pith. sign in

arxiv: 2605.14769 · v2 · pith:HHH5LJUQnew · submitted 2026-05-14 · 💻 cs.LG

Composable Crystals: Controllable Materials Discovery via Concept Learning

Nian Liu , Yuwei Zeng , Ryoji Kubo , Nikita Kazeev , Stephen Gregory Dale , Artem Maevskiy , Pengru Huang , Thomas Laurent
show 2 more authors
Kostya S. Novoselov Xavier Bresson
This is my paper

Pith reviewed 2026-06-30 21:50 UTC · model grok-4.3

classification 💻 cs.LG
keywords crystal generationconcept learningVQ-VAEmaterials discoverycompositional modelsde novo designgenerative models
0
0 comments X

The pith

Recombining concepts discovered by a VQ-VAE allows controllable generation of novel crystals outside the training distribution.

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

The paper establishes a framework in which a vector-quantized variational autoencoder automatically finds a shared set of reusable concepts from crystal structures. These concepts act as building blocks that can be recombined to guide the creation of new crystals. This approach aims to move beyond black-box random sampling by providing control over how generated crystals differ from known ones. Experiments demonstrate that using these compositions improves performance on validity, stability, uniqueness, and novelty metrics, particularly boosting novelty.

Core claim

The central claim is that training a VQ-VAE on crystal data reveals interpretable concepts based on local atomic environments and global symmetry patterns, which generalize across distributions and can be recombined using a composition generator to produce valid, stable, unique, and novel crystals beyond the training set.

What carries the argument

The vector-quantized variational autoencoder that discovers a shared set of reusable crystal concepts serving as building blocks for guided generation.

If this is right

  • Controllable exploration of novel crystals rather than unconstrained random sampling.
  • Up to 53.2% and 51.7% improvement on the V.S.U.N metric on MP-20 and Alex-MP-20 datasets.
  • Particular gains in novelty of generated crystals.
  • Concepts exhibit interpretability from local and global patterns and generalize to different crystal distributions.

Where Pith is reading between the lines

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

  • The iterative refinement of the composition generator using the model's own high-quality samples could create a feedback loop for continuous improvement in generation quality.
  • If the concepts are generalizable, the method might reduce reliance on massive training datasets for materials discovery tasks.
  • The shared concepts could enable transfer between different crystal datasets without full retraining.

Load-bearing premise

The learned concepts from the VQ-VAE can be recombined into valid and stable crystals that do not require additional post-processing or retraining to achieve the claimed controllability.

What would settle it

A test where many concept-recombined crystals turn out to be invalid or unstable, or where the novelty gains disappear when compared to an improved base model without concepts.

Figures

Figures reproduced from arXiv: 2605.14769 by Artem Maevskiy, Kostya S. Novoselov, Nian Liu, Nikita Kazeev, Pengru Huang, Ryoji Kubo, Stephen Gregory Dale, Thomas Laurent, Xavier Bresson, Yuwei Zeng.

Figure 1
Figure 1. Figure 1: Pipelines for (a) learning a VQ-VAE to extract crystal concepts, and (b) training and refining [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Transformer architectures used in (a) the encoder and (b) the decoder of the VQ-VAE, as [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustrations of two learned concepts represented by VQ-VAE codes. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Cosine similarities of concept selection [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustrations of two generated crystals together with their supporting concepts. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The whole pipeline of training VQ-VAE. Directly training a VQ-VAE on crystal structures led to limited reconstruction performance in our preliminary experiments. Therefore, we adopt a three-stage training strategy. We first train a variational autoencoder (VAE) to learn a continuous latent space for local atomic environments. We then apply K-Means clustering to the learned latent embeddings to obtain discr… view at source ↗
read the original abstract

De novo crystal generation, a central task in materials discovery, aims to generate crystals that are simultaneously valid, stable, unique, and novel. Existing methods mainly rely on black-box stochastic sampling, providing limited control over how generated structures move beyond the observed distribution. In this paper, we introduce a concept-based compositional framework for crystal generation. We train a vector-quantized variational autoencoder to automatically discover a shared set of reusable crystal concepts, which serve as building blocks for guided generation. These learned concepts naturally exhibit interpretability from both local atomic environments and global symmetry patterns, and generalize to crystals from different distributions. By recombining such concepts, our framework enables controllable exploration of novel crystals beyond the training distribution, rather than relying solely on unconstrained random sampling. To further improve composition efficiency, we introduce a composition generator and iteratively refine it using high-quality samples generated by the model itself. The resulting concept compositions are then used to condition downstream crystal generation. Numerical experiments on MP-20 and Alex-MP-20 show that compositing concepts separately increase base model up to 53.2% and 51.7% on V.S.U.N metric, with particular gains in novelty.

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 a concept-based framework for de novo crystal generation. A VQ-VAE is trained to discover reusable, interpretable concepts from crystal structures that generalize across distributions. These concepts are recombined via a composition generator that is iteratively refined on self-generated high-quality samples; the resulting compositions condition a downstream generator. Experiments on MP-20 and Alex-MP-20 report that concept compositing improves a base model by up to 53.2% and 51.7% on the V.S.U.N. metric, with largest gains in novelty, enabling controllable exploration beyond the training distribution.

Significance. If the central claims hold without hidden selection effects, the work would offer a concrete advance over black-box sampling methods by providing interpretable, reusable building blocks and explicit controllability. The VQ-VAE concept discovery and reported generalization to out-of-distribution crystals are potentially valuable contributions to materials informatics.

major comments (2)
  1. [Abstract / Methods (composition generator)] Abstract (and the description of the composition generator): the iterative refinement step that uses 'high-quality samples generated by the model itself' to train the composition generator is load-bearing for the controllability claim. It is unclear whether any validity/stability/quality filter is applied during this self-supervised loop; if so, the reported V.S.U.N. gains (53.2% / 51.7%) cannot be attributed solely to recombination of the learned VQ-VAE concepts and the weakest assumption (generalizability without post-hoc mechanisms) is not yet supported.
  2. [Abstract / Experiments] Abstract: the claim that 'compositing concepts separately increase base model up to 53.2% and 51.7% on V.S.U.N metric, with particular gains in novelty' requires an ablation that isolates the effect of concept recombination from any curation performed during iterative refinement. Without such a controlled comparison, the numerical results do not yet establish that the framework enables exploration 'rather than relying solely on unconstrained random sampling.'
minor comments (1)
  1. [Abstract] The V.S.U.N. metric is referenced without an explicit definition or citation in the provided abstract; a clear statement of its components and how novelty is measured would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below and commit to revisions that improve clarity and strengthen the supporting evidence.

read point-by-point responses

  1. Referee: [Abstract / Methods (composition generator)] Abstract (and the description of the composition generator): the iterative refinement step that uses 'high-quality samples generated by the model itself' to train the composition generator is load-bearing for the controllability claim. It is unclear whether any validity/stability/quality filter is applied during this self-supervised loop; if so, the reported V.S.U.N. gains (53.2% / 51.7%) cannot be attributed solely to recombination of the learned VQ-VAE concepts and the weakest assumption (generalizability without post-hoc mechanisms) is not yet supported.

    Authors: We agree that the description of the iterative refinement requires greater precision. The manuscript currently states only that the composition generator is refined on 'high-quality samples generated by the model itself,' without enumerating the exact selection criteria. In the revised version we will expand the Methods section to specify the precise validity, stability, and quality thresholds (if any) applied during self-supervised iteration, thereby clarifying whether post-hoc mechanisms beyond concept recombination are involved. revision: yes

  2. Referee: [Abstract / Experiments] Abstract: the claim that 'compositing concepts separately increase base model up to 53.2% and 51.7% on V.S.U.N metric, with particular gains in novelty' requires an ablation that isolates the effect of concept recombination from any curation performed during iterative refinement. Without such a controlled comparison, the numerical results do not yet establish that the framework enables exploration 'rather than relying solely on unconstrained random sampling.'

    Authors: We acknowledge the value of an explicit ablation that separates the contribution of concept recombination from any curation effects of the iterative loop. We will add this controlled comparison in the revised manuscript, reporting V.S.U.N. metrics for (i) the base generator, (ii) the composition generator without iterative refinement, and (iii) the full pipeline, thereby isolating the impact of learned-concept recombination. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The provided abstract and description contain no equations, parameter-fitting steps, or self-referential reductions where a claimed prediction or result is equivalent to its inputs by construction. The iterative refinement of the composition generator is described at a high level without exhibiting a specific loop that forces outputs from inputs (e.g., no fitted parameter renamed as prediction). No self-citations, uniqueness theorems, or ansatzes are quoted. The framework's claims rest on empirical numerical experiments rather than a closed definitional chain, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Full text unavailable; cannot enumerate free parameters, axioms, or invented entities from abstract alone.

pith-pipeline@v0.9.1-grok · 5772 in / 1144 out tokens · 20704 ms · 2026-06-30T21:50:09.894916+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

44 extracted references · 20 canonical work pages · 6 internal anchors

  1. [1]

    Accurate structure prediction of biomolecular interactions with alphafold 3.Nature, 630(8016):493–500, 2024

    Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J Ballard, Joshua Bambrick, et al. Accurate structure prediction of biomolecular interactions with alphafold 3.Nature, 630(8016):493–500, 2024

    2024

  2. [2]

    Specformer: Spectral graph neural networks meet transformers.arXiv preprint arXiv:2303.01028, 2023

    Deyu Bo, Chuan Shi, Lele Wang, and Renjie Liao. Specformer: Spectral graph neural networks meet transformers.arXiv preprint arXiv:2303.01028, 2023

    work page arXiv 2023

  3. [3]

    Space group informed transformer for crystalline materials generation, 2024.URL https://arxiv

    Zhendong Cao, Xiaoshan Luo, Jian Lv, and Lei Wang. Space group informed transformer for crystalline materials generation, 2024.URL https://arxiv. org/abs/2403.15734, 11

    work page arXiv 2024

  4. [4]

    Space group equivariant crystal diffusion.arXiv preprint arXiv:2505.10994, 2025

    Rees Chang, Angela Pak, Alex Guerra, Ni Zhan, Nick Richardson, Elif Ertekin, and Ryan P Adams. Space group equivariant crystal diffusion.arXiv preprint arXiv:2505.10994, 2025

    work page arXiv 2025

  5. [5]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  6. [6]

    The faiss library

    Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre- Emmanuel Mazaré, Maria Lomeli, Lucas Hosseini, and Hervé Jégou. The faiss library. 2024

    2024

  7. [7]

    A generalization of transformer networks to graphs

    Vijay Prakash Dwivedi and Xavier Bresson. A generalization of transformer networks to graphs. arXiv preprint arXiv:2012.09699, 2020

    work page arXiv 2012

  8. [8]

    Alex M. Ganose, Hrushikesh Sahasrabuddhe, Mark Asta, Kevin Beck, Tathagata Biswas, Alexan- der Bonkowski, Joana Bustamante, Xin Chen, Yuan Chiang, Daryl Chrzan, Jacob Clary, Orion Cohen, Christina Ertural, Max Gallant, Janine George, Sophie Gerits, Rhys Goodall, Rishabh Guha, Geoffroy Hautier, Matthew Horton, Aaron Kaplan, Ryan Kingsbury, Matthew Kuner, B...

    2025

  9. [9]

    Fine-tuned language models generate stable inorganic materials as text, arxiv, 2024.arXiv preprint arXiv:2402.04379, 10

    N Gruver, A Sriram, A Madotto, AG Wilson, CL Zitnick, and Z Ulissi. Fine-tuned language models generate stable inorganic materials as text, arxiv, 2024.arXiv preprint arXiv:2402.04379, 10

    work page arXiv 2024

  10. [10]

    Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  11. [11]

    Classifier-Free Diffusion Guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  12. [12]

    GPT-4o System Card

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv preprint arXiv:2410.21276, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  13. [13]

    Commentary: The materials project: A materials genome approach to accelerating materials innovation.APL materials, 1(1), 2013

    Anubhav Jain, Shyue Ping Ong, Geoffroy Hautier, Wei Chen, William Davidson Richards, Stephen Dacek, Shreyas Cholia, Dan Gunter, David Skinner, Gerbrand Ceder, et al. Commentary: The materials project: A materials genome approach to accelerating materials innovation.APL materials, 1(1), 2013

    2013

  14. [14]

    Crystal structure prediction by joint equivariant diffusion.Advances in Neural Information Processing Systems, 36:17464–17497, 2023

    Rui Jiao, Wenbing Huang, Peijia Lin, Jiaqi Han, Pin Chen, Yutong Lu, and Yang Liu. Crystal structure prediction by joint equivariant diffusion.Advances in Neural Information Processing Systems, 36:17464–17497, 2023

    2023

  15. [15]

    Space group constrained crystal generation.arXiv preprint arXiv:2402.03992, 2024

    Rui Jiao, Wenbing Huang, Yu Liu, Deli Zhao, and Yang Liu. Space group constrained crystal generation.arXiv preprint arXiv:2402.03992, 2024. 11

    work page arXiv 2024

  16. [16]

    All-atom diffusion transformers: Unified generative modelling of molecules and materials.arXiv preprint arXiv:2503.03965, 2025

    Chaitanya K Joshi, Xiang Fu, Yi-Lun Liao, Vahe Gharakhanyan, Benjamin Kurt Miller, Anuroop Sriram, and Zachary W Ulissi. All-atom diffusion transformers: Unified generative modelling of molecules and materials.arXiv preprint arXiv:2503.03965, 2025

    work page arXiv 2025

  17. [17]

    Llm meets diffusion: A hybrid framework for crystal material generation

    Subhojyoti Khastagir, Kishalay Das, Pawan Goyal, Seung-Cheol Lee, Satadeep Bhattacharjee, and Niloy Ganguly. Llm meets diffusion: A hybrid framework for crystal material generation. arXiv preprint arXiv:2510.23040, 2025

    work page arXiv 2025

  18. [18]

    Kresse and J

    G. Kresse and J. Furthmüller. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set.Phys. Rev. B, 54:11169–11186, Oct 1996

    1996

  19. [19]

    Kresse and D

    G. Kresse and D. Joubert. From ultrasoft pseudopotentials to the projector augmented-wave method.Phys. Rev. B, 59:1758–1775, Jan 1999

    1999

  20. [20]

    Symmcd: Symmetry-preserving crystal generation with diffusion models.arXiv preprint arXiv:2502.03638, 2025

    Daniel Levy, Siba Smarak Panigrahi, Sékou-Oumar Kaba, Qiang Zhu, Kin Long Kelvin Lee, Mikhail Galkin, Santiago Miret, and Siamak Ravanbakhsh. Symmcd: Symmetry-preserving crystal generation with diffusion models.arXiv preprint arXiv:2502.03638, 2025

    work page arXiv 2025

  21. [21]

    Flowmm: Generating materials with riemannian flow matching

    Benjamin Kurt Miller, Ricky TQ Chen, Anuroop Sriram, and Brandon M Wood. Flowmm: Generating materials with riemannian flow matching. InForty-first International Conference on Machine Learning, 2024

    2024

  22. [22]

    Improved denoising diffusion probabilistic models

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. InInternational conference on machine learning, pages 8162–8171. PMLR, 2021

    2021

  23. [24]

    Python materials genomics (pymatgen): A robust, open-source python library for materials analysis.Computational Materials Science, 68:314–319, 2013

    Shyue Ping Ong, William Davidson Richards, Anubhav Jain, Geoffroy Hautier, Michael Kocher, Shreyas Cholia, Dan Gunter, Vincent L Chevrier, Kristin A Persson, and Gerbrand Ceder. Python materials genomics (pymatgen): A robust, open-source python library for materials analysis.Computational Materials Science, 68:314–319, 2013

    2013

  24. [25]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023

    2023

  25. [26]

    Perdew, Kieron Burke, and Matthias Ernzerhof

    John P. Perdew, Kieron Burke, and Matthias Ernzerhof. Generalized gradient approximation made simple.Phys. Rev. Lett., 77:3865–3868, Oct 1996

    1996

  26. [27]

    High- resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High- resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  27. [28]

    Improving machine-learning models in materials science through large datasets.Materials Today Physics, 48:101560, 2024

    Jonathan Schmidt, Tiago FT Cerqueira, Aldo H Romero, Antoine Loew, Fabian Jäger, Hai-Chen Wang, Silvana Botti, and Miguel AL Marques. Improving machine-learning models in materials science through large datasets.Materials Today Physics, 48:101560, 2024

    2024

  28. [29]

    Deep unsuper- vised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsuper- vised learning using nonequilibrium thermodynamics. InInternational conference on machine learning, pages 2256–2265. pmlr, 2015

    2015

  29. [30]

    Flowllm: Flow matching for material generation with large language models as base distributions.Advances in Neural Information Processing Systems, 37:46025–46046, 2024

    Anuroop Sriram, Benjamin K Miller, Ricky T Chen, and Brandon M Wood. Flowllm: Flow matching for material generation with large language models as base distributions.Advances in Neural Information Processing Systems, 37:46025–46046, 2024

    2024

  30. [31]

    What-if analysis of large language models: Explore the game world using proactive thinking.arXiv preprint arXiv:2509.04791, 2025

    Yuan Sui, Yanming Zhang, Yi Liao, Yu Gu, Guohua Tang, Zhongqian Sun, Wei Yang, and Bryan Hooi. What-if analysis of large language models: Explore the game world using proactive thinking.arXiv preprint arXiv:2509.04791, 2025

    work page arXiv 2025

  31. [32]

    LLaMA: Open and Efficient Foundation Language Models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timo- thée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971, 2023. 12

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  32. [33]

    Neural discrete representation learning.Advances in neural information processing systems, 30, 2017

    Aaron Van Den Oord, Oriol Vinyals, et al. Neural discrete representation learning.Advances in neural information processing systems, 30, 2017

    2017

  33. [34]

    Attention is all you need.Advances in Neural Information Processing Systems, 2017

    A Vaswani. Attention is all you need.Advances in Neural Information Processing Systems, 2017

    2017

  34. [35]

    Conceptmix: A compositional image generation benchmark with controllable difficulty.Advances in Neural Information Processing Systems, 37:86004–86047, 2024

    Xindi Wu, Dingli Yu, Yangsibo Huang, Olga Russakovsky, and Sanjeev Arora. Conceptmix: A compositional image generation benchmark with controllable difficulty.Advances in Neural Information Processing Systems, 37:86004–86047, 2024

    2024

  35. [36]

    Crys- tal diffusion variational autoencoder for periodic material generation.arXiv preprint arXiv:2110.06197, 2021

    Tian Xie, Xiang Fu, Octavian-Eugen Ganea, Regina Barzilay, and Tommi Jaakkola. Crys- tal diffusion variational autoencoder for periodic material generation.arXiv preprint arXiv:2110.06197, 2021

    work page arXiv 2021

  36. [37]

    Less is more: on the over- globalizing problem in graph transformers.arXiv preprint arXiv:2405.01102, 2024

    Yujie Xing, Xiao Wang, Yibo Li, Hai Huang, and Chuan Shi. Less is more: on the over- globalizing problem in graph transformers.arXiv preprint arXiv:2405.01102, 2024

    work page arXiv 2024

  37. [38]

    On layer normalization in the transformer architecture

    Ruibin Xiong, Yunchang Yang, Di He, Kai Zheng, Shuxin Zheng, Chen Xing, Huishuai Zhang, Yanyan Lan, Liwei Wang, and Tieyan Liu. On layer normalization in the transformer architecture. InInternational conference on machine learning, pages 10524–10533. PMLR, 2020

    2020

  38. [39]

    Geometric latent diffusion models for 3d molecule generation

    Minkai Xu, Alexander S Powers, Ron O Dror, Stefano Ermon, and Jure Leskovec. Geometric latent diffusion models for 3d molecule generation. InInternational Conference on Machine Learning, pages 38592–38610. PMLR, 2023

    2023

  39. [40]

    MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures

    Han Yang, Chenxi Hu, Yichi Zhou, Xixian Liu, Yu Shi, Jielan Li, Guanzhi Li, Zekun Chen, Shuizhou Chen, Claudio Zeni, et al. Mattersim: A deep learning atomistic model across elements, temperatures and pressures.arXiv preprint arXiv:2405.04967, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  40. [41]

    Skill-mix: A flexible and expandable family of evaluations for ai models.arXiv preprint arXiv:2310.17567, 2023

    Dingli Yu, Simran Kaur, Arushi Gupta, Jonah Brown-Cohen, Anirudh Goyal, and Sanjeev Arora. Skill-mix: A flexible and expandable family of evaluations for ai models.arXiv preprint arXiv:2310.17567, 2023

    work page arXiv 2023

  41. [42]

    Mattergen: a generative model for inorganic materials design.arXiv preprint arXiv:2312.03687, 2023

    Claudio Zeni, Robert Pinsler, Daniel Zügner, Andrew Fowler, Matthew Horton, Xiang Fu, Sasha Shysheya, Jonathan Crabbé, Lixin Sun, Jake Smith, et al. Mattergen: a generative model for inorganic materials design.arXiv preprint arXiv:2312.03687, 2023

    work page arXiv 2023

  42. [43]

    Can large language models improve the adversarial robustness of graph neural networks? InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    Zhongjian Zhang, Xiao Wang, Huichi Zhou, Yue Yu, Mengmei Zhang, Cheng Yang, and Chuan Shi. Can large language models improve the adversarial robustness of graph neural networks? InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 1, pages 2008–2019, 2025

    2008

  43. [44]

    Graph-GRPO: Training Graph Flow Models with Reinforcement Learning

    Baoheng Zhu, Deyu Bo, Delvin Ce Zhang, and Xiao Wang. Graph-grpo: Training graph flow models with reinforcement learning.arXiv preprint arXiv:2603.10395, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  44. [45]

    Local structure order parameters and site fingerprints for quantification of coordination environment and crystal structure similarity.RSC advances, 10(10):6063–6081, 2020

    Nils ER Zimmermann and Anubhav Jain. Local structure order parameters and site fingerprints for quantification of coordination environment and crystal structure similarity.RSC advances, 10(10):6063–6081, 2020. 13 A Visualizing More Local Patterns encoded in Concepts In Section 4, we visualize the top-5 local atomic environments associated with two learned...

    2020