pith. machine review for the scientific record. sign in

arxiv: 2605.07227 · v1 · submitted 2026-05-08 · ❄️ cond-mat.mtrl-sci

Recognition: 2 theorem links

· Lean Theorem

Physics Aware Representation Learning on Electronic Charge Density for Materials Property Prediction

Albert Linda, Kammampati Sai Kumar, Shubham Kumar Maurya, Somnath Bhowmick

Pith reviewed 2026-05-11 02:02 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci
keywords charge densityautoencodermaterials property predictiondensity functional theorymechanical propertiesthermodynamic propertiesdimensionality reductioncrystal structures
0
0 comments X

The pith

A convolutional autoencoder compresses three-dimensional electronic charge density into a compact latent space that accurately predicts mechanical and thermodynamic properties of inorganic crystals.

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

The paper establishes that the electronic charge density, computed via density functional theory, serves as a direct and transferable input for predicting crystal properties once its high-dimensional grid is reduced by unsupervised learning. A three-dimensional convolutional autoencoder shrinks the original 128 cubed grid to a 16 by 16 by 16 by 16 latent representation while keeping reconstruction errors low across many crystal systems. Regression models trained on this latent vector, either alone or augmented with composition descriptors, reach R-squared values of 0.94 for bulk modulus, 0.88 for Young's modulus, 0.87 for shear modulus, 0.96 for formation energy, and 0.89 for Debye temperature on roughly 6059 compounds. Because the method uses only a fraction of the compute required for full density functional theory, it opens a route to rapid screening that still rests on the underlying quantum-mechanical charge distribution.

Core claim

The central claim is that an unsupervised three-dimensional convolutional autoencoder can compress DFT-derived electronic charge-density grids from 128x128x128 to 16x16x16x16 while preserving the physically meaningful features needed for downstream regression, yielding R2 scores of 0.94 for bulk modulus K, 0.88 for Young's modulus E, 0.87 for shear modulus G, 0.96 for formation energy Eform, and 0.89 for Debye temperature Theta when the latent vectors are supplied to LightGBM or attention-based 3D CNN models, with further gains when composition-based MAGPIE descriptors are added.

What carries the argument

A three-dimensional convolutional autoencoder that learns a 16x16x16x16 latent representation from 128x128x128 charge-density grids while maintaining negligible reconstruction error across diverse crystal symmetries.

If this is right

  • Latent charge-density vectors alone suffice for high-accuracy prediction of bulk, shear, and Young's moduli as well as formation energy and Debye temperature.
  • Augmenting the latent vectors with composition descriptors raises predictive accuracy still further.
  • The workflow requires approximately one-twenty-fifth the computational cost of a complete DFT calculation for each new material.
  • The same compression step succeeds uniformly across multiple crystal symmetries in a large inorganic dataset.

Where Pith is reading between the lines

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

  • The same autoencoder architecture could be applied to other spatially resolved quantum fields, such as electrostatic potentials, to predict additional properties without new training from scratch.
  • Because the latent space is continuous and low-dimensional, it might support gradient-based optimization to generate hypothetical charge densities that optimize a target property.
  • If the learned features align with interpretable quantities such as bond lengths or valence electron counts, the model could guide physical intuition rather than function only as a black-box predictor.
  • Extending the approach to defective or disordered structures would test whether the compression remains robust when perfect periodicity is broken.

Load-bearing premise

The compressed latent representation retains every physically relevant aspect of the full charge-density grid that determines the five target properties across different crystal structures.

What would settle it

Repeating the training and evaluation on an independent set of several thousand compounds drawn from crystal systems or chemical families absent from the original 6059-compound collection and observing R2 values consistently below 0.70 for the same properties would falsify the claim that the latent features are sufficient and transferable.

Figures

Figures reproduced from arXiv: 2605.07227 by Albert Linda, Kammampati Sai Kumar, Shubham Kumar Maurya, Somnath Bhowmick.

Figure 1
Figure 1. Figure 1: The original charge density data (128 × 128 × 128) is passed through the encoder to obtain a compressed latent representation (16×16×16×16). The autoencoder is trained by minimizing the reconstruction loss between the input and output. The compressed latent representation is utilized for two purposes: (a) One can decode and get back the original charge density, which is readable by the VESTA software, 25 (… view at source ↗
Figure 2
Figure 2. Figure 2: Representative crystal structures and their distribution across crystal systems. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Statistical distributions of five key material properties: bulk modulus (K), Young’s [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Schematic representation of the 3D convolutional autoencoder architecture. The [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of the charge density distributions for representative cubic compounds. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Schematic of the hybrid feature-fusion workflow for materials property prediction, [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Attention 3D CNN model: The architecture consists of stacked residual 3D convo [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: LightGBM model: Actual versus predicted values for the five material properties [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Attention CNN model: Actual versus predicted values for the five material prop [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
read the original abstract

The fundamental quantity governing the mechanical and thermodynamic properties of a crystalline solid is its electronic charge density. Yet, its direct use for the rapid prediction of materials properties remains challenging due to its high dimensionality. Here, we present a physics-informed deep learning framework that directly predicts mechanical and thermodynamic properties from the three-dimensional electronic charge density derived from density functional theory (DFT). The proposed approach first utilizes a three-dimensional convolutional autoencoder for unsupervised dimensionality reduction, compressing a high-resolution charge-density grid (128 x 128 x 128) into a compact latent representation (16 x 16 x 16 x 16) while preserving physically meaningful features, as confirmed by negligible reconstruction errors across diverse crystal systems. The compressed latent-space representation of charge density is then used by two different regression models for property prediction: Light Gradient Boosting Machine (LightGBM) and Attention-based 3D Convolutional Neural Networks (Att CNN), and their performance is compared. Combining composition-based descriptors (Material Agnostic Platform for Informatics and Exploration or MAGPIE) with electronic charge density data further improves the model accuracy. Using a dataset of about 6059 inorganic compounds spanning multiple crystal symmetries, the models achieve strong predictive performance for bulk modulus K (R2 = 0.94), Young's modulus E (R2 = 0.88), shear modulus G (R2 = 0.87), formation energy Eform (R2 = 0.96), and Debye temperature {\Theta} (R2 = 0.89). This work establishes electronic charge density as a transferable, physics-grounded descriptor for materials property prediction, requiring ~ 1/25 the computational resources of full-fledged DFT calculations.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript presents a physics-informed deep learning framework that compresses 128³ DFT-derived electronic charge-density grids into a 16×16×16×16 latent representation via a 3D convolutional autoencoder, then uses this compressed representation (alone or combined with MAGPIE composition descriptors) as input to LightGBM and attention-based 3D CNN regressors to predict bulk modulus K, Young's modulus E, shear modulus G, formation energy Eform, and Debye temperature Θ. On a dataset of ~6059 inorganic compounds spanning multiple crystal systems, the models report R² values of 0.94 (K), 0.88 (E), 0.87 (G), 0.96 (Eform), and 0.89 (Θ), with the claim that the approach requires ~1/25 the computational cost of full DFT while establishing charge density as a transferable descriptor.

Significance. If the reported performance is shown to generalize without data leakage and if the latent space demonstrably retains property-relevant charge-density features, the work would provide a practical route to property prediction that bypasses full electronic-structure calculations for screening. The use of a sizable, symmetry-diverse dataset, the direct comparison of two regression architectures, and the optional fusion with composition-based features are positive elements. The central limitation is that reconstruction error alone does not establish retention of the specific density variations that govern elastic and vibrational properties.

major comments (2)
  1. [Results (paragraphs presenting R² values and dataset description)] The results section reporting R² = 0.94 for K (and the corresponding values for E, G, Eform, Θ) provides no information on train/test splits, cross-validation strategy, or error bars. Because the autoencoder and downstream regressors are trained on the same ~6059-compound dataset, the absence of these details leaves open the possibility of leakage or overfitting, directly undermining confidence in the quoted predictive performance.
  2. [Methods (autoencoder architecture) and Results (latent-space usage)] The claim that the 16×16×16×16 latent representation 'preserves physically meaningful features' rests solely on negligible reconstruction error (Abstract and Methods). Reconstruction fidelity does not guarantee that density gradients, curvatures, or bonding signatures relevant to elastic moduli and Debye temperature are retained; an ablation study, latent-dimension importance analysis, or comparison against property-specific descriptors is required to support the central assertion that the compressed space is sufficient for the reported regressions.
minor comments (2)
  1. [Abstract] The abstract states the framework is 'physics-informed,' yet the autoencoder training is purely unsupervised with no explicit physics-based loss terms or constraints; this terminology should be clarified or qualified.
  2. [Figure captions and Methods] Figure captions and text should explicitly state the number of compounds used for autoencoder training versus regression training to allow readers to assess potential overlap.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which have helped us identify areas for improvement in clarity and validation. We address each major comment point by point below and have revised the manuscript accordingly.

read point-by-point responses

  1. Referee: [Results (paragraphs presenting R² values and dataset description)] The results section reporting R² = 0.94 for K (and the corresponding values for E, G, Eform, Θ) provides no information on train/test splits, cross-validation strategy, or error bars. Because the autoencoder and downstream regressors are trained on the same ~6059-compound dataset, the absence of these details leaves open the possibility of leakage or overfitting, directly undermining confidence in the quoted predictive performance.

    Authors: We agree that the original manuscript did not provide sufficient detail on the data partitioning and validation procedures, which is a valid concern for assessing potential leakage or overfitting. In the revised manuscript, we have expanded the Methods section to explicitly describe our protocol: the dataset was randomly partitioned into an 80/20 train/test split (with stratification by crystal system to maintain diversity), the autoencoder was trained solely on the training portion for unsupervised compression, and the downstream regressors (LightGBM and Att CNN) were trained and evaluated using 5-fold cross-validation on the training set with the test set held completely out. We now report mean R² values along with standard deviations across the folds as error bars in the updated results tables and figures. These changes eliminate ambiguity regarding data leakage and strengthen confidence in the reported performance. revision: yes

  2. Referee: [Methods (autoencoder architecture) and Results (latent-space usage)] The claim that the 16×16×16×16 latent representation 'preserves physically meaningful features' rests solely on negligible reconstruction error (Abstract and Methods). Reconstruction fidelity does not guarantee that density gradients, curvatures, or bonding signatures relevant to elastic moduli and Debye temperature are retained; an ablation study, latent-dimension importance analysis, or comparison against property-specific descriptors is required to support the central assertion that the compressed space is sufficient for the reported regressions.

    Authors: We acknowledge that reconstruction error alone does not fully establish retention of the specific charge-density features (e.g., gradients and bonding signatures) that govern the target properties, and the original manuscript's reliance on this metric for the central claim was insufficient. While the manuscript already shows that the latent representation yields strong predictive performance and that fusing it with MAGPIE descriptors further improves accuracy (indicating complementary information beyond composition), this indirect evidence is not conclusive. In the revised manuscript, we have added an ablation study comparing regression performance using the learned latent vectors versus randomized latent vectors of the same dimensionality, as well as a latent-dimension sensitivity analysis that perturbs individual dimensions and quantifies the resulting change in predicted properties. These additions provide direct support that the compressed representation retains property-relevant physical information. revision: yes

Circularity Check

0 steps flagged

No circularity: unsupervised compression followed by independent supervised regression

full rationale

The derivation chain consists of (1) unsupervised training of a 3D convolutional autoencoder solely on 128^3 DFT charge-density grids to produce a 16x16x16x16 latent vector, justified by reconstruction error, and (2) separate supervised regression (LightGBM or Att-CNN) trained on the latent vectors plus optional MAGPIE descriptors to predict the five target properties. The reported R2 values are test-set performance metrics of the regression step; they are not obtained by fitting any parameter to the targets and then relabeling the fit as a prediction. No equation equates a derived quantity to its own input by construction, no self-citation supplies a load-bearing uniqueness theorem, and the latent-space features are learned from the density input alone rather than from the property labels. The workflow is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that charge density encodes the target properties, plus several free architectural choices whose values are not reported.

free parameters (2)
  • latent grid size 16x16x16x16
    Chosen to balance compression and reconstruction fidelity; value is not derived from first principles.
  • autoencoder depth, filter counts, and training hyperparameters
    Control reconstruction quality and downstream R2 but are selected empirically.
axioms (1)
  • domain assumption DFT-computed charge density is a sufficient descriptor for mechanical and thermodynamic properties
    Invoked when the authors state that charge density is the fundamental quantity governing the properties.

pith-pipeline@v0.9.0 · 5622 in / 1646 out tokens · 47610 ms · 2026-05-11T02:02:19.261089+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

74 extracted references · 74 canonical work pages

  1. [1]

    Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties , author =. Phys. Rev. Lett. , volume =. 2018 , month =. doi:10.1103/PhysRevLett.120.145301 , url =

    work page doi:10.1103/physrevlett.120.145301 2018

  2. [2]

    Projector augmented-wave method , author =. Phys. Rev. B , volume =. 1994 , month =. doi:10.1103/PhysRevB.50.17953 , url =

    work page doi:10.1103/physrevb.50.17953 1994

  3. [3]

    Generalized Gradient Approximation Made Simple , author =. Phys. Rev. Lett. , volume =. 1996 , month =. doi:10.1103/PhysRevLett.77.3865 , url =

    work page doi:10.1103/physrevlett.77.3865 1996

  4. [4]

    , title =

    Cuevas-Zuviría, Bruno and Pacios, Luis F. , title =. Journal of Chemical Information and Modeling , volume =. 2020 , doi =

    work page 2020

  5. [5]

    and Son, Steven F

    Casey, Alex D. and Son, Steven F. and Bilionis, Ilias and Barnes, Brian C. , title =. Journal of Chemical Information and Modeling , volume =. 2020 , doi =

    work page 2020

  6. [6]

    A general-purpose machine learning framework for predicting properties of inorganic materials , volume=

    Ward, Logan and Agrawal, Ankit and Choudhary, Alok and Wolverton, Christopher , year=. A general-purpose machine learning framework for predicting properties of inorganic materials , volume=. npj Computational Materials , publisher=

    work page

  7. [7]

    Reproducibility in materials informatics: lessons from ‘A general-purpose machine learning framework for predicting properties of inorganic materials’

    Persaud, Daniel and Ward, Logan and Hattrick-Simpers, Jason. Reproducibility in materials informatics: lessons from ‘A general-purpose machine learning framework for predicting properties of inorganic materials’. Digital Discovery. 2024. doi:10.1039/D3DD00199G

    work page doi:10.1039/d3dd00199g 2024

  8. [8]

    VESTA : a three-dimensional visualization system for electronic and structural analysis

    Momma, Koichi and Izumi, Fujio. VESTA : a three-dimensional visualization system for electronic and structural analysis. Journal of Applied Crystallography. 2008. doi:10.1107/S0021889808012016 , url =

    work page doi:10.1107/s0021889808012016 2008

  9. [9]

    and Misra, Shikhar

    Kumar, Rajan and Joshi, Ablokit and Khan, Salman A. and Misra, Shikhar. Automated extraction of synthesis parameters of pulsed laser-deposited materials from scientific literature. Digital Discovery. 2024. doi:10.1039/D4DD00051J

    work page doi:10.1039/d4dd00051j 2024

  10. [10]

    Journal of Chemical Information and Modeling , volume =

    Sinha, Prashant and Joshi, Ablokit and Dey, Rik and Misra, Shikhar , title =. Journal of Chemical Information and Modeling , volume =. 2024 , doi =

    work page 2024

  11. [11]

    Journal of Computational Chemistry , volume=

    Ab-initio simulations of materials using VASP: Density-functional theory and beyond , author=. Journal of Computational Chemistry , volume=. 2008 , url=

    work page 2008

  12. [12]

    Ramesh and K

    Yusuf Khan and Shahid Bashir and Maryam Hina and S. Ramesh and K. Ramesh and M.A. Mujtaba and Indranil Lahiri and S. Ramesh , title =. Journal of Materials Science , year =

    work page

  13. [13]

    Chinese Physics , volume=

    First-principles study on the charge density and the bulk modulus of the transition metals and their carbides and nitrides , author=. Chinese Physics , volume=

    work page

  14. [14]

    Modern Charge-Density Analysis , editor =

    Cesare Pisani and Roberto Dovesi and Alessandro Erba and Paolo Giannozzi , title =. Modern Charge-Density Analysis , editor =. 2012 , publisher =. doi:10.1007/978-90-481-3836-4_2 , url =

    work page doi:10.1007/978-90-481-3836-4_2 2012

  15. [15]

    Nature Communications , volume=

    Insightful classification of crystal structures using deep learning , author=. Nature Communications , volume=. 2018 , url=

    work page 2018

  16. [16]

    Integrating Materials and Manufacturing Innovation , volume=

    High throughput quantitative metallography for complex microstructures using deep learning: A case study in ultrahigh carbon steel , author=. Integrating Materials and Manufacturing Innovation , volume=. 2020 , url=

    work page 2020

  17. [17]

    npj Computational Materials , volume =

    Yabo Dan and Yong Zhao and Xiang Li and Shaobo Li and Ming Hu and Jianjun Hu , title =. npj Computational Materials , volume =. 2020 , publisher =

    work page 2020

  18. [18]

    Faiz and Bhowmick, Somnath

    Linda, Albert and Akhtar, Md. Faiz and Bhowmick, Somnath. Deformation in Metals: Insights from ab-initio Calculations. Proceedings of the International Conference on Metallurgical Engineering and Centenary Celebration. 2024

    work page 2024

  19. [19]

    Materials Characterization , volume=

    A robust method of denoising experimental micrographs using deep learning , author=. Materials Characterization , volume=. 2023 , publisher=

    work page 2023

  20. [20]

    Accelerating microstructure modeling via machine learning: A method combining Autoencoder and ConvLSTM , author =. Phys. Rev. Mater. , volume =. 2023 , month =. doi:10.1103/PhysRevMaterials.7.083802 , url =

    work page doi:10.1103/physrevmaterials.7.083802 2023

  21. [21]

    Physica Scripta , abstract =

    Ahmad, Owais and Panwar, Vishal and Das, Kaushik and Mukherjee, Rajdip and Bhowmick, Somnath , title =. Physica Scripta , abstract =. 2025 , month =. doi:10.1088/1402-4896/ade832 , url =

    work page doi:10.1088/1402-4896/ade832 2025

  22. [22]

    Deep learning-driven prediction of microstructure evolution via latent space interpolation , author =. Phys. Rev. Mater. , volume =. 2025 , month =. doi:10.1103/5ngk-4v9j , url =

    work page doi:10.1103/5ngk-4v9j 2025

  23. [23]

    arXiv preprint arXiv:2411.12280 , year=

    Large Language Models for Material Property Predictions: elastic constant tensor prediction and materials design , author=. arXiv preprint arXiv:2411.12280 , year=

    work page arXiv

  24. [24]

    APL Machine Learning , volume =

    Elastic constants from charge density distribution in FCC high-entropy alloys using CNN and DFT , author =. APL Machine Learning , volume =. 2024 , publisher =. doi:10.1063/5.0229105 , url =

    work page doi:10.1063/5.0229105 2024

  25. [25]

    Metals , volume=

    Machine Learning-Based Prediction of Elastic Properties Using Reduced Datasets of Accurate Calculations Results , author=. Metals , volume=. 2024 , publisher=. doi:10.3390/met14040438 , url=

    work page doi:10.3390/met14040438 2024

  26. [26]

    Levämäki and F

    H. Levämäki and F. Tasnádi and D. G. Sangiovanni and L. J. S. Johnson and R. Armiento and I. A. Abrikosov , title =. npj Computational Materials , volume =. 2022 , url =

    work page 2022

  27. [27]

    The Journal of Physical Chemistry C , year=

    Zhao, Yong and Yuan, Kunpeng and Liu, Yinqiao and Louis, Steph-Yves and Hu, Ming and Hu, Jianjun , title=. The Journal of Physical Chemistry C , year=. doi:10.1021/acs.jpcc.0c02348 , url=

    work page doi:10.1021/acs.jpcc.0c02348

  28. [28]

    Diamond and Related Materials , volume=

    Explorative prediction of novel superhard carbon allotropes with larger cell: Density functional theory-assisted deep learning , author=. Diamond and Related Materials , volume=. 2024 , publisher=. doi:10.1016/j.diamond.2024.111320 , url=

    work page doi:10.1016/j.diamond.2024.111320 2024

  29. [29]

    Materialia , volume=

    Charge-density based convolutional neural networks for stacking fault energy prediction in concentrated alloys , author=. Materialia , volume=. 2022 , publisher=. doi:10.1016/j.mtla.2022.101620 , url=

    work page doi:10.1016/j.mtla.2022.101620 2022

  30. [30]

    International Journal of Hydrogen Energy , year =

    Xuhao Liu and Zilin Yan and Zheng Zhong , title =. International Journal of Hydrogen Energy , year =. doi:10.1016/j.ijhydene.2021.04.033 , url =

    work page doi:10.1016/j.ijhydene.2021.04.033 2021

  31. [31]

    Scientific Reports , volume =

    Iman Peivaste and Saba Ramezani and Ghasem Alahyarizadeh and Reza Ghaderi and Ahmed Makradi and Salim Belouettar , title =. Scientific Reports , volume =. 2024 , publisher =. doi:10.1038/s41598-023-50893-9 , url =

    work page doi:10.1038/s41598-023-50893-9 2024

  32. [32]

    Kalidindi , title =

    Pranoy Ray and Kamal Choudhary and Surya R. Kalidindi , title =. Integrating Materials and Manufacturing Innovation , volume =. 2025 , publisher =. doi:10.1007/s40192-024-00389-9 , url =

    work page doi:10.1007/s40192-024-00389-9 2025

  33. [33]

    npj Computational Materials , volume =

    Seonghwan Kim and Byung Do Lee and Min Young Cho and Myoungho Pyo and Young-Kook Lee and Woon Bae Park and Kee-Sun Sohn , title =. npj Computational Materials , volume =. 2024 , publisher =

    work page 2024

  34. [34]

    Machine Learning: Science and Technology , volume =

    Deepak Kamal and Anand Chandrasekaran and Rohit Batra and Rampi Ramprasad , title =. Machine Learning: Science and Technology , volume =. 2020 , publisher =. doi:10.1088/2632-2153/ab7a3a , url =

    work page doi:10.1088/2632-2153/ab7a3a 2020

  35. [35]

    Siboni and Bai-Xiang Xu and Dierk Raabe , title =

    Jaber Rezaei Mianroodi and Shahed Rezaei and Nima H. Siboni and Bai-Xiang Xu and Dierk Raabe , title =. npj Computational Materials , volume =. 2022 , publisher =

    work page 2022

  36. [36]

    arXiv:2109.03020v1 [cond-mat.mtrl-sci] , year =

    Bernhard Eidel , title =. arXiv:2109.03020v1 [cond-mat.mtrl-sci] , year =

    work page arXiv

  37. [37]

    Pelegri , title =

    Xuehai Wu and Parameshwaran Pasupathy and Assimina A. Pelegri , title =. Computer Methods and Programs in Biomedicine , volume =. 2024 , doi =

    work page 2024

  38. [38]

    Efficient Data Compression for 3D Sparse TPC via Bicephalous Convolutional Autoencoder , year=

    Huang, Yi and Ren, Yihui and Yoo, Shinjae and Huang, Jin , booktitle=. Efficient Data Compression for 3D Sparse TPC via Bicephalous Convolutional Autoencoder , year=

    work page

  39. [39]

    Significant Dimension Reduction of 3D Brain MRI using 3D Convolutional Autoencoders , year=

    Arai, Hayato and Chayama, Yusuke and Iyatomi, Hitoshi and Oishi, Kenichi , booktitle=. Significant Dimension Reduction of 3D Brain MRI using 3D Convolutional Autoencoders , year=

    work page

  40. [40]

    Auto-encoder based dimensionality reduction , journal =

    Yasi Wang and Hongxun Yao and Sicheng Zhao , keywords =. Auto-encoder based dimensionality reduction , journal =. 2016 , note =. doi:https://doi.org/10.1016/j.neucom.2015.08.104 , url =

    work page doi:10.1016/j.neucom.2015.08.104 2016

  41. [41]

    Dimensionality Reduction Using 3D Residual Autoencoder for Hyperspectral Image Classification , year=

    Pande, Shivam and Banerjee, Biplab , booktitle=. Dimensionality Reduction Using 3D Residual Autoencoder for Hyperspectral Image Classification , year=

    work page

  42. [42]

    Proceedings of the 7th International Conference on Internet Multimedia Computing and Service (ICIMCS '15) , pages =

    Yasi Wang and Hongxun Yao and Sicheng Zhao and Ying Zheng , title =. Proceedings of the 7th International Conference on Internet Multimedia Computing and Service (ICIMCS '15) , pages =. 2015 , doi =

    work page 2015

  43. [43]

    Autoencoder-based Dimensionality Reduction for QSAR Modeling , year=

    Alsenan, Shrooq and Al-Turaiki, Isra and Hafez, Alaaeldin , booktitle=. Autoencoder-based Dimensionality Reduction for QSAR Modeling , year=

    work page

  44. [44]

    Density-Functional Theory of Atoms and Molecules , author=

    work page

  45. [45]

    Electronic Structure: Basic Theory and Practical Methods , author=

    work page

  46. [46]

    Annalen der Physik , volume=

    Zur Quantentheorie der Molekeln , author=. Annalen der Physik , volume=. 1927 , doi=

    work page 1927

  47. [47]

    Physical Review , volume=

    Inhomogeneous Electron Gas , author=. Physical Review , volume=. 1964 , publisher=

    work page 1964

  48. [48]

    Physical Review , volume=

    Self-Consistent Equations Including Exchange and Correlation Effects , author=. Physical Review , volume=. 1965 , publisher=

    work page 1965

  49. [49]

    Physical Review B , volume=

    Self-interaction correction to the exchange-correlation approximation for inhomogeneous electron systems , author=. Physical Review B , volume=. 1981 , publisher=

    work page 1981

  50. [50]

    Physical Review Letters , volume=

    Generalized Gradient Approximation Made Simple , author=. Physical Review Letters , volume=. 1996 , publisher=

    work page 1996

  51. [51]

    Physical Review B , volume=

    Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set , author=. Physical Review B , volume=. 1996 , publisher=

    work page 1996

  52. [52]

    Physical Review B , volume=

    Ab initio molecular dynamics for liquid metals , author=. Physical Review B , volume=. 1993 , url=

    work page 1993

  53. [53]

    APL Materials , volume=

    The Materials Project: A materials database and online interface for high-throughput materials computation , author=. APL Materials , volume=. 2013 , url=

    work page 2013

  54. [54]

    Physical Review B , volume=

    Projector augmented-wave method , author=. Physical Review B , volume=. 1994 , publisher=

    work page 1994

  55. [55]

    Physical Review B , volume=

    From ultrasoft pseudopotentials to the projector augmented-wave method , author=. Physical Review B , volume=. 1999 , publisher=

    work page 1999

  56. [56]

    Physical Review B , volume=

    Special points for Brillouin-zone integrations , author=. Physical Review B , volume=. 1976 , publisher=

    work page 1976

  57. [57]

    Computational Materials Science , volume =

    Calculation of elastic constants of 4d transition metals , author =. Computational Materials Science , volume =. 2015 , publisher =

    work page 2015

  58. [58]

    1986 , publisher =

    Theory of Elasticity , author =. 1986 , publisher =

    work page 1986

  59. [59]

    Proceedings of the 9th Python in Science Conference , year =

    Wes McKinney , title =. Proceedings of the 9th Python in Science Conference , year =

    work page

  60. [60]

    Journal of Machine Learning Research , volume=

    Scikit-learn: Machine Learning in Python , author=. Journal of Machine Learning Research , volume=

    work page

  61. [61]

    Proceedings of the Physical Society

    The elastic behaviour of a crystalline aggregate , author=. Proceedings of the Physical Society. Section A , volume=. 1952 , publisher=

    work page 1952

  62. [62]

    and Toher, Cormac and Curtarolo, Stefano and Ceder, Gerbrand and Persson, Kristin A

    de Jong, Maarten and Chen, Wei and Angsten, Thomas and Jain, Anubhav and Notestine, Randy and Gamst, Anthony and Sluiter, Marcel and Ande, Chaitanya Krishna and van der Zwaag, Sybrand and Plata, Jose J. and Toher, Cormac and Curtarolo, Stefano and Ceder, Gerbrand and Persson, Kristin A. and Asta, Mark , title =. Scientific Data , volume =. 2015 , publishe...

    work page doi:10.1038/sdata.2015.9 2015

  63. [63]

    Larsen and Jens Jørgen Mortensen and Jakob Blomqvist and Ivano E

    Ask H. Larsen and Jens Jørgen Mortensen and Jakob Blomqvist and Ivano E. Castelli and Rasmus Christensen and Marcin Dułak and Jesper Friis and Michael N. Groves and Bjørk Hammer and Cory Hargus and Eric D. Hermes and Peter C. Jennings and Peter Bjerre Jensen and James Kermode and John R. Kitchin and Esben Lundgaard and Andrew Maxson and Thomas Olsen and L...

    work page doi:10.1088/1361-648x/aa680e 2017

  64. [64]

    Reducing the dimensionality of data with neural networks

    Reducing the dimensionality of data with neural networks , author =. Science , volume =. 2006 , publisher =. doi:10.1126/science.1127647 , url =

    work page doi:10.1126/science.1127647 2006

  65. [65]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

    3D Convolutional Neural Networks for Human Action Recognition , author =. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =. 2013 , publisher =

    work page 2013

  66. [66]

    2015 , howpublished =

    TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems , author =. 2015 , howpublished =

    work page 2015

  67. [67]

    and Bursztein, E

    O’Malley, L. and Bursztein, E. and Chollet, F. , title =. 2019 , url =

    work page 2019

  68. [68]

    Samaei and Tu Phan and Maarten

    George Kim and Haoyan Diao and Chanho Lee and A.T. Samaei and Tu Phan and Maarten. First-principles and machine learning predictions of elasticity in severely lattice-distorted high-entropy alloys with experimental validation , journal =. 2019 , issn =. doi:https://doi.org/10.1016/j.actamat.2019.09.026 , url =

    work page doi:10.1016/j.actamat.2019.09.026 2019

  69. [69]

    Johnson and Raymundo Arróyave , keywords =

    Guillermo Vazquez and Prashant Singh and Daniel Sauceda and Richard Couperthwaite and Nicholas Britt and Khaled Youssef and Duane D. Johnson and Raymundo Arróyave , keywords =. Efficient machine-learning model for fast assessment of elastic properties of high-entropy alloys , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.actamat.2022.117924 , url =

    work page doi:10.1016/j.actamat.2022.117924 2022

  70. [70]

    Machine learning elastic constants of multi-component alloys , journal =

    Vivek Revi and Saurabh Kasodariya and Anjana Talapatra and Ghanshyam Pilania and Alankar Alankar , keywords =. Machine learning elastic constants of multi-component alloys , journal =. 2021 , issn =. doi:https://doi.org/10.1016/j.commatsci.2021.110671 , url =

    work page doi:10.1016/j.commatsci.2021.110671 2021

  71. [71]

    Ross and Iyer, Krithika and Held, Leander I

    Grant, Michael and Kunz, M. Ross and Iyer, Krithika and Held, Leander I. and Tasdizen, Tolga and Aguiar, Jeffery A. and Dholabhai, Pratik P. , title=. Journal of Materials Research , year=. doi:10.1557/s43578-022-00557-7 , url=

    work page doi:10.1557/s43578-022-00557-7

  72. [72]

    Khakurel, Hrishabh and Taufique, M. F. N. and Roy, Ankit and Balasubramanian, Ganesh and Ouyang, Gaoyuan and Cui, Jun and Johnson, Duane D. and Devanathan, Ram , title=. Scientific Reports , year=. doi:10.1038/s41598-021-96507-0 , url=

    work page doi:10.1038/s41598-021-96507-0

  73. [73]

    Hayashi and K

    G. Hayashi and K. Suzuki and T. Terai and H. Fujii and M. Ogura and K. Sato , title =. Science and Technology of Advanced Materials: Methods , volume =. 2022 , publisher =. doi:10.1080/27660400.2022.2125853 , URL =

    work page doi:10.1080/27660400.2022.2125853 2022

  74. [74]

    Predicting elastic properties of refractory high-entropy alloys via machine-learning approach , journal =

    Wei Mei and Gaoshang Zhang and Kuang Yu , keywords =. Predicting elastic properties of refractory high-entropy alloys via machine-learning approach , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.commatsci.2023.112249 , url =

    work page doi:10.1016/j.commatsci.2023.112249 2023