivaste and Nima H. Siboni and Ghasem Alahyarizadeh and Reza Ghaderi and Bob Svendsen and Dierk Raabe and Jaber Rezaei Mianroodi , doi =. Machine-learning-based surrogate modeling of microstructure evolution using phase-field , volume =. Computational Materials Science , keywords =

  7. [7]

    Hu and S

    C. Hu and S. Martin and R. Dingreville , doi =. Accelerating phase-field predictions via recurrent neural networks learning the microstructure evolution in latent space , volume =. Computer Methods in Applied Mechanics and Engineering , keywords =

  8. [8]

    CNN-Based Surrogate for the Phase Field Damage Model: Generalization across Microstructure Parameters for Composite Materials , volume =

    Yuxiang Gao and Matthew Berger and Ravindra Duddu , doi =. CNN-Based Surrogate for the Phase Field Damage Model: Generalization across Microstructure Parameters for Composite Materials , volume =. Journal of Engineering Mechanics , month =

Show all 85 references
  1. [9]

    Physics-embedded graph network for accelerating phase-field simulation of microstructure evolution in additive manufacturing , volume =

    Tianju Xue and Zhengtao Gan and Shuheng Liao and Jian Cao , doi =. Physics-embedded graph network for accelerating phase-field simulation of microstructure evolution in additive manufacturing , volume =. npj Computational Materials , keywords =

  2. [10]

    Stewart and Rémi Dingreville , doi =

    David Montes de Oca Zapiain and James A. Stewart and Rémi Dingreville , doi =. Accelerating phase-field-based microstructure evolution predictions via surrogate models trained by machine learning methods , volume =. npj Computational Materials , keywords =

  3. [11]

    Hore and S

    S. Hore and S. K. Das and S. Banerjee and S. Mukherjee , doi =. A multiscale coupled Monte Carlo model to characterize microstructure evolution during hot rolling of Mo-TRIP steel , volume =. Acta Materialia , keywords =

  4. [12]

    Modeling the austenite-ferrite isothermal transformation in an Fe-C binary system and experimental verification , volume =

    Mingming Tong and Dianzhong Li and Yiyi Li and Jun Ni , doi =. Modeling the austenite-ferrite isothermal transformation in an Fe-C binary system and experimental verification , volume =. Metallurgical and Materials Transactions A: Physical Metallurgy and Materials Science , keywords =

  5. [13]

    Yazdipour and C

    N. Yazdipour and C. H.J. Davies and P. D. Hodgson , doi =. Microstructural modeling of dynamic recrystallization using irregular cellular automata , volume =. Computational Materials Science , keywords =

  6. [14]

    Bos and M

    C. Bos and M. G. Mecozzi and J. Sietsma , doi =. A microstructure model for recrystallisation and phase transformation during the dual-phase steel annealing cycle , volume =. Computational Materials Science , keywords =

  7. [15]

    Bos and M

    C. Bos and M. G. Mecozzi and D. N. Hanlon and M. P. Aarnts and J. Sietsma , doi =. Application of a three-dimensional microstructure evolution model to identify key process settings for the production of dual-phase steels , volume =. Metallurgical and Materials Transactions A:...

  8. [16]

    and Wang, L

    Yang, M. and Wang, L. and Yan, W. , year =. Phase-field modeling of grain evolutions in additive manufacturing from nucleation, growth, to coarsening , journal =

  9. [17]

    and Batra, R

    Ramprasad, R. and Batra, R. and Pilania, G. and Mannodi-Kanakkithodi, A. and Kim, C. , year =. Machine learning in materials informatics: recent applications and prospects , journal =

  10. [18]

    and Tanaka, Y

    Takahashi, K. and Tanaka, Y. , year =. Materials informatics: a journey towards material design and synthesis , journal =

  11. [19]

    Deringer, V. L. , year =. Modelling and understanding battery materials with machine-learning-driven atomistic simulations , journal =

  12. [20]

    Schleder, G. R. and Padilha, A. C. M. and Acosta, C. M. and Costa, M. and Fazzio, A. , year =. From DFT to machine learning: recent approaches to materials science–a review , journal =

  13. [21]

    and others , year =

    Meng, L. and others , year =. Machine Learning in Additive Manufacturing: A Review , journal =

  14. [22]

    and Tan, X

    Wang, C. and Tan, X. P. and Tor, S. B. and Lim, C. S. , year =. Machine learning in additive manufacturing: State-of-the-art and perspectives , journal =

  15. [23]

    GrainNN: A neighbor-aware long short-term memory network for predicting microstructure evolution during polycrystalline grain formation , volume =

    Yigong Qin and Stephen DeWitt and Balasubramaniam Radhakrishnan and George Biros , doi =. GrainNN: A neighbor-aware long short-term memory network for predicting microstructure evolution during polycrystalline grain formation , volume =. Computational Materials Science , keywords =

  16. [24]

    and Pokharel, R

    Pandey, A. and Pokharel, R. , year =. Machine learning based surrogate modeling approach for mapping crystal deformation in three dimensions , journal =

  17. [25]

    and Acar, P

    Senthilnathan, A. and Acar, P. and De Graef, M. , year =. Markov Random Field based microstructure reconstruction using the principal image moments , journal =

  18. [26]

    and Randazzo, E

    Mordvintsev, A. and Randazzo, E. and Niklasson, E. and Levin, M. , year =. Growing Neural Cellular Automata , journal =. doi:10.23915/distill.00023 , url =

    doi

  19. [27]

    Neural cellular automata for solidification microstructure modelling , volume =

    Jian Tang and Siddhant Kumar and Laura De Lorenzis and Ehsan Hosseini , doi =. Neural cellular automata for solidification microstructure modelling , volume =. Computer Methods in Applied Mechanics and Engineering , keywords =

  20. [28]

    Denoising Diffusion Probabilistic Models , volume =

    Jonathan Ho and Ajay Jain and Pieter Abbeel , doi =. Denoising Diffusion Probabilistic Models , volume =. Advances in Neural Information Processing Systems , month =

  21. [29]

    Improved Denoising Diffusion Probabilistic Models , volume =

    Alex Nichol and Prafulla Dhariwal , doi =. Improved Denoising Diffusion Probabilistic Models , volume =. Proceedings of Machine Learning Research , month =

  22. [30]

    Guided diffusion for fast inverse design of voxel-based mechanical metamaterials , volume =

    Yanyan Yang and Lili Wang and Xiaoya Zhai and Kai Chen and Wenming Wu and Yunkai Zhao and Falai Chen and Ligang Liu and Xiao Ming Fu , doi =. Guided diffusion for fast inverse design of voxel-based mechanical metamaterials , volume =. Smart Materials in Manufacturing , keywords =

  23. [31]

    and Kochmann, DM

    Bastek, JH. and Kochmann, DM. , year =. Inverse design of nonlinear mechanical metamaterials via video denoising diffusion models , journal =

  24. [32]

    Physics-Informed Diffusion Models , url =

    Bastek, Jan-Hendrik and Sun, WaiChing and Kochmann, Dennis , booktitle =. Physics-Informed Diffusion Models , url =

  25. [33]

    Predictive microstructure image generation using denoising diffusion probabilistic models , volume =

    Erfan Azqadan and Hamid Jahed and Arash Arami , doi =. Predictive microstructure image generation using denoising diffusion probabilistic models , volume =. Acta Materialia , keywords =

  26. [34]

    Understanding Diffusion Models: A Unified Perspective , url =

    Calvin Luo , doi =. Understanding Diffusion Models: A Unified Perspective , url =

  27. [35]

    Kirka , doi =

    Patxi Fernandez-Zelaia and Jiahao Cheng and Jason Mayeur and Amir Koushyar Ziabari and Michael M. Kirka , doi =. Digital polycrystalline microstructure generation using diffusion probabilistic models , volume =. Materialia , keywords =

  28. [36]

    Reconstructing Microstructures From Statistical Descriptors Using Neural Cellular Automata , volume =

    Paul Seibert and Alexander Raßloff and Yichi Zhang and Karl Kalina and Paul Reck and Daniel Peterseim and Markus Kästner , doi =. Reconstructing Microstructures From Statistical Descriptors Using Neural Cellular Automata , volume =. Integrating Materials and Manufacturing Inno...

  29. [37]

    Conditional diffusion-based microstructure reconstruction , volume =

    Christian Düreth and Paul Seibert and Dennis Rücker and Stephanie Handford and Markus Kästner and Maik Gude , doi =. Conditional diffusion-based microstructure reconstruction , volume =. Materials Today Communications , keywords =

  30. [38]

    Microstructure reconstruction of 2D/3D random materials via diffusion-based deep generative models , volume =

    Xianrui Lyu and Xiaodan Ren , doi =. Microstructure reconstruction of 2D/3D random materials via diffusion-based deep generative models , volume =. Scientific Reports 2024 14:1 , keywords =

    2024

  31. [39]

    U-net: Convolutional networks for biomedical image segmentation , volume =

    Olaf Ronneberger and Philipp Fischer and Thomas Brox , doi =. U-net: Convolutional networks for biomedical image segmentation , volume =. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) , pages =

  32. [40]

    Kingma and Max Welling , doi =

    Diederik P. Kingma and Max Welling , doi =. Auto-Encoding Variational Bayes , url =. 2nd International Conference on Learning Representations, ICLR 2014 - Conference Track Proceedings , month =

    2014

  33. [41]

    Deep Residual Learning for Image Recognition , url =

    Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun , doi =. Deep Residual Learning for Image Recognition , url =. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition , month =

  34. [42]

    Kingma and Tim Salimans and Ben Poole and Jonathan Ho , doi =

    Diederik P. Kingma and Tim Salimans and Ben Poole and Jonathan Ho , doi =. Variational Diffusion Models , volume =. Advances in Neural Information Processing Systems , month =

  35. [43]

    Machine learning-based microstructure prediction during laser sintering of alumina , volume =

    Jianan Tang and Xiao Geng and Dongsheng Li and Yunfeng Shi and Jianhua Tong and Hai Xiao and Fei Peng , doi =. Machine learning-based microstructure prediction during laser sintering of alumina , volume =. Scientific Reports , keywords =

  36. [44]

    Epting and Hokon Kim and Harry W

    Tim Hsu and William K. Epting and Hokon Kim and Harry W. Abernathy and Gregory A. Hackett and Anthony D. Rollett and Paul A. Salvador and Elizabeth A. Holm , doi =. Microstructure Generation via Generative Adversarial Network for Heterogeneous, Topologically Complex 3D Materia...

  37. [45]

    Choi and H

    Sehyun Chun and Sidhartha Roy and Yen Thi Nguyen and Joseph B. Choi and H. S. Udaykumar and Stephen S. Baek , doi =. Deep learning for synthetic microstructure generation in a materials-by-design framework for heterogeneous energetic materials , volume =. Scientific Reports , ...

  38. [46]

    Towards automatic feature extraction and sample generation of grain structure by variational autoencoder , volume =

    Ye Ji and Arnd Koeppe and Patrick Altschuh and Deepalaxmi Rajagopal and Yinghan Zhao and Weijin Chen and Yi Zhang and Yue Zheng and Britta Nestler , doi =. Towards automatic feature extraction and sample generation of grain structure by variational autoencoder , volume =. Comp...

  39. [47]

    Exploration of optimal microstructure and mechanical properties in continuous microstructure space using a variational autoencoder , volume =

    Yongju Kim and Hyung Keun Park and Jaimyun Jung and Peyman Asghari-Rad and Seungchul Lee and Jin You Kim and Hwan Gyo Jung and Hyoung Seop Kim , doi =. Exploration of optimal microstructure and mechanical properties in continuous microstructure space using a variational autoen...

  40. [48]

    Buzzy and Andreas E

    Michael O. Buzzy and Andreas E. Robertson and Surya R. Kalidindi , doi =. Statistically conditioned polycrystal generation using denoising diffusion models , volume =. Acta Materialia , keywords =

  41. [49]

    Vlassis and Wai Ching Sun and Khalid A

    Nikolaos N. Vlassis and Wai Ching Sun and Khalid A. Alshibli and Richard A. Regueiro , doi =. Synthesizing realistic sand assemblies with denoising diffusion in latent space , volume =. International Journal for Numerical and Analytical Methods in Geomechanics , keywords =

  42. [50]

    Bos and J

    C. Bos and J. Sietsma , doi =. A mixed-mode model for partitioning phase transformations , volume =. Scripta Materialia , keywords =

  43. [51]

    Jane Wang , doi =

    Xin Ding and Yongwei Wang and Kao Zhang and Z. Jane Wang , doi =. CCDM: Continuous Conditional Diffusion Models for Image Generation , url =

  44. [52]

    Classifier-Free Diffusion Guidance , url =

    Jonathan Ho and Tim Salimans , doi =. Classifier-Free Diffusion Guidance , url =

  45. [53]

    Welch and Z

    Xin Ding and Yongwei Wang and Zuheng Xu and William J. Welch and Z. Jane Wang , doi =. CcGAN: Continuous Conditional Generative Adversarial Networks for Image Generation , volume =. IEEE Transactions on Pattern Analysis and Machine Intelligence , keywords =

  46. [54]

    Denoising Diffusion Implicit Models , url =

    Jiaming Song and Chenlin Meng and Stefano Ermon , doi =. Denoising Diffusion Implicit Models , url =. ICLR 2021 - 9th International Conference on Learning Representations , month =

    2021

  47. [55]

    A deep adversarial approach for the generation of synthetic titanium alloy microstructures with limited training data , volume =

    Gowtham Nimmal Haribabu and Jeyapriya Thimukonda Jegadeesan and Chiranjib Bhattacharya and Bikramjit Basu , doi =. A deep adversarial approach for the generation of synthetic titanium alloy microstructures with limited training data , volume =. Computational Materials Science ...

  48. [56]

    White and Gowtham Nimmal Haribabu and Jeyapriya Thimukonda Jegadeesan and Bikramjit Basu and Philip J

    Michael D. White and Gowtham Nimmal Haribabu and Jeyapriya Thimukonda Jegadeesan and Bikramjit Basu and Philip J. Withers and Chris P. Race , doi =. Exploring descriptors for titanium microstructure via digital fingerprints from variational autoencoders , volume =. Computation...

  49. [57]

    Attention Is All You Need , url =

    Ashish Vaswani and Google Brain and Noam Shazeer and Niki Parmar and Jakob Uszkoreit and Llion Jones and Aidan N Gomez and Łukasz Kaiser and Illia Polosukhin , doi =. Attention Is All You Need , url =

  50. [58]

    Group Normalization , volume =

    Yuxin Wu and Kaiming He , doi =. Group Normalization , volume =. International Journal of Computer Vision , keywords =

  51. [59]

    Sigmoid-weighted linear units for neural network function approximation in reinforcement learning , volume =

    Stefan Elfwing and Eiji Uchibe and Kenji Doya , doi =. Sigmoid-weighted linear units for neural network function approximation in reinforcement learning , volume =. Neural Networks , keywords =

  52. [60]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , volume =

    Sergey Ioffe and Christian Szegedy , doi =. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , volume =. 32nd International Conference on Machine Learning, ICML 2015 , month =

    2015

  53. [61]

    ACDC: A Structured Efficient Linear Layer , url =

    Marcin Moczulski and Misha Denil and Jeremy Appleyard and Nando De Freitas , doi =. ACDC: A Structured Efficient Linear Layer , url =. 4th International Conference on Learning Representations, ICLR 2016 - Conference Track Proceedings , month =

    2016

  54. [62]

    Deep Learning using Rectified Linear Units (ReLU) , url =

    Abien Fred Agarap , doi =. Deep Learning using Rectified Linear Units (ReLU) , url =

  55. [63]

    2015 , eprint=

    An Introduction to Convolutional Neural Networks , author=. 2015 , eprint=

    2015

  56. [64]

    The Theoretical Expressiveness of Maxpooling , url =

    Kyle Matoba and Kyle Matoba@epfl Ch and Nikolaos Dimitriadis and François Fleuret , doi =. The Theoretical Expressiveness of Maxpooling , url =

  57. [65]

    Nearest Neighbor Value Interpolation , volume =

    Olivier Rukundo and Hanqiang Cao , doi =. Nearest Neighbor Value Interpolation , volume =. International Journal of Advanced Computer Science and Applications , keywords =

  58. [66]

    Skip Connections Matter: On the Transferability of Adversarial Examples Generated with ResNets , url =

    Dongxian Wu and Yisen Wang and Shu Tao Xia and James Bailey and Xingjun Ma , doi =. Skip Connections Matter: On the Transferability of Adversarial Examples Generated with ResNets , url =. 8th International Conference on Learning Representations, ICLR 2020 , month =

    2020

  59. [67]

    Kingma and Jimmy Lei Ba , doi =

    Diederik P. Kingma and Jimmy Lei Ba , doi =. Adam: A Method for Stochastic Optimization , url =. 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings , month =

    2015

  60. [68]

    Freeman and Taesung Park , doi =

    Tianwei Yin and Michaël Gharbi and Richard Zhang and Eli Shechtman and Frédo Durand and William T. Freeman and Taesung Park , doi =. One-step Diffusion with Distribution Matching Distillation , url =. Proceedings of the IEEE Computer Society Conference on Computer Vision and P...

  61. [69]

    Freeman , doi =

    Tianwei Yin and Michaël Gharbi and Taesung Park and Richard Zhang and Eli Shechtman and Frédo Durand and William T. Freeman , doi =. Improved Distribution Matching Distillation for Fast Image Synthesis , volume =. Advances in Neural Information Processing Systems , month =

  62. [70]

    2014 , eprint=

    Generative Adversarial Networks , author=. 2014 , eprint=

    2014

  63. [71]

    2021 , publisher =

    Standard Test Methods for Determining Average Grain Size , author =. 2021 , publisher =. doi:10.1520/E0112-13R21 , url =

    2021 doi

  64. [72]

    and Khanbareh, H

    Gutiérrez, M.A. and Khanbareh, H. and van der Zwaag, S. , year =. Computational modeling of structure formation during dielectrophoresis in particulate composites , volume =. doi:10.1016/j.commatsci.2015.10.011 , journal =

    2015 doi

  65. [73]

    Inverse design of microstructures using conditional continuous normalizing flows , volume =

    Hossein Mirzaee and Serveh Kamrava , doi =. Inverse design of microstructures using conditional continuous normalizing flows , volume =. Acta Materialia , keywords =

  66. [74]

    , title =

    Farahani, H. , title =. doi:https://doi.org/10.4233/uuid:26f3b0db-4c77-4564-8a59-a802fff39028 , url =

    doi

  67. [75]

    Oliphant and Matt Haberland and Tyler Reddy and David Cournapeau and Evgeni Burovski and Pearu Peterson and Warren Weckesser and Jonathan Bright and Stéfan J

    Pauli Virtanen and Ralf Gommers and Travis E. Oliphant and Matt Haberland and Tyler Reddy and David Cournapeau and Evgeni Burovski and Pearu Peterson and Warren Weckesser and Jonathan Bright and Stéfan J. van der Walt and Matthew Brett and Joshua Wilson and K. Jarrod Millman a...

  68. [76]

    Effect of manganese bands on austenite formation of low carbon steels in dual phase steel manufacture , volume =

    Bharath Bandi and Joost Van Krevel and Sukalpan Nandi and Zulfiqar Husain and Prakash Srirangam , doi =. Effect of manganese bands on austenite formation of low carbon steels in dual phase steel manufacture , volume =. Materials Today: Proceedings , keywords =

  69. [77]

    NICE: Non-linear Independent Components Estimation , url =

    Laurent Dinh and David Krueger and Yoshua Bengio , doi =. NICE: Non-linear Independent Components Estimation , url =. 3rd International Conference on Learning Representations, ICLR 2015 - Workshop Track Proceedings , month =

    2015

  70. [78]

    Variational Inference with Normalizing Flows , volume =

    Danilo Jimenez Rezende and Shakir Mohamed , doi =. Variational Inference with Normalizing Flows , volume =. 32nd International Conference on Machine Learning, ICML 2015 , month =

    2015

  71. [79]

    1994 , isbn =

    Haykin, Simon , title =. 1994 , isbn =

    1994

  72. [80]

    Wang, Phil , title =

  73. [81]

    Thermo-Calc and DICTRA, computational tools for materials science , volume =

    Andersson, J-O and Helander, Thomas and H\". Thermo-Calc and DICTRA, computational tools for materials science , volume =. Calphad , publisher =. 2002 , month = june, pages =. doi:10.1016/s0364-5916(02)00037-8 , number =

    2002 doi

  74. [82]

    3D Denoising Diffusion Probabilistic Models for 3D microstructure image generation of fuel cell electrodes , volume =

    Abdelouahid Bentamou and Stephane Chretien and Yann Gavet , doi =. 3D Denoising Diffusion Probabilistic Models for 3D microstructure image generation of fuel cell electrodes , volume =. Computational Materials Science , keywords =

  75. [83]

    3D multiphase heterogeneous microstructure generation using conditional latent diffusion models , volume =

    Nirmal Baishnab and Ethan Herron and Aditya Balu and Soumik Sarkar and Adarsh Krishnamurthy and Baskar Ganapathysubramanian , doi =. 3D multiphase heterogeneous microstructure generation using conditional latent diffusion models , volume =. Digital Discovery , month =

  76. [84]

    , author =

    Accelerate: Training and inference at scale made simple, efficient and adaptable. , author =

  77. [85]

    Exponential Moving Average of Weights in Deep Learning: Dynamics and Benefits , volume =

    Daniel Morales-Brotons and Thijs Vogels and Hadrien Hendrikx , doi =. Exponential Moving Average of Weights in Deep Learning: Dynamics and Benefits , volume =. Transactions on Machine Learning Research , month =

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.