pith. sign in

arxiv: 2606.00776 · v1 · pith:U7PZSZSRnew · submitted 2026-05-30 · 💻 cs.LG

Latent Diffusion Pretraining for Crystal Property Prediction

Shrimon Mukherjee , Kishalay Das , Partha Basuchowdhuri , Pawan Goyal , Niloy Ganguly This is my paper

Pith reviewed 2026-06-28 19:28 UTC · model grok-4.3

classification 💻 cs.LG
keywords latent diffusioncrystal property predictionpretrainingvariational autoencodergraph neural networksmaterials discoveryDFT
0
0 comments X

The pith

A latent diffusion model in VAE space pretrains crystal encoders to outperform baselines on property prediction.

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

The paper introduces CrysLDNet to address data scarcity in crystal property prediction by pretraining on unlabeled data. It uses a VAE to map 3D crystal structures to a smooth latent space and applies a diffusion model there to learn structural and chemical semantics. The pretrained graph encoder is then finetuned on labeled data for specific tasks. Experiments on JARVIS and MP datasets show improvements of 4.26% and 4.90% over training from scratch and other pretrained models. The approach also works well with limited data and can help correct DFT errors using experimental data.

Core claim

By combining a variational autoencoder with a diffusion model in the latent space of crystal structures, the framework enables effective capture of semantics from large-scale unlabeled data, resulting in improved performance for downstream property prediction tasks compared to training from scratch or other pretraining strategies.

What carries the argument

The VAE encoder that maps 3D crystal structures into a smooth latent space, allowing the diffusion process to learn representations for the graph encoder.

If this is right

  • The method reduces reliance on scarce labeled data for crystal property prediction.
  • Learned representations are robust in sparse-data conditions.
  • Representations can correct DFT errors when finetuned with limited experimental data.
  • Outperforms on popular DFT datasets like JARVIS and MP.

Where Pith is reading between the lines

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

  • If the latent space is smooth enough, similar latent diffusion pretraining could apply to other graph-structured data like molecules.
  • Scaling the unlabeled dataset size might yield further gains in representation quality.
  • Integrating this with other pretraining techniques could lead to even stronger baselines.

Load-bearing premise

The VAE encoder produces a sufficiently smooth latent space in which the diffusion process can effectively capture structural and chemical semantics from unlabeled crystal data.

What would settle it

Training the same model but with a non-smooth latent space from a different autoencoder and observing no performance gain over baselines would falsify the claim.

Figures

Figures reproduced from arXiv: 2606.00776 by Kishalay Das, Niloy Ganguly, Partha Basuchowdhuri, Pawan Goyal, Shrimon Mukherjee.

Figure 1
Figure 1. Figure 1: Overview of our proposed pretrain–finetune framework, CrysLDNet. In the pretraining stage, a VAE encodes crystal structures into latent representations, on which the LDM is applied to refine the encoded latent space. CrysLDNet is designed to be fully agnostic to the choice of different backbone encoder architectures. The pretrained encoder is then finetuned on labeled data for downstream tasks. (VAE), whic… view at source ↗
Figure 2
Figure 2. Figure 2: Performance comparison (MAE) under limited training data. MAE on four JARVIS properties using 20%, 40%, 60%, and 80% of finetuning data, comparing supervised baselines (PDDFormer, Matformer) with their corresponding CrysLDNet variants. GATGNN (Louis et al., 2020), DimeNet++ (Gasteiger et al., 2020), ALIGNN (Choudhary & DeCost, 2021), Mat￾former (Yan et al., 2022), Equiformer (Liao & Smidt, 2023), PotNet (L… view at source ↗
Figure 3
Figure 3. Figure 3: Expressiveness of pretrained representations on the GNoME dataset. Comparison of A,X and L reconstruction between CrysDiff, DPF, and CrysLDNet. set. For comparison, we consider two supervised baselines, PDDFormer and Matformer, and evaluate them against their corresponding CrysLDNet variants, where PDDFormer and Matformer are used as the backbone encoders in CrysLDNet, respectively. The results, shown in … view at source ↗
Figure 4
Figure 4. Figure 4: (a) A periodic crystal structure represented as a point cloud of atoms arranged in repeating patterns, along with a magnified view of a unit cell. (b) A multigraph representation of the unit cell. (c) Rotational, translational, and periodic symmetries of the crystal. Limitations and Future Work CrysLDNet is pretrained on the GNoME dataset (380K crystals), which already constitutes a large-scale pretraining… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of loss curves between LDM (Without VAE Loss) and CrysLDNet. E. Joint VAE–Flow Training Stability Joint VAE–Flow training can become unstable and may cause the encoder to collapse, but this usually happens when the encoder is trained from scratch with random initialization. A common and effective solution is to pretrain the encoder so that it starts from a meaningful state. In our case, the enco… view at source ↗
Figure 6
Figure 6. Figure 6: Mean latent variance during Stage 2 diffusion training. The latent variance remains stable and consistently non-zero throughout training, indicating that the latent space does not collapse and continues to preserve meaningful representation diversity. the latent variance remains stable and clearly non-zero (approximately 1.9–2.05) across epochs, indicating that the latent space preserves meaningful express… view at source ↗
Figure 7
Figure 7. Figure 7: Local perturbation stability analysis. (a) Pooled latent shift grows smoothly with perturbation scale. (b) Cosine similarity between original and perturbed latent representations remains high even under larger perturbations, indicating a smooth and robust latent space. I.1. Performance Across Structural Complexity We performed additional experiments by stratifying the training and test crystals into three … view at source ↗
Figure 8
Figure 8. Figure 8: Effect of fine-tuning data size on MAE for Bandgap (MBJ), Bulk Modulus, Shear Modulus, and SLME across crystal structures of varying complexity (N ≤ 5, 5 < N < 10, N > 10) on JARVIS, comparing CrysLDNet vs PDDFormer. The results reveal a clear monotonic increase in relative improvement, rising from 2.29% for simple structures to 5.33% for moderately complex crystals and reaching 11.10% for highly complex s… view at source ↗
read the original abstract

Fast and accurate prediction of crystal properties is a central challenge in new materials design. Graph neural networks and Transformer-based models have emerged as powerful tools for this task due to their ability to encode the local structural environment of atoms within a crystal. However, these models are data-hungry, and in practice, labeled data for crystal properties are scarce. Pretraining-finetuning strategies, particularly those based on diffusion models, have shown promise in addressing these limitations. In this work, we introduce a novel latent diffusion based pretraining framework, CrysLDNet, designed to mitigate data scarcity. Our approach integrates a Variational Autoencoder (VAE) with a diffusion model during the pretraining stage. The VAE encoder maps 3D crystal structures into a smooth latent space within which the diffusion process is applied. This latent diffusion pretraining enables the graph encoder to effectively capture structural and chemical semantics from large-scale unlabeled data, which can then be finetuned for specific property prediction tasks. Comprehensive experiments on popular DFT datasets for property prediction reveal that CrysLDNet significantly outperforms both training-from-scratch and pretrained baselines, with improvements of 4.26% and 4.90% on the JARVIS and MP datasets, respectively. Additionally, the learned representations remain robust in sparse-data conditions and are expressive enough to correct DFT errors when finetuned with limited experimental data. Code is available at: https://github.com/shrimonmuke0202/CrysLDNet.git.

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 / 1 minor

Summary. The paper introduces CrysLDNet, a latent diffusion pretraining framework that combines a VAE encoder mapping 3D crystal structures to a smooth latent space with a diffusion model applied in that space. The pretrained graph encoder is then finetuned for crystal property prediction tasks. The central empirical claim is that this approach outperforms training-from-scratch and other pretrained baselines by 4.26% on the JARVIS dataset and 4.90% on the MP dataset, with additional robustness in sparse-data regimes and the ability to correct DFT errors when finetuned on limited experimental data.

Significance. If the performance gains and robustness claims hold after proper validation, the work would offer a practical way to leverage large unlabeled crystal datasets for property prediction, addressing data scarcity in materials science. The explicit release of code at https://github.com/shrimonmuke0202/CrysLDNet.git is a strength that enables reproducibility and further testing of the latent-diffusion mechanism.

major comments (2)
  1. [Abstract] Abstract: the reported improvements of 4.26% (JARVIS) and 4.90% (MP) are presented without any description of the exact baselines, statistical significance tests, data splits, or ablation studies, so the support for the central empirical claim cannot be assessed from the provided text.
  2. [Abstract] Abstract (pretraining stage description): the assertion that the VAE produces a 'smooth latent space' in which diffusion captures structural and chemical semantics rests on an unvalidated assumption; no reconstruction fidelity, latent continuity metrics, or ablation that isolates the diffusion stage versus VAE-only pretraining is supplied, making it impossible to attribute gains to the proposed mechanism.
minor comments (1)
  1. The manuscript would benefit from a dedicated section or table enumerating all baselines, hyperparameters, and evaluation protocols to allow direct replication of the reported percentages.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. We agree that additional context is needed to support the central claims and will revise the abstract accordingly. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the reported improvements of 4.26% (JARVIS) and 4.90% (MP) are presented without any description of the exact baselines, statistical significance tests, data splits, or ablation studies, so the support for the central empirical claim cannot be assessed from the provided text.

    Authors: We agree that the abstract should be more self-contained. In the revised version we will expand it to name the baselines (training-from-scratch and prior pretrained models), state that results are means over five random seeds with standard deviations, reference the standard JARVIS and MP splits used, and note that full ablation tables appear in Section 4. This change will allow readers to evaluate the empirical support directly from the abstract. revision: yes

  2. Referee: [Abstract] Abstract (pretraining stage description): the assertion that the VAE produces a 'smooth latent space' in which diffusion captures structural and chemical semantics rests on an unvalidated assumption; no reconstruction fidelity, latent continuity metrics, or ablation that isolates the diffusion stage versus VAE-only pretraining is supplied, making it impossible to attribute gains to the proposed mechanism.

    Authors: We acknowledge that the abstract currently states the smoothness assumption without supporting detail. The full manuscript reports VAE reconstruction errors and an ablation of VAE-only versus full latent-diffusion pretraining (Sections 4.3–4.4), but these are not referenced in the abstract. We will revise the abstract either to qualify the claim or to add a short parenthetical reference to the supporting experiments, thereby making the attribution to the diffusion stage explicit. revision: yes

Circularity Check

0 steps flagged

No derivation chain; empirical results on external datasets

full rationale

The paper describes an empirical pretraining framework (VAE + latent diffusion) and reports performance gains from finetuning on JARVIS and MP datasets. No equations, parameter fits, or self-citations are presented that reduce the claimed improvements to quantities defined by the model's own inputs or prior outputs. The central claims rest on external benchmark comparisons rather than any self-referential derivation, satisfying the condition for a self-contained result.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the VAE producing a usable latent space and on standard assumptions of diffusion models and graph encoders; no explicit free parameters or invented entities are named in the abstract.

axioms (1)
  • domain assumption A variational autoencoder can map discrete 3D crystal structures into a smooth continuous latent space suitable for diffusion.
    Invoked to justify applying diffusion in the latent space during pretraining.

pith-pipeline@v0.9.1-grok · 5811 in / 1203 out tokens · 42633 ms · 2026-06-28T19:28:40.878565+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

133 extracted references · 30 canonical work pages · 9 internal anchors

  1. [1]

    Nature , volume=

    Machine learning for molecular and materials science , author=. Nature , volume=. 2018 , publisher=

    2018

  2. [2]

    Nature materials , volume=

    Cryptic crystallography , author=. Nature materials , volume=. 2002 , publisher=

    2002

  3. [3]

    KDD 2024, PhD Consortium , year=

    Learning robust representation of crystal materials for property prediction , author=. KDD 2024, PhD Consortium , year=

    2024

  4. [4]

    Genome , number=

    Materials Genome Initiative for Global Competitiveness , author=. Genome , number=

  5. [5]

    Apl Materials , volume=

    Perspective: Web-based machine learning models for real-time screening of thermoelectric materials properties , author=. Apl Materials , volume=. 2016 , publisher=

    2016

  6. [6]

    Nature materials , volume=

    Design of efficient molecular organic light-emitting diodes by a high-throughput virtual screening and experimental approach , author=. Nature materials , volume=. 2016 , publisher=

    2016

  7. [7]

    Nature communications , volume=

    Accelerated search for materials with targeted properties by adaptive design , author=. Nature communications , volume=. 2016 , publisher=

    2016

  8. [8]

    Nature , volume=

    Scaling deep learning for materials discovery , author=. Nature , volume=. 2023 , publisher=

    2023

  9. [9]

    Proceedings of the 40th International Conference on Machine Learning , pages =

    Efficient Approximations of Complete Interatomic Potentials for Crystal Property Prediction , author =. Proceedings of the 40th International Conference on Machine Learning , pages =. 2023 , editor =

    2023

  10. [10]

    Decoupled Weight Decay Regularization

    Decoupled weight decay regularization , author=. arXiv preprint arXiv:1711.05101 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [11]

    A disciplined approach to neural network hyper-parameters: Part 1 -- learning rate, batch size, momentum, and weight decay

    A disciplined approach to neural network hyper-parameters: Part 1--learning rate, batch size, momentum, and weight decay , author=. arXiv preprint arXiv:1803.09820 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    arXiv preprint arXiv:2107.13586 , year=

    Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing , author=. arXiv preprint arXiv:2107.13586 , year=

    work page arXiv

  13. [13]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Bert: Pre-training of deep bidirectional transformers for language understanding , author=. arXiv preprint arXiv:1810.04805 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    OpenAI blog , volume=

    Language models are unsupervised multitask learners , author=. OpenAI blog , volume=

  15. [15]

    Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

    Gppt: Graph pre-training and prompt tuning to generalize graph neural networks , author=. Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

  16. [16]

    Proceedings of the ACM Web Conference 2023 , pages=

    Graphprompt: Unifying pre-training and downstream tasks for graph neural networks , author=. Proceedings of the ACM Web Conference 2023 , pages=

    2023

  17. [17]

    Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

    All in one: Multi-task prompting for graph neural networks , author=. Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

  18. [18]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Hgprompt: Bridging homogeneous and heterogeneous graphs for few-shot prompt learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  19. [19]

    Advances in Neural Information Processing Systems , volume=

    Universal prompt tuning for graph neural networks , author=. Advances in Neural Information Processing Systems , volume=

  20. [20]

    Nature communications , volume=

    Accelerated discovery of stable lead-free hybrid organic-inorganic perovskites via machine learning , author=. Nature communications , volume=. 2018 , publisher=

    2018

  21. [21]

    Nature communications , volume=

    Universal fragment descriptors for predicting properties of inorganic crystals , author=. Nature communications , volume=. 2017 , publisher=

    2017

  22. [22]

    npj Computational Materials , volume=

    Identifying Pb-free perovskites for solar cells by machine learning , author=. npj Computational Materials , volume=. 2019 , publisher=

    2019

  23. [23]

    Physical Review B , volume=

    Including crystal structure attributes in machine learning models of formation energies via Voronoi tessellations , author=. Physical Review B , volume=. 2017 , publisher=

    2017

  24. [24]

    Scientific reports , volume=

    A statistical learning framework for materials science: application to elastic moduli of k-nary inorganic polycrystalline compounds , author=. Scientific reports , volume=. 2016 , publisher=

    2016

  25. [25]

    Physical review letters , volume=

    Prediction of low-thermal-conductivity compounds with first-principles anharmonic lattice-dynamics calculations and Bayesian optimization , author=. Physical review letters , volume=. 2015 , publisher=

    2015

  26. [26]

    Physical Review B , volume=

    Structure classification and melting temperature prediction in octet AB solids via machine learning , author=. Physical Review B , volume=. 2015 , publisher=

    2015

  27. [27]

    Physical Review B , volume=

    Representation of compounds for machine-learning prediction of physical properties , author=. Physical Review B , volume=. 2017 , publisher=

    2017

  28. [28]

    Physical Review B , volume=

    Prediction model of band gap for inorganic compounds by combination of density functional theory calculations and machine learning techniques , author=. Physical Review B , volume=. 2016 , publisher=

    2016

  29. [29]

    Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties , author=. Phys. Rev. Lett. , volume=. 2018 , publisher=

    2018

  30. [30]

    Graph networks as a universal machine learning framework for molecules and crystals , author=. Chem. Mater. , volume=. 2019 , publisher=

    2019

  31. [31]

    Developing an improved crystal graph convolutional neural network framework for accelerated materials discovery , volume=

    Park,Cheol Woo and Wolverton,Chris , year=. Developing an improved crystal graph convolutional neural network framework for accelerated materials discovery , volume=. Physical Review Materials , publisher=. doi:10.1103/physrevmaterials.4.063801 , number=

    work page doi:10.1103/physrevmaterials.4.063801

  32. [32]

    Science Advances , volume=

    Crystal graph attention networks for the prediction of stable materials , author=. Science Advances , volume=. 2021 , publisher=

    2021

  33. [33]

    npj Computational Materials , volume=

    Atomistic Line Graph Neural Network for improved materials property predictions , author=. npj Computational Materials , volume=. 2021 , publisher=

    2021

  34. [34]

    npj Computational Materials , volume =

    Das, Kishalay and Samanta, Bidisha and Goyal, Pawan and others , title =. npj Computational Materials , volume =. 2022 , doi =

    2022

  35. [35]

    Physical Chemistry Chemical Physics , volume=

    Graph convolutional neural networks with global attention for improved materials property prediction , author=. Physical Chemistry Chemical Physics , volume=. 2020 , publisher=

    2020

  36. [36]

    Advances in Neural Information Processing Systems , volume=

    Periodic graph transformers for crystal material property prediction , author=. Advances in Neural Information Processing Systems , volume=

  37. [37]

    International Conference on Machine Learning , pages=

    Efficient approximations of complete interatomic potentials for crystal property prediction , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  38. [38]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Crysgnn: Distilling pre-trained knowledge to enhance property prediction for crystalline materials , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  39. [39]

    Uncertainty in Artificial Intelligence , pages=

    Crysmmnet: multimodal representation for crystal property prediction , author=. Uncertainty in Artificial Intelligence , pages=. 2023 , organization=

    2023

  40. [40]

    ACM Computing Surveys , volume=

    Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing , author=. ACM Computing Surveys , volume=. 2023 , publisher=

    2023

  41. [41]

    CoRR , volume =

    P-tuning v2: Prompt tuning can be comparable to fine-tuning universally across scales and tasks , author=. arXiv preprint arXiv:2110.07602 , year=

    work page arXiv

  42. [42]

    Prefix-Tuning: Optimizing Continuous Prompts for Generation

    Prefix-tuning: Optimizing continuous prompts for generation , author=. arXiv preprint arXiv:2101.00190 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  43. [43]

    The Power of Scale for Parameter-Efficient Prompt Tuning

    The power of scale for parameter-efficient prompt tuning , author=. arXiv preprint arXiv:2104.08691 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    arXiv preprint arXiv:2203.17274 , year=

    Exploring visual prompts for adapting large-scale models , author=. arXiv preprint arXiv:2203.17274 , year=

    work page arXiv

  45. [45]

    European Conference on Computer Vision , pages=

    Visual prompt tuning , author=. European Conference on Computer Vision , pages=. 2022 , organization=

    2022

  46. [46]

    npj Computational Materials , volume=

    Crystal twins: self-supervised learning for crystalline material property prediction , author=. npj Computational Materials , volume=. 2022 , publisher=

    2022

  47. [47]

    2018 , publisher=

    Kittel's Introduction to Solid State Physics , author=. 2018 , publisher=

    2018

  48. [48]

    APL materials , volume=

    Commentary: The Materials Project: A materials genome approach to accelerating materials innovation , author=. APL materials , volume=. 2013 , publisher=

    2013

  49. [49]

    npj computational materials , volume=

    The joint automated repository for various integrated simulations (JARVIS) for data-driven materials design , author=. npj computational materials , volume=. 2020 , publisher=

    2020

  50. [50]

    Photosynthesis research , volume=

    Density functional theory , author=. Photosynthesis research , volume=. 2009 , publisher=

    2009

  51. [51]

    arXiv preprint arXiv:2212.10556 , year=

    Unleashing the power of visual prompting at the pixel level , author=. arXiv preprint arXiv:2212.10556 , year=

    work page arXiv

  52. [52]

    arXiv preprint arXiv:2306.15706 , year=

    Approximated Prompt Tuning for Vision-Language Pre-trained Models , author=. arXiv preprint arXiv:2306.15706 , year=

    work page arXiv

  53. [53]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Adaptergnn: Parameter-efficient fine-tuning improves generalization in gnns , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  54. [54]

    Konstantin Rusch and Michael Bronstein and Andreea Deac and Marc Lackenby and Siddhartha Mishra and Petar Veli

    Francesco Di Giovanni and T. Konstantin Rusch and Michael Bronstein and Andreea Deac and Marc Lackenby and Siddhartha Mishra and Petar Veli. How does over-squashing affect the power of. Transactions on Machine Learning Research , issn=. 2024 , url=

    2024

  55. [55]

    arXiv preprint arXiv:2309.02769 , year=

    Unifying over-smoothing and over-squashing in graph neural networks: A physics informed approach and beyond , author=. arXiv preprint arXiv:2309.02769 , year=

    work page arXiv

  56. [56]

    , author=

    Hardness, softness, and the fukui function in the electronic theory of metals and catalysis. , author=. Proceedings of the National Academy of Sciences , volume=. 1985 , publisher=

    1985

  57. [57]

    The Twelfth International Conference on Learning Representations , year=

    Complete and Efficient Graph Transformers for Crystal Material Property Prediction , author=. The Twelfth International Conference on Learning Representations , year=

  58. [58]

    The Journal of chemical physics , volume=

    Schnet--a deep learning architecture for molecules and materials , author=. The Journal of chemical physics , volume=. 2018 , publisher=

    2018

  59. [59]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    A diffusion-based pre-training framework for crystal property prediction , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  60. [60]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    A Denoising Pre-training Framework for Accelerating Novel Material Discovery , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  61. [61]

    Advances in neural information processing systems , volume=

    Generative modeling by estimating gradients of the data distribution , author=. Advances in neural information processing systems , volume=

  62. [62]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Score-based generative modeling through stochastic differential equations , author=. arXiv preprint arXiv:2011.13456 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  63. [63]

    Advances in neural information processing systems , volume=

    Improved techniques for training score-based generative models , author=. Advances in neural information processing systems , volume=

  64. [64]

    Nature Machine Intelligence , volume=

    CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling , author=. Nature Machine Intelligence , volume=. 2023 , publisher=

    2023

  65. [65]

    Advances in neural information processing systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in neural information processing systems , volume=

  66. [66]

    Advances in Neural Information Processing Systems , volume=

    Structured denoising diffusion models in discrete state-spaces , author=. Advances in Neural Information Processing Systems , volume=

  67. [67]

    All-atom

    Joshi, Chaitanya K and Fu, Xiang and Liao, Yi-Lun and Gharakhanyan, Vahe and Miller, Benjamin Kurt and Sriram, Anuroop and Ulissi, Zachary Ward , booktitle=. All-atom. 2025 , organization=

    2025

  68. [68]

    Progress in Materials Science , volume=

    Recent progress in metallurgical thermochemistry , author=. Progress in Materials Science , volume=. 1969 , publisher=

    1969

  69. [69]

    1993 , publisher=

    Materials Thermochemistry , author=. 1993 , publisher=

    1993

  70. [70]

    Physical Review B , volume=

    Properties of intrinsic point defects in silicon determined by zinc diffusion experiments under nonequilibrium conditions , author=. Physical Review B , volume=. 1995 , publisher=

    1995

  71. [71]

    Progress in Energy and Combustion Science , volume=

    Understanding NOx formation in nonpremixed flames: experiments and modeling , author=. Progress in Energy and Combustion Science , volume=. 1995 , publisher=

    1995

  72. [72]

    npj Computational Materials , volume=

    The Open Quantum Materials Database (OQMD): assessing the accuracy of DFT formation energies , author=. npj Computational Materials , volume=. 2015 , publisher=

    2015

  73. [73]

    2008 , publisher=

    Group Theory: Application to the Physics of Condensed Matter , author=. 2008 , publisher=

    2008

  74. [74]

    2016 , publisher=

    Group theory in a nutshell for physicists , author=. 2016 , publisher=

    2016

  75. [75]

    Physical review , volume=

    Self-consistent equations including exchange and correlation effects , author=. Physical review , volume=. 1965 , publisher=

    1965

  76. [76]

    Physical Review B , volume=

    How to represent crystal structures for machine learning: Towards fast prediction of electronic properties , author=. Physical Review B , volume=. 2014 , publisher=

    2014

  77. [77]

    International conference on machine learning , pages=

    Equivariant diffusion for molecule generation in 3d , author=. International conference on machine learning , pages=. 2022 , organization=

    2022

  78. [78]

    arXiv preprint arXiv:2209.15408 , year=

    Equivariant energy-guided sde for inverse molecular design , author=. arXiv preprint arXiv:2209.15408 , year=

    work page arXiv

  79. [79]

    arXiv preprint arXiv:1909.00949 , year=

    Data-driven approach to encoding and decoding 3-d crystal structures , author=. arXiv preprint arXiv:1909.00949 , year=

    work page arXiv 1909

  80. [80]

    Matter , volume=

    Inverse design of solid-state materials via a continuous representation , author=. Matter , volume=. 2019 , publisher=

    2019

Showing first 80 references.