Pith. sign in

REVIEW 3 major objections 6 minor 65 references

A Decision Transformer Approach to Grain Boundary Network Optimization

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A Decision Transformer trained on human gameplay can solve grain boundary network design problems at solution quality comparable to simulated annealing, but with three orders of magnitude fewer iterations, and it transfers to a…

desk verdict Solid, novel application of Decision Transformers to grain boundary network optimization, but the central efficiency and quality claims rest on an unablated manual-rotation oracle. read the letter →

arxiv 2412.15393 v1 pith:3XBFSGOJ submitted 2024-12-19 cond-mat.mtrl-sci physics.comp-ph

Christopher W. Adair , Oliver K. Johnson This is my paper · ORCID
classification cond-mat.mtrl-sciphysics.comp-ph
keywords grainboundarynetworkDecisionTransformermicrostructuredesignhumancomputationgamesimulatedannealingstructure-propertymodeleffectivediffusivitysequencemodeling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that a Decision Transformer trained on human player trajectories from a microstructure design video game can become an efficient solver for grain boundary network design. It reports that the trained model reaches solutions of quality comparable to simulated annealing (about 92 percent of SA's value) while using roughly three orders of magnitude fewer iterations, and that it generalizes to a higher-fidelity constitutive model without retraining. A sympathetic reader would care because high-dimensional grain boundary network design is otherwise hard to optimize directly, and human-in-the-loop data is expensive; a learned policy that distills human strategies into a fast, transferable optimizer would make such design practical.

What carries the argument

The load-bearing machinery is the Decision Transformer architecture adapted to grain graph states: inputs are interleaved state (quaternion grain orientations), return (normalized effective diffusivity $D_\mathrm{eff}$), and one-hot actions over grains, with a block causal mask that lets all grains attend to each other's concurrent actions, and a Laplacian position encoding built from the grain boundary network's weighted Laplacian that carries connectivity information into the transformer. At prediction time the model outputs action-type probabilities and an expected return, and a flat return bias of 0.1 is added to steer it toward expert-like trajectories; the manual rotation action is implemented separately by a local gradient ascent that optimizes the properties of the selected grain's connections. This combination of learned grain selection and action choice plus a hand-coded local rotation is what carries the reported quality and efficiency.

What would settle it

Run the trained policy on the evaluation microstructures with the manual-rotation action replaced by random rotations or by a different local update rule, such as a fixed small rotation or gradient ascent on a mismatched constitutive model, while keeping grain selection and action-type choices fixed; if solution quality drops to near random-search levels, the central claim that the learned decision strategy carries the optimization collapses.

Watch

Extended reading notes

Core claim

The central discovery is that sequence modeling of human optimization trajectories works as a materials design optimizer: a Decision Transformer, trained on 897 trajectories of grain-orientation decisions collected from video game players, predicts the next grain to act on and the action type with 84 percent validation accuracy and produces effective diffusivity values that are on average 92 percent as high as simulated annealing solutions, while needing far fewer decision steps. The same trained model, evaluated with a different, higher-fidelity grain boundary diffusivity model and never retrained, still matches SA's relative quality and actually improves its efficiency advantage over SA. The authors also show that the model performs comparably on microstructures whose grain counts lie outside the training range, which they take as evidence that the learned decisions capture generalizable structure-property relationships rather than memorized trajectories.

Load-bearing premise

When the model chooses a manual rotation, the rotation itself is not predicted by the learned model but is supplied by a hand-coded local gradient-ascent routine, so the reported solution quality assumes that routine faithfully reproduces how human players rotated grains.

Editorial extensions

If this is right

  • After training, the learned policy can replace slow stochastic global search for grain boundary network design problems, since it reaches SA-quality solutions with orders of magnitude fewer expensive model evaluations.
  • A policy trained on a computationally cheap toy constitutive model can be deployed on a more expensive high-fidelity model without retraining, which matters when the target model has no training data of its own.
  • Because the model uses all crystallographic orientations and long-range connectivity through Laplacian position encoding, its decisions can reflect non-local grain boundary network effects rather than only nearest-neighbor properties.
  • The model appears to handle grain counts outside its training distribution, suggesting the approach could extend to statistically relevant volume element sizes used in microstructure design.

Reading between the lines

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

  • A direct test of where the credit lies would be to ablate the hand-coded local gradient ascent, for example by replacing it with random rotations while keeping the learned grain and action selection fixed; the drop in quality would separate the learned policy's contribution from the helper routine's.
  • The same training pipeline should transfer to other grain boundary properties such as corrosion resistance or thermal transport, provided a suitable constitutive structure-property model and a human-playable score exist.
  • The attention weights could be mined as a hypothesis generator for physical motifs: grains with high attention despite not being nearest neighbors may point to network-level controls on effective diffusivity, a claim the paper hints at but does not establish.
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 a Decision Transformer (DT) model trained on human gameplay trajectories from the grain-boundary-network design game "Operation: Forge the Deep." The task is to maximize the effective hydrogen diffusivity of a polycrystal by sequentially rotating individual grain orientations. The model receives per-grain quaternion states, a normalized return, and per-grain one-hot actions, with Laplacian and time position encodings. The authors compare the resulting trajectories with human players and simulated annealing (SA) on four held-out microstructures under a simple "Linear" diffusivity model, then evaluate the same model without retraining under the higher-fidelity Borisov/BRK constitutive model. They report 84% validation accuracy on held-out player decisions, solution quality about 92% of SA, large efficiency gains over SA, and generalization to a different constitutive model and to microstructure sizes outside the training range.

Significance. If the central claims are substantiated, the work would be a valuable demonstration that a sequence-modeling agent can distill human optimization heuristics for a high-dimensional microstructure design problem and transfer them across physics models. The paper has notable strengths: the data and code are publicly available, the problem formulation is clearly tied to a physical homogenization model, the evaluation uses held-out microstructures and multiple trained replicates, and the comparison against SA is a meaningful external baseline. However, the current evidence supports a hybrid controller rather than the trained policy alone, because the continuous rotation step is supplied by a hand-coded local optimizer. An ablation quantifying the marginal contribution of the learned policy is needed before the paper's headline claims can be accepted as stated.

major comments (3)
  1. [Section 3.4.9, Tables 3 and 4] The evaluation protocol replaces the manual-rotation action with a hand-coded local gradient ascent. As stated in Section 3.4.9, the DT predicts only the action type and the selected grain; for a manual-rotation action, the actual orientation change is computed by an external algorithm that is not learned and is not part of the trained model. Since the rotation is the decision variable of the optimization, the reported solution quality (Tables 3 and 4), the efficiency gains, and the cross-model transfer to Borisov/BRK all characterize a hybrid system. The paper does not report a control in which the same local-gradient oracle is paired with random grain selection, nor an ablation that prevents the model from choosing manual rotation. Without such an ablation, the claim that the Decision Transformer itself learns to solve GBN design problems is not established. The 84% validation accuracy is also partially inherited from this setup: the model is only asked to reproduce the action label, not the continuous rotation that the players actually performed.
  2. [Abstract, Section 6, Tables 3 and 4] The abstract and conclusions state that the ML model "requires three orders of magnitude fewer iterations" than SA, but the manuscript does not report the total number of iterations used by either method, nor the ratios. The reported quantities in Tables 3 and 4 are differences in steps (e.g., 126, 4861, 2492, and 1427 for the best model on the Linear model), which are not consistent with a 1000-fold improvement. Moreover, the median ML model is sometimes slower than SA (Table 3: -26, -596, and -86 steps for the 10-, 25-, and 30-grain cases; Table 4: -411, -774, and -753 steps for the 15- and 25-grain cases). The efficiency claim should be supported by reporting the actual step counts and their ratios, or the claim should be qualified to refer to the best model and specific conditions.
  3. [Section 5.1.1, Figures 8 and 9] The claim of generalization to microstructure sizes outside the training range is supported only by visual comparison of return trajectories. The retraining experiments that removed small- or large-grain trajectories are not accompanied by quantitative metrics such as final normalized return, step-to-threshold, or comparison with SA. Without numerical reporting, the assertion that performance is "comparable" to the fully trained model (Figures 8 and 9) is not verifiable. Please provide quantitative results for these ablations.
minor comments (6)
  1. [Sections 3.3 and 3.4.8] The number of collected trajectories is given as 879 in Section 3.3 but 897 in Section 3.4.8; please reconcile these numbers.
  2. [Section 3.1.3] The text contains typographical errors: "Dirichelt" and "Dirchlet" should be "Dirichlet."
  3. [Figure 9 caption] The caption refers to the "Bulatov/BRK model," while the rest of the manuscript uses "Borisov/BRK"; please make the naming consistent.
  4. [Section 3.4.9] The justification for modeling manual rotations as local gradient ascent is based on qualitative observation of player behavior. Since the same local-gradient algorithm also exists as a separate action in the action space, the two actions collapse at evaluation time; this should be discussed explicitly when interpreting action-level accuracy and policy behavior.
  5. [Section 3.2] The SA parameters (Cauchy schedule constants, initial temperature, number of steps, and stopping criteria) are not fully specified. Reporting these would improve reproducibility, especially because the efficiency comparison is central to the paper's claims.
  6. [Section 3.4.3] The choice to use the first 4 Laplacian eigenvectors is stated but not justified; since this truncation is a modeling assumption that could affect generalization, a brief rationale or sensitivity check would be helpful.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Decision Transformer is validated against held-out player decisions and an external simulated-annealing baseline, and the flagged manual-rotation oracle is a correctness/attribution concern rather than a circular reduction.

full rationale

The paper's central derivation chain is self-contained with respect to its claims: the ML model is trained on human player trajectories and evaluated against held-out player trajectories and an external simulated annealing baseline, with solution quality measured by the normalized effective diffusivity that is not encoded into the model except as the return label. The return normalization (Section 3.4.1) is an explicit upper-bound scaling definition, and the inference bias of 0.1 (Section 3.4.9) is a stated heuristic justified by the distribution of top-player step counts; neither is fitted to the target solutions. Self-citations to the prior game [23,43] and to the homogenization model [5,7] supply data and constitutive equations, but the current head-to-head comparisons with SA and the held-out generalization tests do not reduce to those citations. The closest thing to a concern is the manual-rotation oracle in Section 3.4.9: 'Rather than training the ML model to predict exactly what manual rotation would be taken for a given step, we instead model it' as a local gradient ascent. This means the continuous orientation update is not learned, and an ablation (for example, random grain selection with the same oracle) would be needed to quantify the learned policy's marginal contribution. That is a correctness and attribution limitation, not a circular step: no parameter is fitted to the reported output, and no equation defining the target result is reused as the prediction. Therefore no circular reduction can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central result rests on the correctness of the homogenization model, the constitutive models, and the quality of the human player data; no new physical entities are introduced. The main free parameters are the inference-time return bias and the unspecified parameters of the manual-rotation oracle and SA baseline, which shape the reported performance.

free parameters (4)
  • Inference-time return bias = 0.1
    Added to the expected return at prediction time to bias the model toward expert-like trajectories (Section 3.4.9). Chosen from the distribution of top players' step counts; directly shapes all ML evaluation trajectories.
  • Manual rotation local gradient ascent parameters = not reported
    The manual rotation action is implemented as a local gradient ascent on the selected grain's boundary properties; its step size, iteration count, or stopping criterion are not specified (Section 3.4.9). These parameters affect the orientation changes applied and thus the final solution quality.
  • SA Cauchy annealing schedule parameters = not reported
    The simulated annealing baseline uses a Cauchy schedule with unspecified starting temperature, cooling rate, and step distribution (Section 3.2). The efficiency comparison between ML and SA depends on this tuning.
  • Linear model scaling beta = 10^7
    Arbitrary scaling constant in the toy constitutive model (Eq. 1); cancels in normalized returns, so not load-bearing, but listed for completeness.
assumptions (5)
  • domain assumption The homogenization model of Johnson et al. correctly computes effective diffusivity of a GBN from the mesh and boundary diffusivities.
    Used in Eq. 2-4 to define the objective; the paper relies on [57] for the relation without re-deriving it.
  • domain assumption The Borisov/BRK model predicts hydrogen diffusivity in nickel grain boundaries.
    Adopted from [52] as the high-fidelity constitutive model; the generalization test assumes this model is the target of interest.
  • domain assumption Player trajectories from the video game contain transferable optimization heuristics.
    The entire training signal is human decisions; if these are noisy or non-representative, the learned policy inherits that.
  • ad hoc to paper The first 4 graph Laplacian eigenvectors capture the relevant connectivity information for decision-making.
    Position encoding uses the first 4 eigenvectors (Section 3.4.3); no justification for why 4 is sufficient.
  • ad hoc to paper The action space of 5 actions is sufficient to express useful optimization moves.
    The model can only choose among do-nothing, manual rotation, local ascent, undo, random orientation (Section 3.4.1); any better move must be expressible as one of these.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Decision Transformer Approach to Grain Boundary Network Optimization." pith.science (2026). https://pith.science/paper/3XBFSGOJ

@misc{pith2026241215393,
  author       = {Pith},
  title        = {Pith review of: A Decision Transformer Approach to Grain Boundary Network Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3XBFSGOJ}},
  note         = {Machine review of arXiv:2412.15393}
}
read the original abstract

As microstructure property models improve, additional information from crystallographic degrees of freedom and grain boundary networks (GBNs) can be included in microstructure design problems. However, the high dimensional nature of including this information precludes the use of many common optimization approaches and requires less efficient methods to generate quality designs. Previous work demonstrated that human-in-the-loop optimization, instantiated as a video game, achieved high-quality, efficient solutions to these design problems. However, such data is expensive to obtain. In the present work, we show how a Decision Transformer machine learning (ML) model can be used to learn from the optimization trajectories generated by human players, and subsequently solve materials design problems. We compare the ML optimization trajectories against players and a common global optimization algorithm: simulated annealing (SA). We find that the ML model exhibits a validation accuracy of 84% against player decisions, and achieves solutions of comparable quality to SA (92%), but does so using three orders of magnitude fewer iterations. We find that the ML model generalizes in important and surprising ways, including the ability to train using a simple constitutive structure-property model and then solve microstructure design problems for a different, higher-fidelity, constitutive structure-property model without any retraining. These results demonstrate the potential of Decision Transformer models for the solution of materials design problems.

Figures

Figures reproduced from arXiv: 2412.15393 by the authors.

Figure 1
Figure 1. A visual description of how the GBN design problem was formulated in a video game context. (a) Shows an [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Borisov/BRK diffusivity model for H in Ni GBs. Subplots show the GB diffusivity as a function of the GB [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Block diagram of the GBN design Decision Trans [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Loss and accuracy values during training of the [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Time history of returns for each method (ML model, players, and SA) using the Linear constitutive model [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Distribution of problem sizes (nGrains) in the training data for the ML model. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Time history of returns for the ML model and SA using the Borisov/BRK constitutive model on each of the 4 [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Trajectories of a model that is only trained on [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: Trajectories of a model that is only trained on [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 41 canonical work pages

  1. [1]

    jun Liu, G

    D. jun Liu, G. Tian, G. feng Jin, W. Zhang, Y. hong Zhang, Characterization of local- ized corrosion pathways in 2195-T8 Al–Li alloys exposed to acidic solution, Defence Technology 25 (2023) 152–165. URL: http: //dx.doi.org/10.1016/j.dt.2022.05.004. doi:10.1016/j.dt.2022.05.004. 19

    work page doi:10.1016/j.dt.2022.05.004 2023

  2. [2]

    Sayet, B

    J. Sayet, B. O. Hoch, A. Oudriss, J. Bouhat- tate, X. Feaugas, Multi-scale approach to hydrogen diffusion in FCC polycrystalline structure with binary classification of grain boundaries in continuum model, Materi- als Today Communications 34 (2023) 105021. URL: https://doi.org/10.1016/j.mtcomm. 2022.105021. doi: 10.1016/j.mtcomm.2022. 105021

    arXiv 2023

  3. [3]

    Bechtle, M

    S. Bechtle, M. Kumar, B. P. Somerday, M. E. Launey, R. O. Ritchie, Grain-boundary engi- neering markedly reduces susceptibility to in- tergranular hydrogen embrittlement in metal- lic materials, Acta Materialia 57 (2009) 4148–4157. URL: http://dx.doi.org/10. 1016/j.actamat.2009.05.012. doi:10.1016/ j.actamat.2009.05.012

    work page 2009

  4. [4]

    B. D. Snow, S. G. Baird, C. Kurniawan, D. T. Fullwood, E. R. Homer, O. K. John- son, Grain boundary structure-property model inference using polycrystals: The un- derdetermined case, Acta Materialia 209 (2021) 116769. URL: https://doi.org/10. 1016/j.actamat.2021.116769. doi:10.1016/ j.actamat.2021.116769

    work page arXiv 2021

  5. [5]

    O. K. Johnson, J. M. Lund, T. R. Critch- field, Spectral graph theory for characteri- zation and homogenization of grain boundary networks, Acta Materialia 146 (2018) 42–54. doi:10.1016/j.actamat.2017.11.054

    work page doi:10.1016/j.actamat.2017.11.054 2018

  6. [6]

    W. Feng, Z. Wang, Q. Sun, Y. He, Y. Sun, Effect of thermomechanical processing via rotary swaging on grain boundary charac- ter distribution and intergranular corrosion in 304 austenitic stainless steel, Journal of Materials Research and Technology 19 (2022) 2470–2482. URL: https://doi.org/ 10.1016/j.jmrt.2022.06.032. doi:10.1016/ j.jmrt.2022.06.032

    work page doi:10.1016/j.jmrt.2022.06.032 2022

  7. [7]

    T. R. Critchfield, O. K. Johnson, Repre- sentative and statistical volume elements for grain boundary networks: A stereological ap- proach, Acta Materialia 188 (2020) 166–180. doi:10.1016/j.actamat.2019.12.029

    work page doi:10.1016/j.actamat.2019.12.029 2020

  8. [8]

    Kamaya, A procedure for estimating Young’s modulus of textured polycrys- talline materials, International Jour- nal of Solids and Structures 46 (2009) 2642–2649

    M. Kamaya, A procedure for estimating Young’s modulus of textured polycrys- talline materials, International Jour- nal of Solids and Structures 46 (2009) 2642–2649. URL: http://dx.doi.org/ 10.1016/j.ijsolstr.2009.02.013. doi:10.1016/j.ijsolstr.2009.02.013

    work page doi:10.1016/j.ijsolstr.2009.02.013 2009

Show all 65 references
  1. [9]

    A. V. Sergueeva, N. A. Mara, A. K. Mukherjee, Universal features of grain boundary networks in FCC materials, Journal of Materials Science 40 (2005) 847–852. URL: http://www.springerlink. com/index/10.1007/s10853-006-0697-0 . doi:10.1007/s10853-006-0697-0

    2005 doi

  2. [10]

    Watanabe, S

    T. Watanabe, S. Tsurekawa, Control of brittleness and development of desirable me- chanical properties in polycrystalline systems by grain boundary engineering, Acta Ma- terialia 47 (1999) 4171–4185. doi: 10.1016/ S1359-6454(99)00275-X

    1999

  3. [11]

    Zheng, X.-g

    H. Zheng, X.-g. Li, R. Tran, C. Chen, M. Horton, D. Winston, K. Aslaug, S. Ping, Acta Materialia Grain boundary properties of elemental metals, Acta Materialia 186 (2020) 40–49. URL: https://doi.org/10. 1016/j.actamat.2019.12.030. doi:10.1016/ j.actamat.2019.12.030

    2020

  4. [12]

    Tcherdyntsev, A

    V. Tcherdyntsev, A. Rodin, The Algorithm to Predict the Grain Boundary Diffusion in Non- Dilute Metallic Systems, Materials 16 (2023). doi:10.3390/ma16041431

    2023 doi

  5. [13]

    S. Hao, H. Li, Effect of twin grain bound- ary on the diffusion of Cu in bulk β-Sn, Computational Materials Science 226 (2023) 112200. URL: https://doi.org/10.1016/ j.commatsci.2023.112200. doi: 10.1016/j. commatsci.2023.112200

    2023

  6. [14]

    Oudriss, J

    A. Oudriss, J. Creus, J. Bouhattate, E. Con- forto, C. Berziou, C. Savall, X. Feau- gas, Grain size and grain-boundary ef- fects on diffusion and trapping of hydrogen in pure nickel, Acta Materialia 60 (2012) 6814–6828. URL: http://dx.doi.org/10. 20 1016/j.actamat.2012.09.004....

    2012

  7. [15]

    Kontis, E

    P. Kontis, E. Chauvet, Z. Peng, J. He, A. K. da Silva, D. Raabe, C. Tassin, J. J. Blandin, S. Abed, R. Dendievel, B. Gault, G. Mar- tin, Atomic-scale grain boundary engineer- ing to overcome hot-cracking in additively- manufactured superalloys, Acta Materialia 177 (2019) 209–2...

    2019 arXiv

  8. [16]

    X. Meng, Z. Liu, B. Cui, D. Qin, H. Geng, W. Cai, L. Fu, J. He, Z. Ren, J. Sui, Grain Boundary Engineering for Achieving High Thermoelectric Performance in n-Type Skutterudites, Advanced Energy Materials 7 (2017). doi:10.1002/aenm.201602582

    2017 doi

  9. [17]

    J. Guo, Q. Lou, Y. Qiu, Z. Y. Wang, Z. H. Ge, J. Feng, J. He, Remarkably enhanced ther- moelectric properties of Bi2S3 nanocompos- ites via modulation doping and grain bound- ary engineering, Applied Surface Science 520 (2020) 146341. URL: https://doi.org/10. 1016/j.apsusc.202...

    2020

  10. [18]

    T. Fast, M. Knezevic, S. R. Kalidindi, Ap- plication of microstructure sensitive design to structural components produced from hexag- onal polycrystalline metals, Computational Materials Science 43 (2008) 374–383. doi: 10. 1016/j.commatsci.2007.12.002

    2008

  11. [19]

    B. B. Zhang, Y. G. Tang, Q. S. Mei, X. Y. Li, K. Lu, Inhibiting creep in nanograined alloys with stable grain boundary networks, Science 378 (2022) 659–663. doi: 10.1126/science. abq7739

    2022 doi

  12. [20]

    Krakow, R

    R. Krakow, R. J. Bennett, D. N. Johnstone, Z. Vukmanovic, W. Solano-Alvarez, S. J. Lain´ e, J. F. Einsle, P. A. Midgley, C. M. Rae, R. Hielscher, On three-dimensional misorien- tation spaces, Proceedings of the Royal Soci- ety A: Mathematical, Physical and Engineer- ing Scienc...

    2017

  13. [21]

    E. R. Homer, S. Patala, J. L. Priede- man, Grain Boundary Plane Orienta- tion Fundamental Zones and Structure- Property Relationships, Scientific Re- ports 5 (2015) 15476. URL: http: //www.nature.com/articles/srep15476. doi:10.1038/srep15476

    2015 doi

  14. [22]

    Lecchini-Visintini, J

    A. Lecchini-Visintini, J. Lygeros, J. Ma- cIejowski, Simulated annealing: Rigor- ous finite-time guarantees for optimization on continuous domains, Advances in Neu- ral Information Processing Systems 20 - Proceedings of the 2007 Conference (2009). arXiv:0709.2989

    2009 arXiv

  15. [23]

    C. W. Adair, H. Evans, E. Beatty, D. L. Hansen, S. Holladay, O. K. Johnson, Mi- crostructure design using a human com- putation game, Materialia 25 (2022) 101544. URL: https://doi.org/10.1016/ j.mtla.2022.101544. doi: 10.1016/j.mtla. 2022.101544

    2022

  16. [24]

    A. J. Quinn, B. B. Bederson, Human Compu- tation: A Survey and Taxonomy of a Grow- ing Field, in: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 2011, pp. 1403–1412. doi: 10.1145/ 1978942.1979148

    2011

  17. [25]

    J. H. M. Jensen, M. Gajdacz, S. Z. Ahmed, J. H. Czarkowski, C. Weidner, J. Rafner, J. J. Sørensen, K. Mølmer, J. F. Sherson, Crowd- sourcing human common sense for quan- tum control, Physical Review Research 3 (2021) 1–23. doi: 10.1103/PhysRevResearch. 3.013057. arXiv:2004.03296

    2021 arXiv

  18. [26]

    Khatib, F

    F. Khatib, F. Dimaio, S. Cooper, M. Kazmier- czyk, M. Gilski, S. Krzywda, H. Zabran- ska, I. Pichova, J. Thompson, Z. Popovi´ c, M. Jaskolski, D. Baker, Crystal structure of a monomeric retroviral protease solved by pro- tein folding game players, Nature Structural and Molecul...

    2010 doi

  19. [27]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, N. Houlsby, AN IM- AGE IS WORTH 16X16 WORDS: TRANS- FORMERS FOR IMAGE RECOGNITION AT SCALE, in: ICLR 2021 - 9th Inter- national Confe...

    2021 arXiv

  20. [28]

    K. H. Lee, O. Nachum, M. Yang, L. Lee, D. Freeman, W. Xu, S. Guadarrama, I. Fis- cher, E. Jang, H. Michalewski, I. Mordatch, Multi-Game Decision Transformers, in: Ad- vances in Neural Information Processing Sys- tems, volume 35, 2022. arXiv:2205.15241

    2022 arXiv

  21. [29]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkor- eit, L. Jones, A. N. Gomez, L. Kaiser, I. Polo- sukhin, Attention is all you need, in: Ad- vances in Neural Information Processing Sys- tems, volume 2017-Decem, 2017, pp. 5999–

    2017

  22. [30]

    L. Chen, K. Lu, A. Rajeswaran, K. Lee, A. Grover, M. Laskin, P. Abbeel, A. Srinivas, I. Mordatch, Decision Transformer: Reinforce- ment Learning via Sequence Modeling, in: Advances in Neural Information Processing Systems, volume 18, 2021, pp. 15084–15097. arXiv:2106.01345

    2021 arXiv

  23. [31]

    Chefer, S

    H. Chefer, S. Gur, L. Wolf, Transformer In- terpretability Beyond Attention Visualization, Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (2021) 782–791. doi: 10.1109/ CVPR46437.2021.00084. arXiv:2012.09838

    2021

  24. [32]

    M. Bhan, N. Achache, V. Legrand, A. Blangero, N. Chesneau, Evaluating Self-attention Interpretability Through Human-Grounded Experimental Protocol, in: Communications in Computer and Informa- tion Science, volume 1903 CCIS, 2023, pp. 26–46. doi: 10.1007/978-3-031-44070-0_2 . a...

    1903 arXiv

  25. [33]

    S. Fu, Y. Arinicheva, C. H¨ uter, M. Finster- busch, R. Spatschek, Grain Boundary Char- acterization and Potential Percolation of the Solid Electrolyte LLZO, Batteries 9 (2023) 1–

    2023

  26. [34]

    doi:10.3390/batteries9040222

    doi

  27. [35]

    Cheng, J

    X. Cheng, J. Zheng, J. Lu, Y. Li, P. Yan, Y. Zhang, Realizing superior cycling stabil- ity of Ni-Rich layered cathode by combina- tion of grain boundary engineering and sur- face coating, Nano Energy 62 (2019) 30–37. URL: https://doi.org/10.1016/j.nanoen. 2019.05.021. doi: 10....

    2019 doi

  28. [36]

    T. P. Swiler, V. Tikare, E. A. Holm, Heteroge- neous diffusion effects i n polycrystalline, Ma- terials Science and Engineering A238 (1997) 85–93

    1997

  29. [37]

    C. W. Adair, O. K. Johnson, Materialia Char- acterizing grain boundary network length fea- tures through a harmonic representation, Ma- terialia 35 (2024) 102121. URL: https://doi. org/10.1016/j.mtla.2024.102121. doi: 10. 1016/j.mtla.2024.102121

    2024

  30. [38]

    V. V. Bulatov, B. W. Reed, M. Ku- mar, Grain boundary energy function for fcc metals, Acta Materialia 65 (2014) 161–175. URL: http://dx.doi. org/10.1016/j.actamat.2013.10.057. doi:10.1016/j.actamat.2013.10.057

    2014 doi

  31. [39]

    R. Quey, P. R. Dawson, F. Barbe, Large- scale 3D random polycrystals for the finite element method: Generation, meshing and remeshing, Computer Methods in Applied Mechanics and Engineering 200 (2011) 1729–

    2011

  32. [40]

    Krause, J

    M. Krause, J. Smeddinck, Human computa- tion games: A survey, European Signal Pro- cessing Conference (2011) 754–758

    2011

  33. [41]

    von Ahn, Games with a Purpose, Computer 39 (2006) 92–94

    L. von Ahn, Games with a Purpose, Computer 39 (2006) 92–94. URL: http://ieeexplore. ieee.org/document/1642623/. doi:10.1109/ MC.2006.196. 22

    2006

  34. [42]

    Kawrykow, G

    A. Kawrykow, G. Roumanis, A. Kam, D. Kwak, C. Leung, C. Wu, E. Zarour, L. Sar- menta, M. Blanchette, J. Waldisp¨ uhl, Phylo: A citizen science approach for improving mul- tiple sequence alignment, PLoS ONE 7 (2012). doi:10.1371/journal.pone.0031362

    2012 doi

  35. [43]

    von Ahn, L

    L. von Ahn, L. Dabbish, Designing games with a purpose, Communications of the ACM 51 (2008) 57. URL: http://portal.acm. org/citation.cfm?doid=1378704.1378719. doi:10.1145/1378704.1378719

    2008

  36. [44]

    V. P. Dwivedi, X. Bresson, A Generalization of Transformer Networks to Graphs, AAAI (2021). URL: http://arxiv.org/abs/2012. 09699. arXiv:2012.09699

    2021 arXiv

  37. [45]

    C. W. Adair, Operation: Forge the Deep,

  38. [46]

    Maskey, A

    S. Maskey, A. Parviz, M. Thiessen, H. St¨ ark, Y. Sadikaj, H. Maron, Generalized Lapla- cian Positional Encoding for Graph Represen- tation Learning, in: NeurIPS Workshop on Symmetry and Geometry in Neural Represen- tations, 2022. URL: http://arxiv.org/abs/ 2210.15956. arXiv:2...

    2022 arXiv

  39. [47]

    W. Park, W. Chang, D. Lee, J. Kim, S.-w. Hwang, GRPE: Relative Posi- tional Encoding for Graph Transformer (2022). URL: http://arxiv.org/abs/2201. 12787. arXiv:2201.12787

    2022 arXiv

  40. [48]

    Mialon, D

    G. Mialon, D. Chen, M. Selosse, J. Mairal, GraphiT: Encoding Graph Structure in Trans- formers (2021) 1–20. URL: http://arxiv. org/abs/2106.05667. arXiv:2106.05667

    2021 arXiv

  41. [49]

    L. G. Harrison, Influence of dislocations on diffusion kinetics in solids with particular ref- erence to the alkali halides, Trans. Fara- day Soc. 57 (1961) 1191–1199. URL: http:// dx.doi.org/10.1039/TF9615701191. doi:10. 1039/TF9615701191

    1961 doi

  42. [50]

    S. Dube, J. Gorimbo, M. Moyo, C. G. Okoye- Chine, X. Liu, Synthesis and application of ni- based membranes in hydrogen separation and purification systems: A review, Journal of En- vironmental Chemical Engineering 11 (2023). doi:10.1016/j.jece.2022.109194

    2023

  43. [51]

    Ramp´ aˇ sek, M

    L. Ramp´ aˇ sek, M. Galkin, V. P. Dwivedi, A. T. Luu, G. Wolf, D. Beaini, Recipe for a General, Powerful, Scalable Graph Trans- former, in: Advances in Neural Information Processing Systems, volume 35, 2022, pp. 1–

    2022

  44. [52]

    D. E. Page, K. F. Varela, O. K. Johnson, D. T. Fullwood, E. R. Homer, Measur- ing simulated hydrogen diffusion in symmet- ric tilt nickel grain boundaries and examin- ing the relevance of the borisov relationship for individual boundary diffusion, Acta Ma- terialia 212 (2021)....

    2021

  45. [53]

    Borisov, V

    V. Borisov, V. Golikov, G. Scherbedinskiy, Re- lation between diffusion coefficients and grain boundary energy, Phys. Met. Metall 17 (1964) 881–885

    1964

  46. [54]

    Luczak, M

    J. Luczak, M. Lieder, Nickel-based cata- lysts for electrolytic decomposition of ammo- nia towards hydrogen production, Advances in Colloid and Interface Science 319 (2023). doi:10.1016/j.cis.2023.102963

    2023

  47. [55]

    Mohles, 3-d front tracking model for in- terfaces with anisotropic energy, Computa- tional Materials Science 176 (2020) 109534

    V. Mohles, 3-d front tracking model for in- terfaces with anisotropic energy, Computa- tional Materials Science 176 (2020) 109534. URL: https://linkinghub.elsevier.com/ retrieve/pii/S0927025620300252. doi: 10. 1016/j.commatsci.2020.109534. 23

    2020

  48. [56]

    S. K. Naghibzadeh, Z. Xu, D. Kinderlehrer, R. Suter, K. Dayal, G. S. Rohrer, Impact of grain boundary energy anisotropy on grain growth, Physical Review Materials 8 (2024). doi:10.1103/PhysRevMaterials.8.093403

    2024 doi

  49. [57]

    Schweizer, A

    P. Schweizer, A. Sharma, L. Peth¨ o, E. Huszar, L. M. Vogl, J. Michler, X. Maeder, Atomic scale volume and grain boundary diffu- sion elucidated by in situ stem, Na- ture Communications 14 (2023). doi:10.1038/ s41467-023-43103-7

    2023

  50. [58]

    Dukkipati, M

    A. Dukkipati, M. N. Murty, S. Bhatnagar, Cauchy annealing schedule: An annealing schedule for Boltzmann selection scheme in evolutionary algorithms, Proceedings of the 2004 Congress on Evolutionary Computation, CEC2004 1 (2004) 55–62. doi: 10.1109/cec. 2004.1330837. arXiv:0408055

    2004 arXiv

  51. [59]

    Dwarampudi, N

    M. Dwarampudi, N. V. S. Reddy, Ef- fects of padding on LSTMs and CNNs (2019). URL: http://arxiv.org/abs/1903. 07288. arXiv:1903.07288

    2019 arXiv

  52. [60]

    Kurniawan, S

    C. Kurniawan, S. Baird, D. T. Fullwood, E. R. Homer, O. K. Johnson, Grain boundary structure–property model infer- ence using polycrystals: the overdeter- mined case, Journal of Materials Science 55 (2020) 1562–1576. URL: https://doi. org/10.1007/s10853-019-04125-z . doi: 10. ...

    2020 doi

  53. [61]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Decoupled weight de- cay regularization, 7th International Confer- ence on Learning Representations, ICLR 2019 (2019). arXiv:1711.05101

    2019 arXiv

  54. [62]

    Adair, O

    C. Adair, O. Johnson, A-Decision- Transformer-Approach-to-Grain-Boundary- Network-Optimization, 2024. URL: https://github.com/ChrisAdair/ A-Decision-Transformer-Approach-to-Grain-Boundary-Network-Optimization . 24

    2024

  55. [63]

    Lee, Mathematical Analysis and Per- formance Evaluation of the GELU Activa- tion Function in Deep Learning, Journal of Mathematics 2023 (2023)

    M. Lee, Mathematical Analysis and Per- formance Evaluation of the GELU Activa- tion Function in Deep Learning, Journal of Mathematics 2023 (2023). doi:10.1155/2023/ 4229924

    2023 doi

  56. [1745]

    cma.2011.01.002

    URL: http://dx.doi.org/10.1016/j. cma.2011.01.002. doi:10.1016/j.cma.2011. 01.002

    2011 doi

  57. [2022]

    com/app/2053100/Operation_Forge_the_ Deep/

    URL: https://store.steampowered. com/app/2053100/Operation_Forge_the_ Deep/

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.