Pith. sign in

REVIEW 5 major objections 6 minor 1 cited by

MetasurfaceViT: A generic AI model for metasurface inverse design

T0 review · 5 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A single Vision Transformer, pretrained on 60 million Jones-matrix samples, performs one-shot metasurface inverse design for arbitrary wavelengths and polarizations with accuracy above 99% on realistic targets.

desk verdict A plausible masked-autoencoder recipe for Jones-matrix inverse design with a large simulated dataset, but the headline 99% accuracy is measured against the same simulator used for training and needs independent validation before it earns the 'generic' label. read the letter →

arxiv 2504.14895 v1 pith:ULNGSXNS submitted 2025-04-21 physics.optics

Jiahao Yan , Jilong Yi , Churong Ma , Yanjun Bao , Qin Chen , Baojun Li This is my paper · ORCID
classification physics.optics
keywords metasurfaceinversedesignVisionTransformerJonesmatrixmaskedpretrainingphysics-informeddataaugmentationpolarizationmultiplexingbroadbandmetalensnanophotonics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

MetasurfaceViT is a single Vision Transformer trained on 60 million Jones-matrix samples for unit cells made of two rotated, paired silicon nanopillars on silica. The central claim is that, after masked pretraining over wavelength and polarization channels and fine-tuning to structural parameters, this one model performs one-shot inverse design for arbitrary wavelengths, polarizations, and application scenarios, producing structural parameters directly from a target Jones matrix (a $20\times6$ array of three amplitude and three phase components across twenty visible wavelengths). The reported evidence is inverse-design L1 loss of 0.0025 on the test set, tandem forward-prediction loss of 0.0048, and accuracy above 99% for physically realistic target matrices, with over 85% accuracy even for ideal matrices. If correct, this removes the need for a separate per-wavelength or per-polarization network and makes the Jones matrix the universal interface between optical function and structure.

What carries the argument

The central object is the unit cell's Jones matrix, coded as a $20\times6$ array that stacks three amplitude components and three phase components across twenty visible wavelengths. A Jones matrix records how a structure transforms the polarization and phase of transmitted light; by rotating and pairing two silicon nanopillars, the authors obtain a family of unit cells whose full Jones matrix is computed from finite-difference time-domain simulated single-pillar responses via matrix operations, yielding tens of millions of labeled samples. The carrying mechanism is masked pretraining over that matrix: the network learns to reconstruct deliberately hidden wavelength rows and polarization components, and this forces it to internalize spectral and polarization correlations that transfer to inverse design after fine-tuning. The ViT encoder uses multi-head self-attention over the projected $20\times6$ entries rather than image patches, and a tandem forward network maps the predicted structural parameters back to a Jones matrix so the designed response can be checked before fabrication.

What would settle it

Fabricate a set of the paper's designed devices, such as the 18-channel multiplexed hologram and the 90-micrometer broadband metalens, and measure their optical response across the six visible wavelengths. If measured channel crosstalk is substantially worse than the simulated holograms, or if the metalens focal length shifts beyond the simulated chromatic aberration, the simulated unit-cell library does not transfer to real devices and the reported accuracy claims would not hold outside simulation.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a wavelength- and polarization-resolved Jones matrix can serve as a universal input-output language for metasurface inverse design when paired with masked autoencoding pretraining. The model treats a target optical response as a $20\times6$ matrix and learns to complete missing entries before being fine-tuned to output the six unit-cell size and orientation parameters. This makes the design task independent of the number of wavelength or polarization channels: any requirement expressible as one or more entries of the matrix can be fed into the same network. The paper verifies this with single-wavelength six-channel printing and hologram multiplexing, 18-channel three-wavelength multiplexing, and a broadband achromatic metalens working at six visible wavelengths, reporting test-set inverse-design loss 0.0025 and tandem loss 0.0048, alongside 99% accuracy on realistic targets and 85% on ideal ones. It also reports that ideal designed Jones matrices are not fully realizable, with L1 loss around 0.13 to 0.15, which it attributes to idealized targets exceeding what the physical unit cell can produce.

Load-bearing premise

The load-bearing premise is that the finite-difference time-domain simulation of a single silicon nanopillar, combined with the analytic rotation and pairing that build the unit cell, faithfully represents what can actually be fabricated and measured across the whole design space; the paper verifies this only in simulation, with no fabricated device.

Editorial extensions

If this is right

  • One-shot design: any target Jones matrix, or a partially specified one, maps to structural parameters through the same fine-tuned model, so new wavelengths or polarization combinations do not require retraining.
  • Higher channel capacity: because each wavelength carries six independent Jones components, the model realizes six-channel multiplexing at one wavelength and eighteen channels at three wavelengths in a single pass, and the capacity scales with the size of the matrix.
  • Built-in validation: the tandem forward predictor gives an immediate check of the designed response before fabrication and quantifies a gap (L1 around 0.13 to 0.15) for ideal targets, identifying when a requested response is not physically realizable.
  • Broadband optics: the workflow yields a 90-micrometer-focal-length metalens focusing across 434 to 760 nm with sub-micrometer full width at half maximum, with chromatic focal shift remaining as a reported limitation.
  • Computational efficiency: a 256-by-256 metasurface with 65,536 unit cells is generated in seconds, which makes iterative design loops practical.

Reading between the lines

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

  • Editorial inference: the same masked-pretraining recipe should transfer to other unit-cell families, such as freeform, multilayer, or tunable structures, because the Jones-matrix format is independent of the specific geometry that generated the training library.
  • Editorial inference: the pretrained reconstructor could operate backwards as a measurement-imputation tool, filling missing wavelength or polarization entries from partial polarimetric data rather than from design targets.
  • Editorial inference: the reported gap between test-set loss (0.0025) and designed-matrix loss (0.13 to 0.15) yields a testable prediction: if target designs are constrained to the physically achievable amplitude-phase manifold of the unit cell, end-to-end accuracy should approach the test-set level; if it does not, the residual error sits in the unit-cell forward model rather than in the inverse n
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The manuscript introduces MetasurfaceViT, a Vision-Transformer-based model for metasurface inverse design. The authors construct a dataset of over 60 million Jones-matrix samples derived from FDTD simulations of paired and rotated silicon nanopillars, pretrain a ViT to reconstruct masked wavelength and polarization channels, and then fine-tune it to map Jones matrices to structural parameters. A tandem workflow with a forward predictor is used to evaluate the designs. The paper claims one-shot inverse design for arbitrary wavelengths, polarizations, and application scenarios, with 'prediction accuracy' exceeding 99% for physically realistic targets and 85% for ideal targets. Two applications are demonstrated: 6-channel and 18-channel multiplexing of printings and holograms, and a broadband achromatic metalens evaluated by 3D FDTD.

Significance. If the claims were fully supported, this would be a notable step toward a foundation-model-style approach for metasurface inverse design: a single pretrained model that handles multiple wavelengths, polarizations, and application types. The use of Jones matrices as a unified representation, the masking-based pretraining strategy, the large data-generation effort, and the public code release are all potentially valuable contributions. The metalens demonstration with full-wave 3D FDTD is a positive attempt at validation. However, the central quantitative claims are not yet established: the meaning of '99% accuracy' is never defined, the reported losses are mostly internal consistency checks against the same simulation family used for training, and no fabricated device or independent physical measurement is presented. The contribution is therefore promising but requires substantially stronger validation and more careful claims before it can be accepted as stated.

major comments (5)
  1. [Abstract; §2.3, Fig. 3d] The headline claim that 'prediction accuracy can exceed 99%' is never defined. Section 2.3 reports L1 losses of 0.0025 for inverse design and 0.0048 for the tandem workflow on test-set Jones matrices, and losses of about 0.13–0.15 for designed (ideal) Jones matrices. If 'accuracy' is being computed as (1 − L1) × 100%, this is not a physically meaningful accuracy metric, since L1 on a mixture of amplitudes and phases (with different units and scales) cannot be converted to a percentage accuracy without a stated normalization. The authors should define the metric precisely, report error bars and sample sizes, and avoid presenting an unnormalized loss as a percentage accuracy.
  2. [§2.3, §2.4, Fig. 3a] The validation of the multiplexing applications is essentially circular. Evaluation Workflow-2 uses a lightweight predictor and a matcher that are trained on the same FDTD-generated data used to train the inverse model; the reported Type1 and Type2 losses therefore measure how consistently the model inverts its own learned simulator, not how accurately the designed metasurface performs physically. For non-metalens applications no independent FDTD simulation or fabricated device is reported. An external validation set generated by full-wave simulation for at least a few representative multiplexing designs is needed to support the claimed physical accuracy.
  3. [§2.3, §2.5, Conclusion] The claim of 'arbitrary wavelength, polarization, and application requirements' is contradicted by the paper's own results. Section 2.3 states that ideal designed Jones matrices 'may not be fully realizable in practice,' and Section 2.5 reports residual chromatic aberration in the broadband achromatic metalens, attributed to strong amplitude constraints. These admissions indicate that the model does not map arbitrary targets to valid physical devices. The claims should be restricted to in-distribution or physically realizable targets, and the realizability limitations should be discussed in the abstract and conclusion rather than only in the results.
  4. [§2.1, §4.3] The description of the 60-million-sample dataset and the 'physics-informed data augmentation' is insufficient to assess generalizability. The paper does not specify the parameter ranges for widths, lengths, and angles, the discretization step sizes, the fabrication constraints (e.g., minimum gap, aspect ratio), the number of FDTD simulations actually performed versus analytically rotated/paired matrices, or how the 1M fine-tuning dataset is sampled from the 60M set. These details are essential because the claimed universality depends entirely on the coverage and physical validity of the training distribution.
  5. [§2.3, Fig. 3d] All reported losses are single point estimates without error bars, standard deviations, or information about the number of test samples. In particular, the distinction between Type1 and Type2 losses is described only in the figure caption, and the text does not state how many designed Jones matrices were evaluated or how the 'designed' targets were generated. Statistical characterization is needed to determine whether the differences between the test-set losses (0.0025) and designed-target losses (0.13–0.15) are meaningful or dominated by target distribution mismatch.
minor comments (6)
  1. [Throughout] The manuscript contains many typographical errors and broken characters (e.g., 'üit cell,' 'ünitäre,' 'ürstly,' 'achormatic' in the Section 2.5 title, and 'nm' instead of 'nm' in several places). A careful proofread is needed.
  2. [Abstract; §1] The phrase 'revolutionary' and 'game-changing' is used repeatedly (Abstract, Introduction, Conclusion). Such promotional language is not appropriate for a technical journal report and should be replaced with concrete descriptions of the contribution.
  3. [§2.2, Fig. 2a] The explanation of the masking strategies in the text does not fully match the figure. For example, Mask-type-2 is described as keeping all amplitudes but phases at one wavelength, while the figure should state which amplitude components are kept. Please clarify the correspondence between the text, the figure, and the actual implementation.
  4. [§2.3] The statement that 'the model can achieve a very low loss after 50 epochs' would benefit from showing the actual loss curves with labeled axes and legends in Figure 3b and 3c; the current figures are difficult to interpret without axis labels.
  5. [§4.3] The Gerchberg-Saxton algorithm is mentioned but not cited, and the exact implementation (e.g., number of iterations, initialization, constraints) is not described. Please add a reference and implementation details.
  6. [§2.4] The 18-channel demonstration is acknowledged to have 'high noise levels' and unclear images in some channels. The authors attribute this to crosstalk from idealized designs, but a quantitative evaluation (e.g., correlation with target images) is missing. Such a metric would help assess the actual performance.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline 99%/85% accuracy is a round-trip self-consistency metric: the inverse network is evaluated against a forward predictor trained on the same FDTD-generated Jones-matrix distribution, not against independent physical validation.

  1. fitted input called prediction [Section 2.3, 'Workflow of Finetuning, Prediction, and Evaluation' (Fig. 3d); Section 2.4]
    "Evaluation Workflow-2, on the other hand, is for conventional metasurfaces, using a predictor (a lightweight trained network) and a matcher (searching in pre-trained data) to map structural parameters to Jones Matrices, followed by visualization steps. ... The low L1 loss of 0.0025 for the inverse design (Jones Matrices -> size parameters) and 0.0048 for the tandem workflow (Jones Matrices -> size parameters -> Jones Matrices) on the test set demonstrates the effectiveness of the model."

    The abstract's claim that 'prediction accuracy can exceed 99%' is drawn from the tandem-workflow L1 losses (JM -> size -> JM), where the second leg is a trained predictor/matcher built from the same FDTD-generated Jones-matrix dataset used to train the inverse network. Both fitted functions are therefore tested against the same simulation surrogate that produced the training data; the low loss measures self-consistency between two fitted functions, not agreement between the predicted metasurface and an independently computed or measured optical response. This is a fitted surrogate being renamed as an 'evaluation' of physical prediction, so the central quantitative claim reduces to an internal consistency check.

full rationale

The machine-learning procedure itself is not circular in the ordinary train/test sense: a held-out test set drawn from the same distribution is legitimate evidence of generalization to that distribution. The circularity lies in how the result is labeled. The paper's 'prediction accuracy' is essentially (1 - L1) on Jones matrices that pass through both the inverse model and a learned forward predictor/matcher, both fitted to the same simulation family. No fabricated device is measured, and the multiplexing demonstrations use the fitted predictor rather than full-wave FDTD to generate the reported images and losses. The only independent full-wave check, Evaluation Workflow-1 for the broadband metalens, is not the source of the 99% claim and in fact shows residual chromatic aberration. The paper itself reports L1 losses of 0.13-0.15 for user-defined ideal targets and attributes them to targets 'not fully realizable in practice,' which undercuts the abstraction-level 'arbitrary wavelength, polarization, and application' claim. Thus one or more of the paper's central quantitative predictions reduce, by construction, to self-consistency between two networks trained on the same data; this is a partial circularity rather than a fully independent verification of the claimed physical capability.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central claim rests on simulated training data, on an implicit accuracy metric, and on the assumption that target Jones matrices generated for applications are realizable by the unit-cell library. No fabricated device or independent measurement is provided, so the axioms are domain assumptions rather than verified facts.

free parameters (1)
  • Neural network weights of the inverse design and forward prediction networks = Not released; trained on 60 million simulated samples
    The model's predictions are entirely determined by these fitted weights; no pretrained checkpoint is published, so the central result is not independently reproducible from the paper alone.
assumptions (3)
  • domain assumption FDTD-computed responses of single Si nanopillars, combined by analytic rotation and pairing, faithfully represent the optical response of fabricated metasurface unit cells.
    The entire dataset and all evaluations inherit this assumption; no fabricated structure is measured. See Methods Section 4.3 and Figure 1.
  • domain assumption Target Jones matrices constructed for applications are physically realizable within the training distribution.
    The paper itself reports L1 losses around 0.13 to 0.15 for ideal designed Jones matrices, so this assumption is only partially satisfied, and validation uses no fabricated sample. See Section 2.3.
  • ad hoc to paper L1 loss on the six Jones-matrix components is a valid proxy for metasurface performance and accuracy.
    The reported accuracy of 99% and 85% is never explicitly defined; if it is 1 minus L1, it ignores phase wrapping, physical realizability, and application-specific tolerances. See Section 2.3 and the Conclusion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MetasurfaceViT: A generic AI model for metasurface inverse design." pith.science (2026). https://pith.science/paper/ULNGSXNS

@misc{pith2026250414895,
  author       = {Pith},
  title        = {Pith review of: MetasurfaceViT: A generic AI model for metasurface inverse design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ULNGSXNS}},
  note         = {Machine review of arXiv:2504.14895}
}
read the original abstract

Metasurfaces, sub-wavelength artificial structures, can control light's amplitude, phase, and polar ization, enabling applications in efficient imaging, holograms, and sensing. Recent years, AI has witnessed remarkable progress and spurred scientific discovery. In metasurface design, optical inverse design has recently emerged as a revolutionary approach. It uses deep learning to create a nonlinear mapping between optical structures and functions, bypassing time-consuming traditional design and attaining higher accuracy. Yet, current deep-learning models for optical design face limitations. They often work only for fixed wavelengths and polarizations, and lack universality as input-output vector size changes may require retraining. There's also a lack of compatibility across different application scenarios. This paper introduces MetasurfaceViT, a revolutionary generic AI model. It leverages a large amount of data using Jones matrices and physics-informed data augmentation. By pre-training through masking wavelengths and polarization channels, it can reconstruct full-wavelength Jones matrices, which will be utilized by fine-tuning model to enable inverse design. Finally, a tandem workflow appended by a forward prediction network is introduced to evaluate performance. The versatility of MetasurfaceViT with high prediction accuracy will open a new paradigm for optical inverse design.

Figures

Figures reproduced from arXiv: 2504.14895 by the authors.

Figure 1
Figure 1. (a) Schematic illustration of the building block of metasurfaces, which consists of Si nanopillars on a [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. (a) Workflow of the pretraining, design, and reconstruction phases. Pretraining: Jones Matrices from the [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. (a) Workflows of the fine-tuning, prediction, and evaluation phases. When the design type is a metalens, [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (a) Schematic illustration of the process for designing and evaluating multiplexing metasurfaces. (b) [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: (a) Schematic representation of the calculation methodology for phase distribution. (b) Layouts of the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. An Agentic Framework for Autonomous Metamaterial Modeling and Inverse Design

    cs.AI 2025-06 conditional novelty 5.0 of 10

    An LLM agent team autonomously trains a surrogate model and performs inverse design for metamaterials, matching human forward accuracy but not inverse accuracy.

Reference graph

Works this paper leans on

34 extracted references · 24 canonical work pages · cited by 1 Pith paper

  1. [1]

    Deep learning enabled inverse design in nanophotonics

    Sunae So, Trevon Badloe, Jaebum Noh, Jorge Bravo-Abad, and Junsuk Rho. Deep learning enabled inverse design in nanophotonics. Nanophotonics, 9(5):1041–1057, 2020

    2020

  2. [2]

    Deep learning in nano-photonics: inverse design and beyond

    Peter R Wiecha, Arnaud Arbouet, Christian Girard, and Otto L Muskens. Deep learning in nano-photonics: inverse design and beyond. Photonics Research, 9(5):B182–B200, 2021

    work page 2021

  3. [3]

    Generative model for the inverse design of metasurfaces

    Zhaocheng Liu, Dayu Zhu, Sean P Rodrigues, Kyu-Tae Lee, and Wenshan Cai. Generative model for the inverse design of metasurfaces. Nano letters, 18(10):6570–6576, 2018

    2018

  4. [4]

    Arbitrary multifunctional vortex beam designed by deep neural network

    Pu Peng, Zhengchang Liu, Changhong Dai, Xiao He, Dongyi Wang, Zhibo Dang, Yuxiang Chen, Yuchen Dai, Han Zhang, Shulin Sun, et al. Arbitrary multifunctional vortex beam designed by deep neural network. Advanced Optical Materials, 12(6):2300158, 2024

    work page 2024

  5. [5]

    Physics-informed inverse design of programmable metasurfaces

    Yucheng Xu, Jia-Qi Yang, Kebin Fan, Sheng Wang, Jingbo Wu, Caihong Zhang, De-Chuan Zhan, Willie J Padilla, Biaobing Jin, Jian Chen, et al. Physics-informed inverse design of programmable metasurfaces. Advanced Science, 11(41):2406878, 2024

    work page 2024

  6. [6]

    Deep-learning assisted polarization holograms

    Jiawei Xi, Jian Shen, Man To Chow, Tan Li, Jack Ng, and Jensen Li. Deep-learning assisted polarization holograms. Advanced Optical Materials, 12(6):2202663, 2024

    work page 2024

  7. [7]

    Deep-learning empowered customized chiral metasurface for calibration-free biosensing

    Nan Zhang, Feng Gao, Ride Wang, Zhonglei Shen, Donghai Han, Yuqing Cui, Liuyang Zhang, Chao Chang, Cheng-wei Qiu, and Xuefeng Chen. Deep-learning empowered customized chiral metasurface for calibration-free biosensing. Advanced Materials, 37(1):2411490, 2025

    work page 2025

  8. [8]

    Deep learning for the design of photonic structures

    Wei Ma, Zhaocheng Liu, Zhaxylyk A Kudyshev, Alexandra Boltasseva, Wenshan Cai, and Yongmin Liu. Deep learning for the design of photonic structures. Nature photonics, 15(2):77–90, 2021

    work page 2021

Show all 34 references
  1. [9]

    Pushing the limits of functionality-multiplexing capability in metasurface design based on statistical machine learning

    Wei Ma, Yihao Xu, Bo Xiong, Lin Deng, Ru-Wen Peng, Mu Wang, and Yongmin Liu. Pushing the limits of functionality-multiplexing capability in metasurface design based on statistical machine learning. Advanced Materials, 34(16):2110022, 2022

    2022

  2. [10]

    Interfacing nanophotonics with deep neural networks: Ai for photonic design and photonic implementation of ai

    Taehyuk Park, Sujoy Mondal, and Wenshan Cai. Interfacing nanophotonics with deep neural networks: Ai for photonic design and photonic implementation of ai. Laser & Photonics Reviews, page 2401520

  3. [11]

    Design of multifunctional color routers with kerker switching using generative adversarial networks

    Jiahao Yan, Dayu Zhu, Yanjun Bao, Qin Chen, Baojun Li, and Wenshan Cai. Design of multifunctional color routers with kerker switching using generative adversarial networks. Laser & photonics reviews, 18(4):2300592, 2024

    2024

  4. [12]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems , 30, 2017

    2017

  5. [13]

    A survey on large language model based autonomous agents

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, et al. A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6):186345, 2024

    2024

  6. [14]

    A survey on vision transformer

    Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, et al. A survey on vision transformer. IEEE transactions on pattern analysis and machine intelligence, 45(1):87–110, 2022

    2022

  7. [15]

    Highly accurate protein structure prediction with alphafold

    John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ronneberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin Žídek, Anna Potapenko, et al. Highly accurate protein structure prediction with alphafold. nature, 596(7873):583–589, 2021

    2021

  8. [16]

    Transformers enable accurate prediction of acute and chronic chemical toxicity in aquatic organisms

    Mikael Gustavsson, Styrbjörn Käll, Patrik Svedberg, Juan S Inda-Diaz, Sverker Molander, Jessica Coria, Thomas Backhaus, and Erik Kristiansson. Transformers enable accurate prediction of acute and chronic chemical toxicity in aquatic organisms. Science Advances, 10(10):eadk6669...

    2024

  9. [17]

    A generative model for inorganic materials design

    Claudio Zeni, Robert Pinsler, Daniel Zügner, Andrew Fowler, Matthew Horton, Xiang Fu, Zilong Wang, Aliaksan- dra Shysheya, Jonathan Crabbé, Shoko Ueda, et al. A generative model for inorganic materials design. Nature, pages 1–3, 2025

    2025

  10. [18]

    Scientific discovery in the age of artificial intelligence

    Hanchen Wang, Tianfan Fu, Yuanqi Du, Wenhao Gao, Kexin Huang, Ziming Liu, Payal Chandak, Shengchao Liu, Peter Van Katwyk, Andreea Deac, et al. Scientific discovery in the age of artificial intelligence. Nature, 620(7972):47–60, 2023

    2023

  11. [19]

    Data-centric artificial intelligence: A survey

    Daochen Zha, Zaid Pervaiz Bhat, Kwei-Herng Lai, Fan Yang, Zhimeng Jiang, Shaochen Zhong, and Xia Hu. Data-centric artificial intelligence: A survey. ACM Computing Surveys, 57(5):1–42, 2025

    2025

  12. [20]

    Scaling laws for neural language models

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models. arXiv preprint arXiv:2001.08361, 2020

    2001 arXiv

  13. [21]

    Scaling vision transformers

    Xiaohua Zhai, Alexander Kolesnikov, Neil Houlsby, and Lucas Beyer. Scaling vision transformers. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 12104–12113, 2022

    2022

  14. [22]

    Transfer learning for metamaterial design and simulation

    Rixi Peng, Simiao Ren, Jordan Malof, and Willie J Padilla. Transfer learning for metamaterial design and simulation. Nanophotonics, 13(13):2323–2334, 2024

    2024

  15. [23]

    Meta-attention network based spectral reconstruction with snapshot near-infrared metasurface

    Haoyang He, Yuzhe Zhang, Yujie Shao, Yan Zhang, Guangzhou Geng, Junjie Li, Xin Li, Yongtian Wang, Liheng Bian, Jun Zhang, et al. Meta-attention network based spectral reconstruction with snapshot near-infrared metasurface. Advanced Materials, 36(23):2313357, 2024

    2024

  16. [24]

    Deep neural network enabled active metasurface embedded design

    Sensong An, Bowen Zheng, Matthew Julian, Calum Williams, Hong Tang, Tian Gu, Hualiang Zhang, Hyun Jung Kim, and Juejun Hu. Deep neural network enabled active metasurface embedded design. Nanophotonics, 11(17):4149–4158, 2022

    2022

  17. [25]

    Frequency transfer and inverse design for metasurface under multi-physics coupling by euler latent dynamic and data-analytical regularizations

    Enze Zhu, Zheng Zong, Erji Li, Yang Lu, Jingwei Zhang, Hao Xie, Ying Li, Wen-Yan Yin, and Zhun Wei. Frequency transfer and inverse design for metasurface under multi-physics coupling by euler latent dynamic and data-analytical regularizations. Nature Communications, 16(1):2251, 2025

    2025

  18. [26]

    Spectral transfer-learning-based metasurface design assisted by complex-valued deep neural network

    Yi Xu, Fu Li, Jianqiang Gu, Zhiwei Bi, Bing Cao, Quanlong Yang, Jiaguang Han, Qinghua Hu, and Weili Zhang. Spectral transfer-learning-based metasurface design assisted by complex-valued deep neural network. Advanced Photonics Nexus, 3(2):026002–026002, 2024

    2024

  19. [27]

    On-demand quick metasurface design with neighborhood attention transformer

    Zhi Sun, Tianyue Li, Shiqi Kuang, Xue Yun, Minru He, Boyan Fu, Yunlai Fu, Tianyu Zhao, Shaowei Wang, Yansheng Liang, et al. On-demand quick metasurface design with neighborhood attention transformer. arXiv preprint arXiv:2412.08405, 2024

    2024 arXiv

  20. [28]

    Deep learning design for multiwavelength infrared image sensors based on dielectric freeform metasurface

    Bo Xiong, Yihao Xu, Wenwen Li, Wei Ma, Tao Chu, and Yongmin Liu. Deep learning design for multiwavelength infrared image sensors based on dielectric freeform metasurface. Advanced optical materials, 12(10):2302200, 2024

    2024

  21. [29]

    Harnessing the missing spectral correlation for metasurface inverse design

    Jie Zhang, Chao Qian, Guangfeng You, Tao Wang, Yasir Saifullah, Reza Abdi-Ghaleh, and Hongsheng Chen. Harnessing the missing spectral correlation for metasurface inverse design. Advanced Science, 11(33):2308807, 2024

    2024

  22. [30]

    Nanophotonic device design based on large language models: multilayer and metasurface examples

    Myungjoon Kim, Hyeonjin Park, and Jonghwa Shin. Nanophotonic device design based on large language models: multilayer and metasurface examples. Nanophotonics, (0), 2025

    2025

  23. [31]

    Optogpt: a foundation model for inverse design in optical multilayer thin film structures

    Taigao Ma, Haozhu Wang, and L Jay Guo. Optogpt: a foundation model for inverse design in optical multilayer thin film structures. Opto-Electronic Advances, 7(7), 2024

    2024

  24. [32]

    Observation of full-parameter jones matrix in bilayer metasurface

    Yanjun Bao, Fan Nan, Jiahao Yan, Xianguang Yang, Cheng-Wei Qiu, and Baojun Li. Observation of full-parameter jones matrix in bilayer metasurface. Nature Communications, 13(1):7550, 2022

    2022

  25. [33]

    Single-shot simultaneous intensity, phase and polarization imaging with metasurface

    Yanjun Bao and Baojun Li. Single-shot simultaneous intensity, phase and polarization imaging with metasurface. National Science Review, 12(3):nwae418, 2025

    2025

  26. [34]

    Toward the capacity limit of 2d planar jones matrix with a single-layer metasurface

    Yanjun Bao, Long Wen, Qin Chen, Cheng-Wei Qiu, and Baojun Li. Toward the capacity limit of 2d planar jones matrix with a single-layer metasurface. Science advances, 7(25):eabh0365, 2021. 11

    2021

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.