Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

PolyMicros: Bootstrapping a Foundation Model for Polycrystalline Material Structure

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A physics-driven augmentation loop converts five experimental microstructures into a training set large enough for a foundation model of polycrystalline materials.

desk verdict Bootstrapping a diffusion prior for polycrystals from five volumes is a real idea; the two zero-shot demos are suggestive but rest on one held-out volume each. read the letter →

arxiv 2506.11055 v1 pith:LO5AJ4ZP submitted 2025-05-22 cs.LG cond-mat.mtrl-sci

Michael Buzzy , Andreas Robertson , Peng Chen , Surya Kalidindi This is my paper
classification cs.LGcond-mat.mtrl-sci
keywords foundationmodelspolycrystallinemicrostructuredataaugmentationdiffusion2-pointspatialstatisticssuper-resolution2D-to-3Dreconstructionmaterialsinformatics
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

PolyMicros starts from five experimental 3D polycrystalline microstructures and claims that a physics-guided augmentation loop can turn hyper-sparse data into a large, statistically diverse training set for a foundation model. The augmentation combines a design-of-experiments sweep over plausible two-point spatial statistics with an ensemble of local generative diffusion models, one per experimental sample, then trains a single unconditional diffusion model on the resulting 30,000 synthetic volumes. The payoff is zero-shot utility: the same pretrained model, conditioned by simple post-training procedures, performs microstructure super-resolution and 2D-to-3D dimensionality expansion on held-out experimental volumes without retraining. If this is right, it establishes a recipe for building generalist generative priors in data-scarce scientific domains rather than requiring million-sample repositories.

What carries the argument

The Local-Global Decomposition (LGD) framework carries the argument: a two-stage generative scheme that decomposes microstructure generation into a global Multi-Output Gaussian Random Field (MOGRF), fixed by target 1- and 2-point statistics, and a local diffusion model that refines neighborhoods. The augmentation step uses the Multi-Output Spectral Mixture Kernel (MOSM) to propose 2000 diverse multi-output covariance fields, sampled by Latin Hypercube Sampling within heuristic parameter bounds, and coordinates five local diffusion models with those global statistics. A modified EDM sampler with replaceable conditioning functions then turns the unconditional PolyMicros prior into task-specific samplers. This decomposition is what lets five examples stand in for a dataset of 30,000.

What would settle it

Train the same augmentation pipeline on five experimental microstructures drawn from a different material class or crystal symmetry, then measure super-resolution and dimensionality-expansion error on a held-out volume from that class; if the pointwise MAPE or target-statistics error degrades by orders of magnitude, the bootstrap does not generalize as claimed.

Watch

Extended reading notes

Core claim

The central claim is that a foundation model for polycrystalline microstructures can be bootstrapped from as few as five experimental volumes by separating global and local structure. Global long-range patterns are represented by two-point spatial statistics generated from a parameterized spectral-mixture kernel with heuristic bounds, while local realism is supplied by five diffusion models, each trained on overlapping patches from one experimental volume. Combinatorial sampling yields 30,000 synthetic volumes of size $128^3$, and training an EDM-style UNet on this dataset produces PolyMicros, a generative prior over polycrystals. The paper demonstrates that this prior, used with masked inpainting or with alternating optimization-and-diffusion sampling, solves microstructure super-resolution (4.05% pointwise MAPE) and 2D-to-3D dimensionality expansion (maximum target-statistics error below $10^{-5}$) with no additional training.

Load-bearing premise

The argument hinges on the heuristic MOSM kernel parameter bounds producing covariances that the five local diffusion models can refine into realistic polycrystals without destroying the requested long-range statistics.

Editorial extensions

If this is right

  • A foundation-model prior for polycrystalline microstructure can be built from single-digit experimental counts, so data-scarce structural materials are no longer ruled out from generalist generative models.
  • The same PolyMicros prior can be reused for other microscopy tasks by swapping conditioning functions, such as denoising or repairing corrupted volumes, as demonstrated by additional inpainting masks in the appendix.
  • Because PolyMicros is an unconditional diffusion prior, it can regularize inverse problems such as microstructure design or property prediction through Bayesian conditioning, without new experimental data.
  • The augmentation recipe transfers to any statistically stationary spatial field, so other scientific domains with scarce observations could apply the same bootstrapping scheme.
  • The held-out results suggest that the prior captures both local grain morphology and long-range statistical patterns, which are precisely the features needed for accelerating 3D experimental microscopy.

Reading between the lines

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

  • The reported numbers are measured against two specific held-out volumes; a harder test would be generalization to an unseen material system whose local neighborhood is absent from the five training sources, a case the paper's own limitation discussion suggests may be constrained.
  • The paper's caveat that augmentation 'cannot extend arbitrarily far' implies a testable prediction: adding more seed experimental volumes should extend the diversity horizon roughly additively, which could be checked with the PCA-variance convergence metric used in the diversity analysis.
  • The success of alternating optimization and diffusion suggests the same prior could be used for matching higher-order statistics, such as three-point correlations, since the conditioning function only needs a differentiable statistic.
  • If the heuristic MOSM parameter bounds are replaced by bounds learned from a larger corpus of experimental covariances, the diversity and physical plausibility of the synthetic dataset could improve beyond what the current trial-and-error bounds achieve.
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

3 major / 6 minor

Summary. The paper proposes PolyMicros, a diffusion-based foundation model for 3D polycrystalline microstructures, trained entirely on a synthetic dataset bootstrapped from only five experimental 3D volumes. The bootstrapping pipeline uses an ensemble of Local-Global Decomposition (LGD) generative models: global 2-point statistics are sampled from a Multi-Output Spectral Mixture (MOSM) kernel parameterized by Latin Hypercube Sampling, and local neighborhood statistics are captured by five diffusion models trained on patches from the five experimental volumes. The two components are combined to generate 30,000 synthetic 128^3 microstructures, which are then used to train PolyMicros. The paper demonstrates zero-shot use of PolyMicros on two tasks: 4x microstructure super-resolution via masked inpainting (4.05% pointwise MAPE on one held-out ferritic steel volume) and 2D-to-3D dimensionality expansion via optimization-based conditioning on three orthogonal 2-point statistics (maximum 2-point statistics error below 1e-5). The authors interpret these results as evidence that a physics-driven augmentation scheme can support a useful generalist prior for data-scarce microstructure science.

Significance. If the central claims hold, this is a valuable contribution to data-efficient scientific machine learning. The core idea—using smaller physics-informed generative models to curate a large synthetic dataset that trains a generalist diffusion prior—addresses a real bottleneck in mesoscale materials science, where experimental 3D microstructure data are extremely scarce. The paper's strengths include concrete quantitative metrics on held-out experimental data, an ablation study of the conditioning schedule in super-resolution, uncertainty estimates from multiple posterior samples, and the stated intention to release models and datasets. The dataset-diversity comparison against MICRO2D, DREAM3D, and an experimental EBSD dataset is also a useful step. However, the current evaluation does not yet establish the breadth of generalization implied by the 'foundation model' framing: each downstream task is demonstrated on essentially a single held-out volume, the dimensionality-expansion metric only measures adherence to the conditioning statistics, and no comparisons are made to existing 2D-to-3D reconstruction methods.

major comments (3)
  1. [§5.2–5.3] The zero-shot claims are supported by exactly one held-out volume per task. The super-resolution study uses the ferritic steel reference [8] (§5.2), and the dimensionality-expansion study appears to use a single reference volume (§5.3 and App. D.2). Since the paper's stated motivation is a generalist prior that transfers across 'a wide range of material classes, manufacturing processes, and local microstructure morphology' (§3), one material system cannot establish such transfer. I recommend evaluating both tasks on multiple held-out experimental volumes spanning different crystal symmetries, grain morphologies, and processing histories, with per-volume errors reported.
  2. [§5.3] The headline dimensionality-expansion metric—maximum 2-point statistics error below 1e-5—measures only how closely the generated volume matches the conditioning statistics, which the conditioning procedure explicitly optimizes. It does not validate the realism of the full 3D structure, and no comparison is made to existing 2D-to-3D methods such as Kench and Cooper [47], DREAM3D-based pipelines [34,35], or the diffusion-based approach of Lee and Yun [50], all of which are cited in App. D.2. The statement that this is 'the first successful attempt at performing the microstructure dimensionality task for polycrystalline materials beyond first-order accuracy' is therefore not supported by the evidence as presented. Please add baseline comparisons and independent realism metrics, for example grain-size distributions, boundary-curvature statistics, or homogenized property predictions.
  3. [§3.1, App. B.1, App. E] The diversity of the synthetic dataset rests on the assumption that the MOSM kernel parameter box, selected through heuristics and 'an aspect of trial and error' (App. B.1), contains a representative set of physically feasible polycrystalline covariances. The paper itself states that the augmentation framework 'cannot extend arbitrarily far' and 'does not extend uniformly' (App. E), and that microstructures far from their seed are 'likely increasingly less plausible.' Because PolyMicros is trained entirely on LGD-generated synthetic volumes, its zero-shot behavior inherits any bias of that generator. I would like to see a direct test of out-of-distribution capability: for example, generate from target 2-point statistics deliberately chosen outside the trained MOSM box, or evaluate on additional experimental volumes whose statistics lie far from the five seeds, and report whether super-resolution and dimensionality-expansion quality degrade. This would speak directly to the weakest load-bearing assumption in the paper.
minor comments (6)
  1. [§5.2] The phrase '4X down-sample' should use a proper multiplication sign and clarify that the down-sampling is along one axis; the abbreviation MAPE should be defined at first use.
  2. [App. D.1] The ablation text says '60 − 100%' with inconsistent spacing and an en dash; this should be cleaned up for readability.
  3. [App. E] There is a typo, 'requirments', that should be corrected to 'requirements'.
  4. [§5.1] The comparison of explained-variance saturation is informative, but the two-dimensional PCA projections in Fig. 3 can obscure overlaps in a high-dimensional statistics space; a quantitative coverage or nearest-neighbor diversity measure would strengthen the diversity claim.
  5. [App. D.2] The term 'in-painting' is hyphenated inconsistently; use a single spelling throughout.
  6. [§1] The capitalization of 'Foundation Models' varies across the abstract and main text; please make it consistent.

Circularity Check

1 steps flagged · score 6.0 of 10

The 2D-to-3D success metric is the conditioning optimizer's own stopping criterion, making that headline result enforced by construction rather than independently predicted.

  1. fitted input called prediction [Sec. 5.3 (Reconstruction from Partial Statistics); App. D.2, Algorithm 5]
    "We achieve a maximum error below 1e−5 on the target statistics across the generated samples. App. D.2 presents comparisons of the full 2-point statistics errors. Overall, this result is incredibly precise, and is the first successful attempt at performing the microstructure dimensionality task for polycrystalline materials beyond first-order accuracy."

    The reported success metric is the objective function of the conditioning procedure, not an independent measurement. Algorithm 5 loops: err ← ∥f̂⊥ − f⊥∥²; while err > ϱ_i: x ← x − α_lr ∇_x err, with threshold schedule ϱ(t) = (N−t)·1e−5 + 1e−7. Thus 'maximum error below 1e−5' is guaranteed by the optimization stopping rule, so the 2-point statistics match is enforced by construction. Presenting this enforced convergence as quantitative confirmation, and as the basis for the 'beyond first-order accuracy' claim, is a fitted input renamed as a prediction: the metric cannot fail under the algorithm and does not independently validate realism or transferability.

full rationale

The PolyMicros pipeline is largely self-contained rather than circular: the synthetic dataset is produced by LGD models from five experimental volumes, PolyMicros is trained on that dataset, and the super-resolution case study is evaluated on a held-out experimental volume [8] that was not used for data generation. The reported 4.05% pointwise MAPE in super-resolution is a genuine held-out prediction, and the diversity analysis in Fig. 3 compares the synthetic dataset against external datasets (HT-EBSD-2D, MICRO2D, DREAM3D), which is independent evidence. The dimensionality-expansion case study, however, contains one concrete circular step: the 'maximum error below 1e−5 on the target statistics' is the stopping criterion of the conditioning optimizer in Algorithm 5, so that particular quantitative claim reduces by construction to the algorithm's own objective. The paper also leans heavily on prior LGD/MOGRF work by the same authors for the core extrapolation assumption, but those citations provide the modeling framework rather than the target result, and the held-out experiments give independent support for the overall utility claim; consequently that self-citation does not by itself raise the score. The score of 6 reflects the one enforced-metric prediction in an otherwise externally evaluated pipeline.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The central claim rests on five main types of free parameters or assumptions. First, the MOSM kernel bounds and the Q choice are tuned heuristically, and the paper itself labels them 'domain expertise as well as an aspect of trial and error'. Second, the LGD skip parameter and the super-resolution mask enforcement fraction are empirically selected. Third, the decomposition of microstructure generation into global Gaussian and local non-Gaussian components is a domain assumption inherited from prior work. Fourth, the assumption that valid polycrystal covariances form a box in MOSM parameter space is explicitly flagged in the paper as likely false in general. Fifth, the evaluations use a single or very few held-out volumes, so the generalization of the reported error numbers is assumed.

free parameters (5)
  • MOSM kernel parameter bounds (A, ς, ω, θ, ϕ, Q) = A elements 1.5-5, ς between -5 and 5, ω between -0.02 and 0.02, θ between -0.5 and 0.5, ϕ between 0 and 2π, Q=4 (132…
    These bounds are not derived from theory but chosen to keep generated samples periodic, within ROGSH value ranges, and qualitatively polycrystalline-like. The paper states the choice of bounds was 'a complex task involving both domain expertise as well as an aspect of trial and error' (App. B.1).
  • LGD sampling skip parameter = skip = 12 diffusion steps out of a 25-step schedule (approximately half the sampling steps)
    The skip parameter controls the balance between adherence to the MOGRF's spatial statistics and realism of the generated samples. It is 'empirically selected' (App. A.3) and affects the realism and statistical fidelity of every synthetic microstructure in the dataset.
  • Fraction of diffusion steps with mask enforcement (super-resolution) = 75% of diffusion steps
    Chosen via a small ablation (60-100% tested) to balance reconstruction accuracy and removal of vertical artifacts. This is a task-level free parameter tuned on the evaluation setup, and it affects the reported 4.05% MAPE.
  • Number of MOSM mixtures Q = 4 mixtures, 132 kernel parameters
    Selected empirically as the minimum Q that can approximate an experimental covariance map. This directly controls the expressiveness and diversity of the generated covariance structures.
  • LHS sampling of 2000 covariance parameter sets = 2000 accepted sets, about 8% rejection rate
    The number 2000 is a design choice, and the rejection rate of about 8% is an empirical post-filtering that removes non-periodic or invalid ROGSH fields. The paper does not specify how much the final dataset depends on this rejection threshold.
assumptions (5)
  • domain assumption Polycrystalline microstructures can be represented as ROGSH fields, and their 1- and 2-point statistics are sufficient to quantify salient differences for generative purposes.
    The entire augmentation pipeline and the downstream conditioning tasks rely on 2-point statistics as the diversity and fidelity measure. The paper cites prior literature (Brown Jr, Torquato, etc.) for the property relevance of these statistics, but the sufficiency of 2-point statistics for generating realistic polycrystals is an assumption.
  • domain assumption The microstructure generation process can be decomposed into a global Gaussian component (MOGRF) and local non-Gaussian corrections (diffusion model), as expressed in Eq. 11.
    This is the core LGD assumption inherited from prior work [69, 15]. The paper assumes the joint distribution factorizes into a Multi-Output Gaussian Random Field and local neighborhood distributions, which is an approximation that may not hold for all polycrystalline morphologies.
  • ad hoc to paper The MOSM kernel parameter box contains a representative and diverse sample of physically feasible polycrystal covariances.
    The paper explicitly acknowledges this is an assumption: 'we are assuming the space of valid polycrystal covariances forms a box in the parameter space of the MOSM kernel' and that this is 'probably not the case for all' (App. B.1). This assumption is central to the claim that the synthetic dataset is physically diverse.
  • domain assumption The spatial statistics and local neighborhood distributions are statistically stationary and periodic.
    Stationarity and periodicity are required for the MOGRF sampling algorithm and for the MOSM kernel to generate periodic covariances. The paper states this as a standard assumption (App. A.4, A.5), and it limits the class of materials and tasks the model can address.
  • domain assumption The experimental reference volumes used for evaluation are representative of the target task distribution.
    The super-resolution and dimensionality expansion evaluations are each performed against a single held-out experimental volume (the ferritic steel from [8] for super-resolution, and one reference for dimensionality expansion). Generalization beyond these volumes is assumed rather than measured.
invented entities (2)
  • PolyMicros foundation model independent evidence
    purpose: A generative prior over polycrystalline microstructures intended to be reused for multiple tasks via post-training conditioning.
    The model is made available publicly, and its value is demonstrated on held-out experimental volumes that were not used in training or data generation. However, the model itself is a learned artifact, not a physical entity, and its generalization to new materials is not yet independently verified by other groups.
  • Synthetic PolyMicros dataset of 30,000 microstructures independent evidence
    purpose: Training corpus for the foundation model, generated by the proposed bootstrapping scheme.
    The dataset is released openly, so others can inspect and evaluate it. But its physical realism is only assessed indirectly through the downstream task performance and PCA-based diversity comparisons.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PolyMicros: Bootstrapping a Foundation Model for Polycrystalline Material Structure." pith.science (2026). https://pith.science/paper/LO5AJ4ZP

@misc{pith2026250611055,
  author       = {Pith},
  title        = {Pith review of: PolyMicros: Bootstrapping a Foundation Model for Polycrystalline Material Structure},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LO5AJ4ZP}},
  note         = {Machine review of arXiv:2506.11055}
}
read the original abstract

Recent advances in Foundation Models for Materials Science are poised to revolutionize the discovery, manufacture, and design of novel materials with tailored properties and responses. Although great strides have been made, successes have been restricted to materials classes where multi-million sample data repositories can be readily curated (e.g., atomistic structures). Unfortunately, for many structural and functional materials (e.g., mesoscale structured metal alloys), such datasets are too costly or prohibitive to construct; instead, datasets are limited to very few examples. To address this challenge, we introduce a novel machine learning approach for learning from hyper-sparse, complex spatial data in scientific domains. Our core contribution is a physics-driven data augmentation scheme that leverages an ensemble of local generative models, trained on as few as five experimental observations, and coordinates them through a novel diversity curation strategy to generate a large-scale, physically diverse dataset. We utilize this framework to construct PolyMicros, the first Foundation Model for polycrystalline materials (a structural material class important across a broad range of industrial and scientific applications). We demonstrate the utility of PolyMicros by zero-shot solving several long standing challenges related to accelerating 3D experimental microscopy. Finally, we make both our models and datasets openly available to the community.

Figures

Figures reproduced from arXiv: 2506.11055 by the authors.

Figure 1
Figure 1. PolyMicros is a Foundation model for polycrystalline materials structure trained on a [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed Statistical Physics driven Data Augmentation algorithm. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Principal Component projections (i.e., PC basis coefficients computed from the 2-point [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: Overview of modular post-training conditioning of the PolyMicros Foundation Model to [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Overview of modular post-training conditioning of the PolyMicros Foundation Model to [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Visualization of the LGD framework’s two stage sampling process. A sample from [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]
Figure 7
Figure 7. Figure 7: Comparisons of an experimental microstructure’s ROGSH auto- and cross-correlations [PITH_FULL_IMAGE:figures/full_fig_p028_7.png]
Figure 8
Figure 8. Figure 8: Ablation testing the level of enforcement of the known values from the mask. After [PITH_FULL_IMAGE:figures/full_fig_p032_8.png]
Figure 9
Figure 9. Figure 9: PolyMicro’s post-training conditioning performance on a variety of potential inpainting [PITH_FULL_IMAGE:figures/full_fig_p033_9.png]
Figure 10
Figure 10. Figure 10: Row One: Target Orthogonal Auto- and Cross-Correlations. Row 2: Average Orthogonal [PITH_FULL_IMAGE:figures/full_fig_p035_10.png]
Figure 11
Figure 11. Figure 11: Example Multi-Output Gaussian Random Field samples. The MOGRF covariance [PITH_FULL_IMAGE:figures/full_fig_p037_11.png]
Figure 12
Figure 12. Figure 12: Six Example Neighborhoods extracted from the AL2219R Experimental Microstructure [PITH_FULL_IMAGE:figures/full_fig_p038_12.png]
Figure 13
Figure 13. Figure 13: Six Example Neighborhoods extracted from the Inconel625AM Experimental Microstruc [PITH_FULL_IMAGE:figures/full_fig_p039_13.png]
Figure 14
Figure 14. Figure 14: Six Example Neighborhoods extracted from the Inconel718W Experimental Microstruc [PITH_FULL_IMAGE:figures/full_fig_p039_14.png]
Figure 15
Figure 15. Figure 15: Six Example Neighborhoods extracted from the NRL Experimental Microstructure [ [PITH_FULL_IMAGE:figures/full_fig_p040_15.png]
Figure 16
Figure 16. Figure 16: Six Example Neighborhoods extracted from the TI64R Experimental Microstructure [ [PITH_FULL_IMAGE:figures/full_fig_p040_16.png]
Figure 17
Figure 17. Figure 17: Example synthetic microstructures (Size of [PITH_FULL_IMAGE:figures/full_fig_p041_17.png]
Figure 18
Figure 18. Figure 18: Example synthetic microstructures (Size of [PITH_FULL_IMAGE:figures/full_fig_p042_18.png]
Figure 19
Figure 19. Figure 19: Example synthetic microstructures (Size of [PITH_FULL_IMAGE:figures/full_fig_p043_19.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Materials Behavior as Mechanism Ensembles: A Probabilistic Framework for Emergent Behaviors

    cond-mat.mtrl-sci 2026-07 conditional novelty 5.0 of 10

    Materials phenomena such as fatigue crack growth are framed as conditional probability landscapes over competing unit mechanisms, to be inferred from multiscale simulation and multimodal data and then optimized toward...

Reference graph

Works this paper leans on

118 extracted references · 47 canonical work pages · cited by 1 Pith paper

  1. [47]

    Kench and S.J

    S. Kench and S.J. Cooper. Generating three-dimensional structures from a two-dimensional slice with generative adversarial network-based dimensionality expansion. Nature Machine Intelligence, 3:299–305, 2021. doi: 10.1038/s42256-021-00322-1

    doi:10.1038/s42256-021-00322-1 2021

  2. [50]

    Multi-plane denoising diffusion-based dimensionality expansion for 2d-to-3d reconstruction of microstructures with harmonized sampling

    Kang-Hyun Lee and Gun Jin Yun. Multi-plane denoising diffusion-based dimensionality expansion for 2d-to-3d reconstruction of microstructures with harmonized sampling. npj Computational Materials, 10(1):99, 2024

    2024

  3. [1]

    Phi-3 technical report: A highly capable language model locally on your phone

    Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadallah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Jianmin Bao, Harkirat Behl, et al. Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219, 2024

    arXiv 2024

  4. [2]

    Phi-4 technical report

    Marah Abdin, Jyoti Aneja, Harkirat Behl, Sébastien Bubeck, Ronen Eldan, Suriya Gunasekar, Michael Harrison, Russell J Hewett, Mojan Javaheripi, Piero Kauffmann, et al. Phi-4 technical report. arXiv preprint arXiv:2412.08905, 2024

    arXiv 2024

  5. [3]

    Adams, S.R

    B.L. Adams, S.R. Kalidindi, and D.T. Fullwood. Microstructure Sensitive Design for Perfor- mance Optimization. Butterworth-Heinemann, Waltham, MA, 2013

    2013

  6. [4]

    Uncertainty propagation in a multiscale calphad-reinforced elastochemical phase-field model

    Vahid Attari, Pejman Honarmandi, Thien Duong, Daniel J Sauceda, Douglas Allaire, and Raymundo Arroyave. Uncertainty propagation in a multiscale calphad-reinforced elastochemical phase-field model. Acta Materialia, 183:452–470, 2020

    2020

  7. [5]

    Universal guidance for diffusion models

    Arpit Bansal, Hong-Min Chu, Avi Schwarzschild, Soumyadip Sengupta, Micah Goldblum, Jonas Geiping, and Tom Goldstein. Universal guidance for diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 843–852, 2023

    2023

  8. [6]

    Barber, J.A

    Z.H. Barber, J.A. Leake, and T.W. Clyne. The doitpoms project: Micrograph library. URL https://www.doitpoms.ac.uk/miclib/index.php

Show all 118 references
  1. [7]

    Elena, Dávid P

    Ilyes Batatia, Philipp Benner, Yuan Chiang, Alin M. Elena, Dávid P. Kovács, Janosh Riebesell, Xavier R. Advincula, Mark Asta, Matthew Avaylon, William J. Baldwin, Fabian Berger, Noam Bernstein, Arghya Bhowmik, Samuel M. Blau, Vlad C ˘arare, James P. Darby, Sandip De, Flaviano ...

  2. [9]

    Bruinsma, Ana Lucic, Megan Stanley, Anna Vaughan, Johannes Brandstetter, Patrick Garvan, Maik Riechert, Jonathan A

    Cristian Bodnar, Wessel P. Bruinsma, Ana Lucic, Megan Stanley, Anna Vaughan, Johannes Brandstetter, Patrick Garvan, Maik Riechert, Jonathan A. Weyn, Haiyu Dong, Jayesh K. Gupta, Kit Thambiratnam, Alexander T. Archibald, Chun-Chieh Wu, Elizabeth Heider, Max Welling, Richard E. ...

    2024 arXiv

  3. [10]

    Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S

    Rishi Bommasani, Drew A. Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S. Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, Erik Brynjolfsson, Shyamal Buch, Dallas Card, Rodrigo Castellon, Niladri Chatterji, Annie Chen, Kathleen Creel, Jare...

  4. [11]

    Multi-task gaussian process prediction

    Edwin V Bonilla, Kian Chai, and Christopher Williams. Multi-task gaussian process prediction. Advances in neural information processing systems, 20, 2007

    2007

  5. [12]

    Brodnik, C

    N.R. Brodnik, C. Muir, N. Tulshibagwale, J. Rossin, M.P. Echlin, C.M. Hamel, S.L.B. Kramer, T.M. Pollock, J.D. Kiser, C. Smith, and S.H. Daly. Perspective: Machine learning in experimen- tal solid mechanics. Journal of the Mechanics and Physics of Solids, 173:105231, 2023. doi...

    2023

  6. [13]

    Language models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901, 2020

    1901

  7. [14]

    Texture Analysis in Materials Science: Mathematic Methods

    H-J Bunge. Texture Analysis in Materials Science: Mathematic Methods. Butterworth & Co., Berlin, 1982

    1982

  8. [15]

    Statistically conditioned polycrystal generation using denoising diffusion models

    Michael O Buzzy, Andreas E Robertson, and Surya R Kalidindi. Statistically conditioned polycrystal generation using denoising diffusion models. Acta Materialia, page 119746, 2024

    2024

  9. [16]

    Active learning for the design of polycrystalline textures using conditional normalizing flows

    Michael O Buzzy, David Montes de Oca Zapiain, Adam P Generale, Surya R Kalidindi, and Hojun Lim. Active learning for the design of polycrystalline textures using conditional normalizing flows. Acta Materialia, 284:120537, 2025

    2025

  10. [17]

    Afrl additive manufacturing modeling series: challenge 4, 3d reconstruction of an in625 high-energy diffraction microscopy sample using multi-modal serial sectioning

    Michael G Chapman, Megna N Shah, Sean P Donegan, J Michael Scott, Paul A Shade, David Menasche, and Michael D Uchic. Afrl additive manufacturing modeling series: challenge 4, 3d reconstruction of an in625 high-energy diffraction microscopy sample using multi-modal serial secti...

    2021

  11. [18]

    Cheng, Y

    S. Cheng, Y . Jiao, and Y . Ren. Data-driven learning of 3-point correlation functions as mi- crostructure representations. Acta Materialia, 229:117800, 2022. doi: 10.1016/j.actamat.2022. 117800

    2022 doi

  12. [19]

    Nucleotide transformer: building and evaluating robust foundation models for human genomics

    Hugo Dalla-Torre, Liam Gonzalez, Javier Mendoza-Revilla, Nicolas Lopez Carranza, Adam Henryk Grzywaczewski, Francesco Oteri, Christian Dallago, Evan Trop, Bernardo P de Almeida, Hassan Sirelkhatim, et al. Nucleotide transformer: building and evaluating robust foundation models...

    2024

  13. [20]

    DeCost and Elizabeth A

    Brain L. DeCost and Elizabeth A. Holm. A large dataset of synthetic sem images of powder materials and their ground truth 3d structures. Data in Brief, 9:727–731, 2016. doi: 10.1016/j. dib.2016.10.011

    2016 doi

  14. [21]

    DeCost and Elizabeth A

    Brian L. DeCost and Elizabeth A. Holm. A computer vision approach for automated analysis and classification of microstructural images. Computational Materials Science, 110:126–133,

  15. [22]

    DeCost, Matthew Hecht, Toby Francis, Bryan A

    Brian L. DeCost, Matthew Hecht, Toby Francis, Bryan A. Webler, Yoosuf N. Picard, and Elizabeth Holm. Uhcsdb: Ultra high carbon steel micrograph database. Integrating Materials and Manufacturing Innovation, 6:197–205, 2017. doi: 10.1007/s40192-017-0097-0. 12

    2017 doi

  16. [23]

    Bert: Pre-training of deep bidirectional transformers for language understanding, 2019

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding, 2019. URL https://arxiv.org/ abs/1810.04805

    2019 arXiv

  17. [24]

    Dimiduk, E.A

    D.M. Dimiduk, E.A. Holm, and S.R. Niezgoda. Perspective on the impact of machine learning, deep learning, and artificial intelligence on materials, processes, and structures engineering. Integrating Materials and Manufacturing Innovation , 7:157–172, 2018. doi: 10.1007/s40192-...

    2018 doi

  18. [25]

    Dureth, P

    C. Dureth, P. Seibert, D. Rucker, S. Handford, M. Kastner, and M. Gude. Conditional diffusion- based microstructure reconstruction. ArXiV, 2022

    2022

  19. [26]

    Falco, J

    S. Falco, J. Jiang, F. De Cola, and N. Petrinic. Generation of 3d polycrystalline microstructures with a conditioned laguerre-voronoi tessellation technique. Computational Materials Science, 136:20–28, 2017. doi: 10.1016/j.commatsci.2017.04.018

    2017 doi

  20. [27]

    High-throughput microstructural characterization and process correlation using automated electron backscatter diffraction

    J Elliott Fowler, Timothy J Ruggles, Dale E Cillessen, Kyle L Johnson, Luis J Jauregui, Robert L Craig, Nathan R Bianco, Amelia A Henriksen, and Brad L Boyce. High-throughput microstructural characterization and process correlation using automated electron backscatter diffract...

    2024

  21. [28]

    Fullwood, B.L

    D.T. Fullwood, B.L. Adams, and S.R. Kalidindi. A strong contrast homogenization formulation for multi-phase anistropic materials. Journal of the Mechanics and Physics of Solids , 56: 2287–2297, 2008

    2008

  22. [29]

    Gao and Y

    Y . Gao and Y . Liu. Relibaility-based topology optimization with stochastic heterogeneous microstructure properties. Materials and Design, 2021. doi: 10.1016/j.matdes.2021.109713

    2021

  23. [30]

    A bayesian approach to designing microstructures and processing pathways for tailored material properties

    Adam Generale, Conlain Kelly, Grayson Harrington, Andreas Robertson, Michael Buzzy, and Surya Kalidindi. A bayesian approach to designing microstructures and processing pathways for tailored material properties. In AI for Accelerated Materials Design - NeurIPS 2023 Workshop,

    2023

  24. [31]

    Generale and S.R

    A.P. Generale and S.R. Kalidindi. Reduced-order models for microstructure-sensitive effective thermal conductivity of woven ceramic matrix composites with residual porosity. Composite Structures, 274:114399, 2021. doi: 10.1016/j.compstruct.2021.114399

    2021

  25. [32]

    Generale, A.E

    A.P. Generale, A.E. Robertson, C. Kelly, and S.R. Kalidindi. Inverse stochastic microstructure design. SSRN: Preprint, 2023. doi: 10.2139/ssrn.4590691

    2023 doi

  26. [33]

    Watmus: Wavelet transformation-induced multi-time scaling for accelerating fatigue simulations at multiple spatial scales

    Somnath Ghosh, Jinlei Shen, Shravan Kotha, and Pritam Chakraborty. Watmus: Wavelet transformation-induced multi-time scaling for accelerating fatigue simulations at multiple spatial scales. Integrating Materials and Manufacturing Innovation, 10:568–587, 2021

    2021

  27. [34]

    Groeber and M.A

    M.A. Groeber and M.A. Jackson. Dream.3d: A digital representation environment for the analysis of microstructure in 3d. Integrating Materials and Manufacturing Innovation, 3:56–72,

  28. [35]

    Groeber, S

    M.A. Groeber, S. Ghosh, M.D. Uchic, and D.M. Dimiduk. A framework for automated analysis and simulation of 3d polycrystalline microstructures. part 2: Synthetic microstructure generation. Acta Materialia, 56:1274–1287, 2008. doi: 10.1016/j.actamat.2007.11.040

    2008 doi

  29. [36]

    Hashemi and S.R

    S. Hashemi and S.R. Kalidindi. A machine learning framework for the temporal evolution of microstructure during static recrystallization of polycrystalline materials simulated by cellular automaton. Computational Materials Science, 188:110132, 2021. doi: 10.1016/j.commatsci. 2...

    2021

  30. [37]

    Kalidindi

    Sepideh Hashemi and Surya R. Kalidindi. Gaussian process autoregression models for the evo- lution of polycrystalline microstructures subjected to arbitrary stretching tensors. International Journal of Plasticity, 162:103532, 2023. doi: 10.1016/j.ijplas.2023.103532

    2023

  31. [38]

    Ho and T

    J. Ho and T. Salimans. Classifier-free diffusion generation. ArXiV, 2022. doi: 10.48550/arXiv. 2207.12598. 13

    2022 doi arXiv

  32. [39]

    J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic models. NeurIPS, 2020

    2020

  33. [40]

    Terramind: Large-scale generative multimodality for earth observation

    Johannes Jakubik, Felix Yang, Benedikt Blumenstiel, Erik Scheurer, Rocco Sedona, Ste- fano Maurogiovanni, Jente Bosmans, Nikolaos Dionelis, Valerio Marsocci, Niklas Kopp, et al. Terramind: Large-scale generative multimodality for earth observation. arXiv preprint arXiv:2504.11...

    2025 arXiv

  34. [41]

    Q-rbsa: high-resolution 3d ebsd map generation using an efficient quaternion transformer network

    Devendra K Jangid, Neal R Brodnik, McLean P Echlin, Chandrakanth Gudavalli, Connor Levenson, Tresa M Pollock, Samantha H Daly, and BS Manjunath. Q-rbsa: high-resolution 3d ebsd map generation using an efficient quaternion transformer network. npj Computational Materials, 10(1)...

    2024

  35. [42]

    Titanium 3d microstructure for physics-based generative models: a dataset and primer

    Devendra Kumar Jangid, Neal R Brodnik, McLean P Echlin, Samantha Daly, Tresa Polloc, and BS Manjunath. Titanium 3d microstructure for physics-based generative models: a dataset and primer. In 1st Workshop on the Synergy of Scientific and Machine Learning Modeling@ ICML

  36. [43]

    Brown Jr

    W.F. Brown Jr. Solid mixture permittivities. Journal of Chemical Physics, 23:1514–1517, 1955

    1955

  37. [44]

    Highly accurate protein structure prediction with alphafold

    John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ron- neberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin Žídek, Anna Potapenko, et al. Highly accurate protein structure prediction with alphafold. nature, 596(7873):583–589, 2021

    2021

  38. [45]

    Microstruc- ture design using machine learning generated low dimensional and continuous design space

    Jaimyun Jung, Jae Ik Yoon, Hyung Keun Park, Hyeontae Jo, and Hyoung Seop Kim. Microstruc- ture design using machine learning generated low dimensional and continuous design space. Materialia, 11:100690, 2020. doi: https://doi.org/10.1016/j.mtla.2020.100690

    2020

  39. [46]

    Elucidating the design space of diffusion-based generative models, 2022

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models, 2022

    2022

  40. [48]

    net/, 2023

    https://openreview. net/, 2023

    2023

  41. [49]

    Sidharth Krishnamoorthi, Ritwik Bandyopadhyay, and Michael D. Sangid. A microstructure- based fatigue model for additively manufactured ti-6al-4v, including the role of prior β bound- aries. International Journal of Plasticity, 163:103569, 2023. doi: https://doi.org/10.1016/j....

    2023

  42. [51]

    Towards an astronomical foundation model for stars with a transformer-based model

    Henry W Leung and Jo Bovy. Towards an astronomical foundation model for stars with a transformer-based model. Monthly Notices of the Royal Astronomical Society , 527(1): 1494–1520, October 2023. ISSN 1365-2966. doi: 10.1093/mnras/stad3015. URL http: //dx.doi.org/10.1093/mnras/stad3015

    2023 doi

  43. [52]

    Score-based diffusion models in function space

    Jae Hyun Lim, Nikola B Kovachki, Ricardo Baptista, Christopher Beckham, Kamyar Az- izzadenesheli, Jean Kossaifi, Vikram V oleti, Jiaming Song, Karsten Kreis, Jan Kautz, et al. Score-based diffusion models in function space. arXiv preprint arXiv:2302.07400, 2023

    2023 arXiv

  44. [53]

    Epting, Noel T

    Rubayyat Mahbub, Tim Hsu, William K. Epting, Noel T. Nuhfer, Gregory A. Hackett, Harry Abernathy, Anthony D. Rollett, Marc De Graef, Shawn Litster, and Paul A. Salvador. A method for quantitative 3d mesoscale analysis of solid oxide fuel cell microstructures using xe-plasma fo...

    2017

  45. [54]

    Neural operator: Learning maps between function spaces with applications to pdes

    Nikola Kovachki, Zongyi Li, Burigede Liu, Kamyar Azizzadenesheli, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Learning maps between function spaces with applications to pdes. Journal of Machine Learning Research, 24(89):1–97, 2023

    2023

  46. [55]

    McDowell

    David L. McDowell. Viscoplasticity of heterogeneous metallic materials.Materials Science and Engineering: R: Reports, 62(3):67–123, 2008. doi: https://doi.org/10.1016/j.mser.2008.04.003

    2008 doi

  47. [56]

    C. Meng, Y . He, Y . Song, J. Song, J. Wu, J. Zhu, and S. Ermon. Sdedit: Guided image synthesis and editing with stochastic differential equations. ArXiv, 2022

    2022

  48. [57]

    4m: Massively multimodal masked modeling

    David Mizrahi, Roman Bachmann, Oguzhan Kar, Teresa Yeo, Mingfei Gao, Afshin Dehghan, and Amir Zamir. 4m: Massively multimodal masked modeling. Advances in Neural Information Processing Systems, 36:58363–58408, 2023

    2023

  49. [58]

    The well: a large-scale collection of diverse physics simulations for machine learning

    Ruben Ohana, Michael McCabe, Lucas Meyer, Rudy Morel, Fruzsina Agocs, Miguel Beneitez, Marsha Berger, Blakesly Burkhart, Stuart Dalziel, Drummond Fielding, et al. The well: a large-scale collection of diverse physics simulations for machine learning. Advances in Neural Informa...

    2024

  50. [59]

    Integrating neural operators with diffusion models improves spectral representation in turbulence modeling

    Vivek Oommen, Aniruddha Bora, Zhen Zhang, and George Em Karniadakis. Integrating neural operators with diffusion models improves spectral representation in turbulence modeling. arXiv preprint arXiv:2409.08477, 2024

    2024 arXiv

  51. [60]

    Marshall and S.R

    A. Marshall and S.R. Kalidindi. Autonomous development of a machine-learning model for the plastic response of two-phase composites from micromechanical finite element models. JOM, 73:2085–2095, 2021. doi: 10.1007/s11837-021-04696-w

    2021 doi

  52. [61]

    Spectral mixture kernels for multi-output gaussian processes,

    Gabriel Parra and Felipe Tobar. Spectral mixture kernels for multi-output gaussian processes,

  53. [62]

    Paulson, M.W

    N.H. Paulson, M.W. Priddy, D.L. McDowell, and S.R. Kalidindi. Reduced-order structure- property linkages for polycrystalline microstructures based on 2-point statistics. Acta Mater., 129:428, 2017. doi: 10.1016/j.actamat.2017.03.009

    2017 doi

  54. [63]

    Gen- erating 3d images of material microstructures from a single 2d image: a denoising diffusion approach

    Johan Phan, Muhammad Sarmad, Leonardo Ruspini, Gabriel Kiss, and Frank Lindseth. Gen- erating 3d images of material microstructures from a single 2d image: a denoising diffusion approach. Scientific Reports, 14(1):6498, 2024

    2024

  55. [64]

    Pilchak, Jared Shank, Joseph C

    Adam L. Pilchak, Jared Shank, Joseph C. Tucker, Shesh Srivatsa, Patrick N. Fagin, and S. Lee Semiatin. A dataset for the development, verification, and validation of microstructure-sensitive process models for near-alpha titanium alloys. Integrating Materials and Manufacturing...

    2016 doi

  56. [65]

    Prasad, N

    M.R.G. Prasad, N. Vajragupta, and A. Hartmaier. Kanapy: A python package for generating complex synthetic polycrystalline microstructures. Journal of Open Source Software, 4:1732,

  57. [66]

    Astroclip: a cross-modal foundation model for galaxies

    Liam Parker, Francois Lanusse, Siavash Golkar, Leopoldo Sarra, Miles Cranmer, Alberto Bietti, Michael Eickenberg, Geraud Krawezik, Michael McCabe, Rudy Morel, et al. Astroclip: a cross-modal foundation model for galaxies. Monthly Notices of the Royal Astronomical Society, 531(...

    2024

  58. [67]

    Metasynth: Meta-prompting-driven agentic scaffolds for diverse synthetic data generation

    Haris Riaz, Sourav Bhabesh, Vinayak Arannil, Miguel Ballesteros, and Graham Horwood. Metasynth: Meta-prompting-driven agentic scaffolds for diverse synthetic data generation. arXiv preprint arXiv:2504.12563, 2025

    2025

  59. [68]

    Robertson and S.R

    A.E. Robertson and S.R. Kalidindi. Efficient generation of n-field microstructures from 2-point statistics using multi-output gaussian random fields. Acta Materialia, 232:117927, 2022. doi: 10.1016/j.actamat.2022.117927

    2022

  60. [69]

    Robertson, Conlain Kelly, Michael Buzzy, and Surya R

    Andreas E. Robertson, Conlain Kelly, Michael Buzzy, and Surya R. Kalidindi. Local–global decompositions for conditional microstructure generation. Acta Materialia, 253:118966, 2023. ISSN 1359-6454. doi: https://doi.org/10.1016/j.actamat.2023.118966. 15

    2023

  61. [70]

    Micro2d: A large, statistically diverse, heterogeneous microstructure dataset.Integrating Materials and Manufacturing Innovation, pages 1–35, 2024

    Andreas E Robertson, Adam P Generale, Conlain Kelly, Michael O Buzzy, and Surya R Ka- lidindi. Micro2d: A large, statistically diverse, heterogeneous microstructure dataset.Integrating Materials and Manufacturing Innovation, pages 1–35, 2024

    2024

  62. [71]

    A monte carlo model for 3d grain evolution during welding

    Theron M Rodgers, John A Mitchell, and Veena Tikare. A monte carlo model for 3d grain evolution during welding. Modelling and Simulation in Materials Science and Engineering, 25 (6):064006, 2017

    2017

  63. [72]

    Rossin, P

    J. Rossin, P. Leser, K. Pusch, C. Frey, S.C. V ogel, A.I. Saville, C. Torbet, A.J. Clarke, S. Daly, and T.M. Pollock. Single crystal elastic constants of additively manufactured components determined by resonant ultrasound spectroscopy. Materials Characterization, 192:112244,

  64. [73]

    Roters, P

    F. Roters, P. Eisenlohr, L. Hantcherli, D.D. Tjahjanto, T.R. Bieler, and D. Raabe. Overview of constitutive laws, kinematics, homogenization and multiscale methods in crystal plasticity finite element modeling: Theory, experiments, applications. Acta Materialia, 58:1152, 2010

    2010

  65. [74]

    Derivative-enhanced deep operator network

    Yuan Qiu, Nolan Bridges, and Peng Chen. Derivative-enhanced deep operator network. Ad- vances in Neural Information Processing Systems, 37:20945–20981, 2024

    2024

  66. [75]

    Safdari, M

    M. Safdari, M. Baniassadi, H. Garmestani, and M.S. Al-Haik. A modified strong-constrast ex- pansion for estimating the effective thermal conductivity of multiphase heterogeneous materials. Journal of Applied Physics, 112:114318, 2012

    2012

  67. [76]

    Generalized spectral kernels, 2015

    Yves-Laurent Kom Samo and Stephen Roberts. Generalized spectral kernels, 2015. URL https://arxiv.org/abs/1506.02236

    2015 arXiv

  68. [77]

    Seibert, M

    P. Seibert, M. Ambati, A. Rabloff, and M. Kastner. Reconstructing random heterogeneous media through differentiable optimization. Computational Materials Science , 196:110455,

  69. [78]

    Seibert, A

    P. Seibert, A. Rabloff, M. Ambati, and M. Kastner. Descriptor-based reconstruction of three- dimensional microstructures through gradient-based optimization. Acta Materialia, 227:117667,

  70. [79]

    Seibert, M

    P. Seibert, M. Husert, M.P. Wollner, K.A. Kalina, and M. Kastner. Fast reconstruction of microstructures with ellipsoidal inclusions using analytic descriptors. ArXiv, 2023. doi: 10.48550/arXiv.2306.08316

    2023 doi arXiv

  71. [80]

    Two-stage 2d-to-3d reconstruction of realistic microstructures: Implementation and numerical validation by effective properties

    Paul Seibert, Alexander Raßloff, Karl A Kalina, Joachim Gussone, Katrin Bugelnig, Martin Diehl, and Markus Kästner. Two-stage 2d-to-3d reconstruction of realistic microstructures: Implementation and numerical validation by effective properties. Computer Methods in Applied Mech...

    2023

  72. [81]

    doi: 10.1016/j.matchar.2022.112244

    2022

  73. [82]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. 2023

    2023

  74. [83]

    Three-dimensional analysis of grain topology and interface curvature in a β-titanium alloy

    DJ Rowenhorst, AC Lewis, and G Spanos. Three-dimensional analysis of grain topology and interface curvature in a β-titanium alloy. Acta Materialia, 58(16):5511–5519, 2010

    2010

  75. [84]

    Provable probabilistic imaging using score-based generative priors

    Yu Sun, Zihui Wu, Yifan Chen, Berthy T Feng, and Katherine L Bouman. Provable probabilistic imaging using score-based generative priors. IEEE Transactions on Computational Imaging, 2024

    2024

  76. [85]

    Torquato

    S. Torquato. Effective stiffness tensor of composite media: 1. exact series expansions. Journal of the Mechanics and Physics of Solids, 45:1421–1448, 1997

    1997

  77. [86]

    Torquato

    S. Torquato. Random Heterogeneous Materials. Springer, New York, NY , 2002. 16

    2002

  78. [87]

    Turner and S

    D. Turner and S. R. Kalidindi. Statistical construction of 3d microstructures from 2-d examplars collected on oblique sections. Acta Materialia, 102:136–148, 2016. doi: 10.1016/j.actamat. 2015.09.011

    2016 doi

  79. [88]

    The mechanistic origins of heterogeneous void growth during ductile failure

    MW Vaughan, H Lim, B Pham, R Seede, AT Polonsky, KL Johnson, and PJ Noell. The mechanistic origins of heterogeneous void growth during ductile failure. Acta Materialia, 274: 119977, 2024

    2024

  80. [89]

    doi: 10.1016/j.actamat.2022.117667

    2022

  81. [90]

    Wilson and R.P

    A.G. Wilson and R.P. Adams. Gaussian process kernels for pattern discovery and extrapolation. In Proceedings of the 30th International Conference on Machine Learning , volume 28 of Proceedings of Machine Learning Research, pages 1067–1075. PMLR, 2013

    2013

  82. [91]

    Target-aware molecule generation for drug design using a chemical language model

    Yingce Xia, Kehan Wu, Pan Deng, Renhe Liu, Yuan Zhang, Han Guo, Yumeng Cui, Qizhi Pei, Lijun Wu, Shufang Xie, Si Chen, Xi Lu, Song Hu, Jinzhi Wu, Chi-Kin Chan, Shuo Chen, Liangliang Zhou, Nenghai Yu, Haiguang Liu, Jinjiang Guo, Tao Qin, and Tie-Yan Liu. Target-aware molecule g...

    2024

  83. [92]

    A noise robust method based on completed local binary patterns for hot-rolled steel strip surface defects

    Kechen Song and Yunhui Yan. A noise robust method based on completed local binary patterns for hot-rolled steel strip surface defects. Applied Surface Science, 285P:858–864, 2013. doi: 10.1016/j.apsusc.2013.09.002

    2013 doi

  84. [93]

    Pdeformer: Towards a foundation model for one-dimensional partial differential equations,

    Zhanhong Ye, Xiang Huang, Leheng Chen, Hongsheng Liu, Zidong Wang, and Bin Dong. Pdeformer: Towards a foundation model for one-dimensional partial differential equations,

  85. [94]

    Multi-modal dataset of a polycrystalline metallic material: 3d microstructure and deformation fields

    Jean-Charles Stinville, JM Hestroffer, Marie-Agathe Charpagne, AT Polonsky, MP Echlin, CJ Torbet, Valéry Valle, KE Nygren, MP Miller, Ottmar Klaas, et al. Multi-modal dataset of a polycrystalline metallic material: 3d microstructure and deformation fields. Scientific Data, 9 (...

    2022

  86. [95]

    Zelaia, J

    P.F. Zelaia, J. Cheng, J.R. Mayeur, A.K. Ziabari, and M.M. Kirka. Digital polycrystalline microstructure generation using diffusion probabilistic models. SSRN, 2023. doi: 10.2139/ssrn. 4419461

    2023 doi

  87. [96]

    A generative model for inorganic materials design

    Claudio Zeni, Robert Pinsler, Daniel Zügner, Andrew Fowler, Matthew Horton, Xiang Fu, Zilong Wang, Aliaksandra Shysheya, Jonathan Crabbé, Shoko Ueda, et al. A generative model for inorganic materials design. Nature, pages 1–3, 2025

    2025

  88. [97]

    process- driven

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3836–3847, 2023. 17 Appendices A Background 19 A.1 Previous Efforts in Big Dataset Desig...

    2023

  89. [100]

    Vlassis and WaiChing Sun

    Nikolaos N. Vlassis and WaiChing Sun. Denoising diffusion algorithm for inverse design of microstructures with fine-tuned nonlinear material properties.Computer Methods in Applied Me- chanics and Engineering, 413:116126, 2023. doi: https://doi.org/10.1016/j.cma.2023.116126

    2023

  90. [103]

    A triple junction energy study using an inclination-dependent anisotropic monte carlo potts grain growth model

    Lin Yang, Vishal Yadav, Joseph Melville, Joel B Harley, Amanda R Krause, and Michael R Tonks. A triple junction energy study using an inclination-dependent anisotropic monte carlo potts grain growth model. Materials & Design, 239:112763, 2024

    2024

  91. [106]

    Vasilakos, and Thippa Reddy Gadekallu

    Gokul Yenduri, Ramalingam M, Chemmalar Selvi G, Supriya Y , Gautam Srivastava, Praveen Kumar Reddy Maddikunta, Deepti Raj G, Rutvij H Jhaveri, Prabadevi B, Weizheng Wang, Athanasios V . Vasilakos, and Thippa Reddy Gadekallu. Generative pre-trained transformer: A comprehensive ...

    2023 arXiv

  92. [110]

    (22) Importantly the EDM formulation identifies key values for various hyper parameters needed both during training and sampling which we adopt without modification [46]

    (20) The denoiser then samples from a data distribution through its relationship to the score via integration of the probability flow ODE: dx = ˙σ(t)σ(t)∇x log p(x; σ(t))dt, (21) where ∇x log p(x; σ) = (D(x; σ) − x)/σ2. (22) Importantly the EDM formulation identifies key value...

  93. [111]

    Our intent is then to produce a diverse sampling of of covariances by sampling within these bounds

    In practice we aim to convert these guidelines into bounds on the parameters which govern the MOSM Kernel, such that parameter sets within these bounds lead to meaningful covariance structures. Our intent is then to produce a diverse sampling of of covariances by sampling with...

  94. [112]

    A simple way to induce periodicity is to enforce that these statistics decay before the edge of the domain, such that the boundary is a homogeneous constant value

    Gaussian mixture covariances A: These covariances control the anisotropic decay of the spatial statistics as the spatial vector r moves away from the meanς. A simple way to induce periodicity is to enforce that these statistics decay before the edge of the domain, such that th...

  95. [113]

    We limit the means to be between -5 and 5

    Gaussian mixture means ς: The mean term influences the frequency content of the correlations, here we choose low absolute values to search only the low frequency space. We limit the means to be between -5 and 5

  96. [114]

    Our goal is to introduce correlations such that the sampled fields do not have values outside of -1 to 1 as the un- symmetrized ROGSH cannot exceed these values

    Weights ω: The weights control the magnitude of the statistics. Our goal is to introduce correlations such that the sampled fields do not have values outside of -1 to 1 as the un- symmetrized ROGSH cannot exceed these values. For a symmetrized system the ROGSH values will be s...

  97. [115]

    Delays θ: The delays result in a spatial offset in the cross correlations, here we limit them to −0.5 to 0.5 such that the delay can be significant (with respect to a domain of -1 to 1), but will not push correlations beyond the edge of the domain and break periodicity

  98. [116]

    Phase ϕ: The phase shift is within a cosine term and is naturally bounded between 0 and 2π

  99. [117]

    We briefly investigate the effect of this complexity qualita- 3Please see App

    Number of mixtures Q: The number of mixtures determines the overall complexity of the generated correlation structure. We briefly investigate the effect of this complexity qualita- 3Please see App. A.4 for the distinction between 2-Point statistics and covariances 4At this poi...

    2000

  100. [118]

    The output is then the synthetically generated dataset D

    G = 2000 global statistics Bglobal, 2) L = 5 local diffusion models Dlocal, 3) as well as two hyperparameters, the oversampling parameter R and the skip parameter. The output is then the synthetically generated dataset D. We chose R = 3 such that three samples of the MOGRF wer...

    2000

  101. [119]

    We observed that the performance of this conditional diffusion process depended heavily on the order of Alg

    to convert the unconditional diffusion process into the specific conditional diffusion processes needed to address super-resolution. We observed that the performance of this conditional diffusion process depended heavily on the order of Alg. 2. It is critical to perform condit...

  102. [2014]

    doi: 10.1186/2193-9772-3-5

    doi

  103. [2015]

    doi: 10.1016/j.commatsci.2015.08.011

    2015 doi

  104. [2017]

    URL https://arxiv.org/abs/1709.01298

    arXiv

  105. [2019]

    doi: 10.21105/joss.01732

    doi

  106. [2021]

    doi: 10.1016/j.commatsci.2021.110455

    2021

  107. [2022]

    URL https://arxiv.org/abs/2108.07258

    arXiv

  108. [2023]

    URL https://openreview.net/forum?id=zZPICTs5gB

  109. [2024]

    URL https://arxiv.org/abs/2401.00096

    arXiv

  110. [2025]

    URL https://arxiv.org/abs/2402.12652

    arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.