Pith. sign in

REVIEW 4 major objections 4 minor 32 references

Transfer Learning for Deep Learning-based Prediction of Lattice Thermal Conductivity

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a two-stage transfer-learning schedule substantially improves deep-learning predictions of lattice thermal conductivity on small, precise datasets, while degrading performance on narrow-range datasets.

desk verdict Two-stage transfer learning gives a big MAPE drop on broad LTC data, but the unreported train/test overlap with the AFLOW pretraining set leaves the generalizability claim unproven. read the letter →

arxiv 2411.18259 v1 pith:66FZ4IUC submitted 2024-11-27 cs.LG physics.comp-ph

L. Klochko , M. d'Aquin , A. Togo , L. Chaput This is my paper · ORCID
classification cs.LGphysics.comp-ph
keywords transferlearninglatticethermalconductivitydeepMEGNetfine-tuningAFLOWAGLmaterialsdiscoveryParAIsite
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 claims that a two-stage transfer-learning schedule materially improves deep-learning predictions of lattice thermal conductivity on small, high-precision datasets: first fine-tune a pre-trained graph neural network on a large dataset of cheap approximate conductivity values, then fine-tune again on the small precise datasets. On the broad-range Dataset1 the validation mean absolute percentage error (MAPE) drops from 0.82 when trained from scratch and 0.76 with standard transfer to 0.34 with the double transfer; the mixed dataset MIX improves from 0.83 to 0.69. The same schedule harms the narrow-range half-Heusler Dataset2 (MAPE rises from 0.42 to 0.78), because the model adapted to a wide range of conductivity values and overfits the specialized set. The authors conclude that double transfer learning helps when the target dataset spans a broad range of materials and values, and can hinder it when the dataset is narrow.

What carries the argument

The central object is ParAIsite: a pre-trained graph neural network (MEGNet, trained on formation energies of 62,315 compounds) whose last hidden layer feeds a three-layer fully connected head of 350 neurons per layer, trained with MAPE loss. The carrying mechanism is a three-step training protocol: (1) train from random weights, (2) fine-tune the pre-trained MEGNet, and (3) fine-tune the Step-2 model first on the large AFLOW AGL approximate dataset, then on the small precise datasets. The AFLOW pre-fine-tuning step is what transfers a broad sense of how structure relates to conductivity, and the paper credits it for the large drop in error on datasets with wide LTC ranges.

What would settle it

Hold out a new set of 50 to 100 materials whose precise DFT anharmonic LTC values are computed after the paper, spanning several structure families and a wide range of conductivities; if a Step-3 model does not clearly beat a Step-2 model on that held-out set, the double transfer does not generalize as claimed.

Watch

Extended reading notes

Core claim

The paper's central claim is that a model (ParAIsite) built from the pre-trained graph network MEGNet with a three-layer MLP head achieves its best precision and generalization when it is first fine-tuned on the 5,578-material AFLOW AGL dataset of quasi-harmonic Debye-Grüneisen approximate conductivities and then fine-tuned again on smaller datasets of DFT-based values. The authors report that this second round of transfer reduces validation MAPE on Dataset1 from 0.76 to 0.34 and on MIX from 0.81 to 0.69, while making training converge faster and show less overfitting; the same procedure raises Dataset2's MAPE from 0.42 to 0.78, showing the limit of the approach when the target dataset is narrow in LTC range and material diversity.

Load-bearing premise

The approach assumes that the 5,578 approximate AFLOW AGL conductivities carry enough of the same physical signal as the precise DFT anharmonic values that learning on them first makes the model better at the precise task.

Editorial extensions

If this is right

  • For small datasets with a broad property range, a cheap approximate dataset can serve as a stepping stone that lets a model learn precise values from few examples.
  • The same protocol is a template for other materials properties where precise ab initio data are scarce but cheap approximations exist.
  • Transfer helps only when the target distribution is wide; applying it to narrow, specialized datasets can hurt, so practitioners should check range and diversity before adopting double fine-tuning.
  • The validated prediction on (BaSbO3)2 suggests the method can flag low-conductivity candidates at scale for later ab initio confirmation.

Reading between the lines

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

  • The gains on Dataset1 are large but reported on a single split protocol with a fixed seed; whether they survive on a wider, independently computed benchmark set is the paper's open question.
  • The failure on Dataset2 suggests the double transfer acts like a prior over the conductivity range; a conditional or range-aware transfer could avoid the negative transfer.
  • If AFLOW AGL values were reweighted or filtered by estimated reliability, the transfer might improve further.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes ParAIsite, a transfer-learning pipeline for predicting lattice thermal conductivity (LTC) from crystal structures. It attaches an MLP head to MEGNet and compares three training protocols: training from scratch (Step 1), fine-tuning a formation-energy-pretrained MEGNet (Step 2), and a two-stage transfer in which the model is first fine-tuned on 5,578 approximate AFLOW AGL LTC values and then fine-tuned on the target datasets (Step 3). The experiments use two small high-quality datasets (96 materials in Dataset1 and 143 half-Heuslers in Dataset2), a combined MIX dataset, and the large approximate AFLOW AGL dataset. Table I reports validation MAPE averaged over nine random 80/20 splits. The headline result is that Step 3 lowers validation MAPE on Dataset1 (0.82 to 0.34) and on MIX (0.83 to 0.69) but worsens it on Dataset2 (0.42 to 0.78). The authors conclude that double transfer learning improves precision, generalizability, and overfitting behavior for diverse datasets, and they report a single external ab initio check on (BaSbO3)2.

Significance. If the central claim is established, the two-stage pretraining recipe would be a practically useful strategy for exploiting large approximate databases to improve models trained on small DFT-derived datasets. The paper has clear strengths: the architecture and training protocol are described in detail, Table I reports standard deviations over nine repeats, the code and data are made available in a public repository, and the negative result on Dataset2 is reported rather than hidden. However, the generalizability claim currently rests on an unverified absence of structure overlap between the AFLOW pretraining set and the evaluation sets, and on a base-model selection performed on the same benchmark that later produces the headline number. Both points need to be addressed before the main conclusion can be considered robust.

major comments (4)
  1. [II.B and Table I] The central claim that Step 3 improves generalizability is not yet supported because the manuscript never reports whether the 5,578 AFLOW AGL materials overlap with Dataset1, Dataset2, MIX, or the external test compound (BaSbO3)2. Dataset1 consists of common rocksalt, zincblende, and wurtzite compounds that are likely present in AFLOW-LIB, and if a target structure appeared in the AFLOW pretraining set, the Step 3 model has already seen that exact graph and an approximate LTC label for it; the improvement could then reflect label familiarity rather than transferable structure-property relations. The cross-dataset rows in Table I do not substitute for an overlap analysis, since the model trained directly on AFLOW in Step 1 already reaches 0.55 MAPE on Dataset1. Because the GitHub repository is provided, this overlap can be computed and should be reported; in the meantime, the phrase 'better generalizability' in Section IV should be conditioned on an overlap-free evaluation.
  2. [Section III, Figs. 3-5] The base model was selected from MatBench candidates using validation performance on Dataset1, and the same Dataset1 numbers are then reported as the headline result in Table I. This makes the absolute Dataset1 MAPE values and the Dataset1-specific comparison vulnerable to selection bias. The authors should either use a nested or separate selection procedure or report results for both candidate bases. In addition, no non-neural baseline (e.g., composition or structure descriptors with a gradient-boosting or ridge regression model) is reported, so the reader cannot separate the effect of the transfer-learning protocol from the effect of having a reasonable regression model on a small, log-scaled problem. Adding such a baseline would make the double-transfer gain more interpretable.
  3. [Section IV] The conclusion that double transfer learning provides 'reduced overfitting' is based only on validation-loss curves. Without training-loss curves or an explicit comparison of the train/validation gap, a validation curve that rises at later epochs can reflect distribution shift, learning-rate effects, or run-to-run variability across the nine splits; it is not direct evidence of overfitting. The claim in Section IV should either be supported with train/validation gap plots or softened to a statement about validation-error stabilization.
  4. [Section II.A] The external validation rests on a single compound, (BaSbO3)2, and the manuscript does not state whether this material appeared in the AFLOW AGL pretraining set or in any training split. The predicted value (1.23 W/mK) differs from the ab initio value (7.1 W/mK) by roughly a factor of six, so describing these as 'on the same order of magnitude' and as evidence of 'agreement' overstates the support. At a minimum, the overlap status should be reported, the run-to-run spread of the predictions should be given, and the conclusion should be framed as an indicative single-case check rather than a validation of the generalizability claim.
minor comments (4)
  1. [Section II.A] The step numbering in the text is inconsistent with Table I: the sentence describing the drop from 82% to 76% calls the pretrained-MEGNet condition 'Step 1' and then refers to 'Step 2' for the double-transfer condition, while Table I labels these as Step 2 and Step 3. Please align the narrative with the table labels.
  2. [Section II.C] The data-processing description states that logarithmic scaling is followed by standardization 'using the parameters of the corresponding dataset' before the nine validation splits are made. If the standardization statistics are computed on the full dataset, this is a mild form of leakage; the authors should state explicitly that scaling parameters are fitted on the training folds only.
  3. [Section II.B] The figures would be easier to interpret if they included a legend or caption specification of the curves (training versus validation) and the error bands; the current text comments on 'validation loss' but does not state whether the shaded areas correspond to standard deviations over the nine runs.
  4. [Section III] The conclusion that transfer learning is 'effective' is stated too broadly in the opening sentence of Section IV; the paper's own Table I shows that Step 3 degrades Dataset2, so the conclusion should be restricted to datasets with a wider LTC range and more structural diversity, as the later sentences in the section already acknowledge.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the transfer-learning comparison is an empirical benchmark, not a derivation that reduces to its inputs.

full rationale

The paper's central claim — that double transfer learning improves precision and generalizability on diverse LTC datasets — is supported by Table I and the training curves. Step 3 uses the AFLOW AGL dataset as a pre-fine-tuning source and then fine-tunes on Dataset1, Dataset2, and MIX; this is a supervised transfer-learning protocol, and the reported validation MAPE is an empirical outcome rather than a quantity defined by the training procedure. No equation in the paper defines a predicted LTC in terms of the target values, and no fitted parameter is relabeled as a prediction. The MEGNet base model is selected partly using Dataset1 (Section II.B), which could introduce optimistic bias, and the paper does not report whether AFLOW AGL contains structures overlapping the target sets, so the generalizability claim is under-supported; however, these are data-leakage or model-selection risks, not circularity. Self-citations (Togo and Chaput for phono3py and Dataset1 sources) are provenance citations for external first-principles calculations and code, not load-bearing argumentative support. The external (BaSbO3)2 check is a genuine independent calculation, though it shares the same uncertainty about AFLOW overlap. Under the stated standards, no circular step can be quoted with a specific reduction, so the score is 0.

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

The central claim rests on empirical data-quality and transferability assumptions rather than on mathematical axioms; no invented physical entities are introduced.

free parameters (3)
  • MLP head architecture (3 dense layers, 350 neurons each) = 3 layers of 350 neurons
    Chosen without reported ablations; the reported gains depend on this architecture. Section II.B.
  • Training length and random seed = 300 epochs, seed 42
    Fixed by the authors; no sensitivity analysis is shown for 300 epochs or the single seed used to generate the 9 splits.
  • Validation split procedure = 9 random 80/20 splits per dataset
    The reported averages and standard deviations come from this specific split scheme; repeated runs use the same fixed seed, so the split variability is only a portion of the possible randomness.
assumptions (4)
  • domain assumption MEGNet's pretrained representations transfer to LTC prediction.
    Section II.B: ParAIsite is built by attaching an MLP to MEGNet's last hidden layer; the authors select MEGNet via a Dataset1 validation test but do not test whether a descriptor-based model would do as well.
  • domain assumption AFLOW AGL approximate LTC values carry transferable signal.
    Section II.C, Step 3: the headline improvement assumes that fine-tuning on 5,578 AGL values, despite their low quality, moves the model toward a useful representation of LTC. This is the load-bearing assumption behind the Step 3 gains.
  • domain assumption DFT anharmonic-lattice-dynamics values are accurate enough to serve as ground truth.
    Section II.A: Dataset1 and Dataset2 are treated as high-quality labels; any systematic error in the phono3py calculations propagates into all reported MAPE values.
  • domain assumption Small datasets are representative enough to support generalization claims.
    Section III: the paper claims improved generalizability, but the claims rest on 96 and 143 materials, and the 9 random 80/20 splits do not test out-of-distribution materials.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transfer Learning for Deep Learning-based Prediction of Lattice Thermal Conductivity." pith.science (2026). https://pith.science/paper/66FZ4IUC

@misc{pith2026241118259,
  author       = {Pith},
  title        = {Pith review of: Transfer Learning for Deep Learning-based Prediction of Lattice Thermal Conductivity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/66FZ4IUC}},
  note         = {Machine review of arXiv:2411.18259}
}
read the original abstract

Machine learning promises to accelerate the material discovery by enabling high-throughput prediction of desirable macro-properties from atomic-level descriptors or structures. However, the limited data available about precise values of these properties have been a barrier, leading to predictive models with limited precision or the ability to generalize. This is particularly true of lattice thermal conductivity (LTC): existing datasets of precise (ab initio, DFT-based) computed values are limited to a few dozen materials with little variability. Based on such datasets, we study the impact of transfer learning on both the precision and generalizability of a deep learning model (ParAIsite). We start from an existing model (MEGNet~\cite{Chen2019}) and show that improvements are obtained by fine-tuning a pre-trained version on different tasks. Interestingly, we also show that a much greater improvement is obtained when first fine-tuning it on a large datasets of low-quality approximations of LTC (based on the AGL model) and then applying a second phase of fine-tuning with our high-quality, smaller-scale datasets. The promising results obtained pave the way not only towards a greater ability to explore large databases in search of low thermal conductivity materials but also to methods enabling increasingly precise predictions in areas where quality data are rare.

Figures

Figures reproduced from arXiv: 2411.18259 by the authors.

Figure 1
Figure 1. FIG. 1. The ParAIsite model architecture. Each dense layer contains 350 neurons. The output property is the thermal [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Results of validation tests of ParAIsite when using top-performing models cataloged on MatBench [31] as pre-trained [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Validation loss (MAPE) for models trained and tested on Dataset1 across training epochs. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Validation loss (MAPE) for models trained and tested on MIX across training epochs. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Validation loss (MAPE) for models trained and tested on Dataset2 across training epochs. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 20 canonical work pages

  1. [1]

    C. Chen, W. Ye, Y. Zuo, C. Zheng, and S. P. Ong, Graph networks as a universal machine learning framework for molecules and crystals, Chemistry of Materials 31, 3564–3572 (2019)

    work page 2019

  2. [2]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, Physics-informed machine learning, Nature Reviews Physics 3, 422–440 (2021)

    work page 2021

  3. [3]

    C. K. H. Borg, E. S. Muckley, C. Nyby, J. E. Saal, L. Ward, A. Mehta, and B. Meredig, Quantifying the performance of machine learning models in materials discovery, Digital Discovery 2, 327–338 (2023)

    work page 2023

  4. [4]

    Asensio Ramos, M

    A. Asensio Ramos, M. C. M. Cheung, I. Chifu, and R. Gafeira, Machine learning in solar physics, Living Reviews in Solar Physics 20, 10.1007/s41116-023-00038-x (2023)

    doi:10.1007/s41116-023-00038-x 2023

  5. [5]

    He, Y.-G

    W.-B. He, Y.-G. Ma, L.-G. Pang, H.-C. Song, and K. Zhou, High-energy nuclear physics meets machine learning, Nuclear Science and Techniques 34, 10.1007/s41365-023-01233-z (2023)

    doi:10.1007/s41365-023-01233-z 2023

  6. [6]

    A. C. Mater and M. L. Coote, Deep learning in chemistry, Journal of Chemical Information and Modeling 59, 2545–2559 (2019)

    work page 2019

  7. [7]

    Vinod, U

    V. Vinod, U. Kleinekath¨ ofer, and P. Zaspel, Optimized multifidelity machine learning for quantum chemistry, Machine Learning: Science and Technology 5, 015054 (2024)

    work page 2024

  8. [8]

    Sarkar, B

    C. Sarkar, B. Das, V. S. Rawat, J. B. Wahlang, A. Nongpiur, I. Tiewsoh, N. M. Lyngdoh, D. Das, M. Bidarolli, and H. T. Sony, Artificial intelligence and machine learning technology driven modern drug discovery and development, International Journal of Molecular Sciences 24, 2026 (2023)

    work page 2023

Show all 32 references
  1. [9]

    Zhang, H

    B. Zhang, H. Shi, and H. Wang, Machine learning and ai in cancer prognosis, prediction, and treatment selection: A critical approach, Journal of Multidisciplinary Healthcare V olume 16, 1779–1791 (2023)

    2023

  2. [10]

    Swanson, E

    K. Swanson, E. Wu, A. Zhang, A. A. Alizadeh, and J. Zou, From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment, Cell 186, 1772–1791 (2023)

    2023

  3. [11]

    Yaqoob, R

    A. Yaqoob, R. Musheer Aziz, and N. K. verma, Applications and techniques of machine learning in cancer classification: A systematic review, Human-Centric Intelligent Systems 3, 588–615 (2023)

    2023

  4. [12]

    Ihalage and Y

    A. Ihalage and Y. Hao, Formula graph self-attention network for representation-domain independent materials discovery, Advanced Science 9, 10.1002/advs.202200164 (2022)

    2022 doi

  5. [13]

    Suzuki, T

    Y. Suzuki, T. Taniai, K. Saito, Y. Ushiku, and K. Ono, Self-supervised learning of materials concepts from crystal structures via deep neural networks, Machine Learning: Science and Technology 3, 045034 (2022)

    2022

  6. [14]

    Zhang and C

    Y. Zhang and C. Ling, A strategy to apply machine learning to small datasets in materials science, npj Computational Materials 4, 10.1038/s41524-018-0081-z (2018)

    2018 doi

  7. [15]

    Y. Qian, Y. Luo, Y. Li, T. Xiong, L. Wang, W. Zhang, S. Gang, X. Li, Q. Jiang, and J. Yang, Enhanced electromagnetic wave absorption, thermal conductivity and flame retardancy of BCN@LDH/EP for advanced electronic packing materials, Chem. Eng. J. 467, 143433 (2023)

    2023

  8. [16]

    Huang, T

    T. Huang, T. Wang, J. Jin, M. Chen, and L. Wu, Design of silicon rubber/bn film with high through-plane thermal conductivity and ultra-low contact resistance, Chemical Engineering Journal 469, 143874 (2023)

    2023

  9. [17]

    C. E. Calderon, J. J. Plata, C. Toher, C. Oses, O. Levy, M. Fornari, A. Natan, M. J. Mehl, G. Hart, M. Buongiorno Nardelli, and S. Curtarolo, The aflow standard for high-throughput materials science calculations, Computational Materials Science 108, 233–238 (2015)

    2015

  10. [18]

    Kirklin, J

    S. Kirklin, J. E. Saal, B. Meredig, A. Thompson, J. W. Doak, M. Aykol, S. R¨ uhl, and C. Wolverton, The open quantum materials database (oqmd): assessing the accuracy of dft formation energies, npj Computational Materials 1, 10.1038/npj- compumats.2015.10 (2015)

    2015 doi

  11. [19]

    A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, and K. A. Persson, Commentary: The materials project: A materials genome approach to accelerating materials innovation, APL Materials 1, 10.1063/1.4812323 (2013)

    2013 doi

  12. [20]

    Choudhary, K

    K. Choudhary, K. F. Garrity, A. C. E. Reid, B. DeCost, A. J. Biacchi, A. R. Hight Walker, Z. Trautt, J. Hattrick-Simpers, A. G. Kusne, A. Centrone, A. Davydov, J. Jiang, R. Pachter, G. Cheon, E. Reed, A. Agrawal, X. Qian, V. Sharma, H. Zhuang, S. V. Kalinin, B. G. Sumpter, G. ...

    2020

  13. [21]

    Blanco, E

    M. Blanco, E. Francisco, and V. Lua˜ na, Gibbs: isothermal-isobaric thermodynamics of solids from energy curves using a quasi-harmonic debye model, Computer Physics Communications 158, 57–72 (2004)

    2004

  14. [22]

    Toher, J

    C. Toher, J. J. Plata, O. Levy, M. de Jong, M. Asta, M. B. Nardelli, and S. Curtarolo, High-throughput computational screening of thermal conductivity, debye temperature, and gr¨ uneisen parameter using a quasiharmonic debye model, Physical Review B 90, 10.1103/physrevb.90.174...

    2014 doi

  15. [23]

    Zhuang, Z

    F. Zhuang, Z. Qi, K. Duan, D. Xi, Y. Zhu, H. Zhu, H. Xiong, and Q. He, A comprehensive survey on transfer learning (2019). 9

    2019

  16. [24]

    S. Kong, D. Guevarra, C. P. Gomes, and J. M. Gregoire, Materials representation and transfer learning for multi-property prediction, Applied Physics Reviews 8, 10.1063/5.0047066 (2021)

    2021 doi

  17. [25]

    Hoffmann, J

    N. Hoffmann, J. Schmidt, S. Botti, and M. A. L. Marques, Transfer learning on large datasets for the accurate prediction of material properties (2023)

    2023

  18. [26]

    A. Seko, A. Togo, H. Hayashi, K. Tsuda, L. Chaput, and I. Tanaka, Prediction of low-thermal-conductivity compounds with first-principles anharmonic lattice-dynamics calculations and bayesian optimization, Physical Review Letters 115, 10.1103/physrevlett.115.205901 (2015)

    2015 doi

  19. [27]

    A. Togo, L. Chaput, and I. Tanaka, Distributions of phonon lifetimes in brillouin zones, Physical Review B 91, 10.1103/physrevb.91.094306 (2015)

    2015 doi

  20. [28]

    A. Togo, L. Chaput, T. Tadano, and I. Tanaka, Implementation strategies in phonopy and phono3py, J. Phys. Condens. Matter 35, 353001 (2023)

    2023

  21. [29]

    Chaput, Direct solution to the linearized phonon boltzmann equation, Phys

    L. Chaput, Direct solution to the linearized phonon boltzmann equation, Phys. Rev. Lett. 110, 265506 (2013)

    2013

  22. [30]

    Miyazaki, T

    H. Miyazaki, T. Tamura, M. Mikami, K. Watanabe, N. Ide, O. M. Ozkendir, and Y. Nishino, Machine learning based prediction of lattice thermal conductivity for half-heusler compounds using atomic information, Scientific Reports 11, 10.1038/s41598-021-92030-4 (2021)

    2021 doi

  23. [31]

    A. Dunn, Q. Wang, A. Ganose, D. Dopp, and A. Jain, Benchmarking materials property prediction methods: the matbench test set and automatminer reference algorithm, npj Computational Materials 6, 10.1038/s41524-020-00406-3 (2020)

    2020 doi

  24. [32]

    A. Y.-T. Wang, S. K. Kauwe, R. J. Murdock, and T. D. Sparks, Compositionally restricted attention-based network for materials property predictions, npj Computational Materials 7, 77 (2021)

    2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.