Pith. sign in

REVIEW 3 major objections 5 minor 71 references

Inverse Design of Metamaterials with Manufacturing-Guiding Spectrum-to-Structure Conditional Diffusion Model

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read DiffMeta turns a target spectrum into fabricable metamaterial designs

desk verdict A well-executed conditional-diffusion inverse-design paper with a real fabricated demo, but the RCWA oracle is both load-bearing and under-validated, and the edge over the VAE baseline is thinner than the text suggests. read the letter →

arxiv 2506.07083 v1 pith:3QV7UTQH submitted 2025-06-08 physics.optics cs.LG

Jiawen Li , Jiang Guo , Yuanzhe Li , Zetian Mao , Jiaxing Shen , Tashi Xu , Diptesh Das , Jinming He
show 4 more authors
Run Hu Yaerim Lee Koji Tsuda Junichiro Shiomi
This is my paper
classification physics.opticscs.LG
keywords metamaterialsinversedesignconditionaldiffusionmodelthermalcamouflagespectrum-to-structurefree-formmetasurfacesizeparametersrigorouscoupled-waveanalysis
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper introduces DiffMeta, a conditional diffusion model that tackles the one-to-many inverse design problem for metamaterials: given a target infrared emission spectrum, it generates both the free-form top-layer pattern and the four size parameters of a metal-insulator-metal unit cell. The claim is that this spectrum-to-structure mapping is accurate enough that generated designs re-simulate to the target spectrum, and diverse enough to inform practical fabrication choices. Compared with conditional VAE and GAN variants on 6000 rigorous coupled-wave analysis (RCWA) test spectra, DiffMeta reports the lowest pattern error, size-parameter error, and spectrum error. The authors validate the full pipeline by fabricating a free-form thermal camouflage emitter whose measured emissivity is roughly 0.1 in the 3–5 and 8–13 µm atmospheric windows while carrying about 80% of blackbody heat flux in the 5–8 µm band.

What carries the argument

The central mechanism is conditional denoising: a U-Net, a convolutional encoder-decoder network, removes Gaussian noise from a pattern image while the target spectrum is injected through cross-attention layers, and a CNN head reads an intermediate U-Net layer to output the four geometric parameters. The spectrum-structure pairs, generated by rigorous coupled-wave analysis, supply both the training signal and the evaluation oracle, letting one target spectrum branch into many valid structures whose optical responses are checked by re-simulation.

What would settle it

Fabricate a batch of DiffMeta-designed structures for a different target spectrum than the thermal camouflage case and measure their FTIR emissivity; if the average deviation from the target exceeds the RCWA-simulated error by a large margin, or if the diversity-guided tolerance predictions are not reflected in measured performance, the method's central claim fails.

Watch

Extended reading notes

Core claim

DiffMeta couples a U-Net denoiser to a spectral encoder through cross-attention, so the noise removed at each reverse-diffusion step is conditioned on the desired emissivity spectrum rather than on a single global guidance scale. A CNN attached to the U-Net's middle layer predicts the four size parameters (pitch, top-layer height, dielectric spacer height, and reflector thickness) jointly with pattern generation. The paper's claim is that this conditional denoising solves the one-to-many problem: on 6000 RCWA-simulated test spectra it reports lower pattern error (0.0047), size-parameter error (0.0136), and spectrum error (0.0619) than conditional VAE and GAN baselines, while generating visibly diverse patterns. Sampling 1000 structures for one thermal-camouflage target spectrum shows the pitch and pattern height are narrowly distributed whereas spacer and reflector thickness spread widely, which the authors interpret as a manufacturing tolerance guide. A fabricated version of the design shows emissivity near 0.091 in the 3–5 µm band and 0.103 in the 8–13 µm band, with strong 5–8 µm emission.

Load-bearing premise

The entire inverse design pipeline relies on rigorous coupled-wave analysis (RCWA) simulated spectra as ground truth labels and as the evaluation oracle, and if that simulator does not faithfully predict the emissivity of experimentally fabricated structures the claimed spectral accuracy and manufacturing guidance break down.

Editorial extensions

If this is right

  • The same target spectrum yields multiple distinct, optically equivalent structures, so designers can choose among solutions based on fabrication cost rather than rerunning optimization.
  • The parameter distributions from 1000 samples tell the fabricator which dimensions are critical: pitch and top-pattern height need tight control, while spacer and reflector thickness tolerate wider variation.
  • The framework outperforms conditional VAE and GAN baselines on spectrum error and pattern recovery, suggesting diffusion models are a better fit for high-degree-of-freedom metamaterial inverse design.
  • The free-form top layer excites surface plasmons and magnetic polaritons that broaden 5–8 µm emission while suppressing 3–5 and 8–13 µm emission, making the generated device suitable for thermal camouflage at 180 °C.

Reading between the lines

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

  • Extending beyond the paper, the same conditional diffusion setup should transfer to other wavelength bands and material stacks, because the conditioning is just a spectrum vector and the output space is free-form patterns.
  • The diversity distributions could be used as a testable yield predictor: fabricating multiple structures from the tight and broad parameter regions should reproduce the predicted performance spread.
  • A future closed-loop version that fine-tunes on experimentally measured FTIR spectra could shrink the residual discrepancy the paper attributes to fabrication rounding and material impurities.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript presents DiffMeta, a conditional diffusion model for the inverse design of free-form metal-insulator-metal metamaterial unit cells. A U-Net with cross-attention spectrum conditioning generates a 2D top-layer pattern from a target emissivity spectrum, and a CNN predicts four geometric size parameters from intermediate U-Net features. The authors compare DiffMeta with conditional VAE and GAN baselines on a test set of 6000 RCWA-simulated spectra, report pattern, size-parameter, and spectrum errors, analyze 1000 generated designs to infer parameter tolerances, and fabricate a thermal-camouflage emitter that shows low emissivity in the 3-5 and 8-13 um windows and enhanced emission near 5-8 um. They also analyze electromagnetic field profiles to attribute the emission peaks to surface plasmon and magnetic polariton resonances.

Significance. The paper addresses an important and active problem: machine-learning inverse design of metamaterials with one-to-many solution sets and manufacturing constraints. Its strengths include public release of code and data, side-by-side comparison with VAE/GAN baselines, a concrete fabrication demonstration with thermal IR imaging, and physical mechanism analysis. The manufacturing-tolerance analysis is a useful idea. However, the headline claims of superior spectral accuracy and of experimentally validated manufacturing guidance are only partially supported: the reported accuracy advantage over VAEMeta is small and unaccompanied by uncertainty quantification, and the simulator-experiment correspondence is validated with a single sample. If the authors add statistical rigor and a more direct validation of simulator fidelity, the framework could be a solid contribution.

major comments (3)
  1. [Performance of DiffMeta, Table 1] The central claim of superior spectral prediction accuracy rests on a single point estimate: DiffMeta spectrum error 0.0619 versus VAEMeta 0.0650, with no error bars, repeated runs, or significance test. Since the difference is only about 5% relative and could easily change with random seed or training run, the current evidence does not establish that DiffMeta is more accurate than VAEMeta in spectrum prediction. Please report mean and standard deviation over multiple training runs and apply a paired statistical test (e.g., Wilcoxon signed-rank test on per-spectrum errors) for all three metrics in Table 1.
  2. [Methods/Spectrum Simulation and Physical Mechanism Analysis, Figure 6C] The RCWA simulation is explicitly performed under p-polarized normal incidence, while the experimental FTIR emissivity measurement is not described in terms of polarization or angular collection. If the measured emissivity is unpolarized or angle-averaged, the p-polarized normal-incidence proxy may not correspond to the experimental quantity, and this mismatch would propagate through training labels and all evaluation metrics. The single fabricated sample in Figure 6C and Figure S4, with deviations attributed to fabrication imperfections and material impurities, is insufficient to validate the simulator across the design space or to support the claimed quantitative figures (emissivity near 0.1 in the 3-5 and 8-13 um windows, and about 80% blackbody heat flux in 5-8 um). Please specify the exact measurement configuration, quantify simulation-measurement agreement on more than one sample, and assess how sensitive the reported metrics are to the p-polarized normal-incidence assumption.
  3. [Diversity in Generation and Figure 5] The evaluation in Table 1 rewards agreement with a single ground-truth pattern, which is in tension with the paper's one-to-many premise: a structurally different but spectrally equivalent solution is penalized by the pattern error. The diversity and manufacturing-tolerance analysis in Figure 5 is also presented qualitatively, with statements such as 'over 90% of patterns are concentrated' but no quantitative diversity metric, no confidence intervals on the parameter distributions, and no experimental verification that the broad tolerances inferred for phi3 and phi4 are robust to fabrication variation. Please add quantitative diversity and coverage metrics for the generated samples and, if possible, fabricate multiple structures to test whether the advertised tolerance analysis holds in practice.
minor comments (5)
  1. [Experimental Realization, Figure 6F] The vacuum condition is given as '1 x 10^4 hPa', which is atmospheric pressure; this appears to be a typo, likely intended as 10^-4 hPa or a comparable high-vacuum value. Please correct the unit and value.
  2. [Experimental Realization, Figure 6C] The text states that three resonant peaks at 5.406, 6.488, and 7.241 um are indicated by the simulated spectral radiance flux, but the figure appears to show both simulation and experimental curves; please clarify which curve is used to define these peak wavelengths.
  3. [Diversity in Generation, Figure 4] The claim that GANMeta exhibits significant mode collapse and that VAEMeta produces overly similar patterns is supported only by visual inspection of five samples; a quantitative diversity metric such as pairwise structural similarity or coverage of the ground-truth distribution would make the comparison more convincing.
  4. [Methods and Supplemental Information] The main text does not state the dataset size or the train/validation/test split, which are important for assessing the fairness of the comparison with VAEMeta and GANMeta; please include these numbers in the main text or clearly reference the exact Supplemental section.
  5. [Performance of DiffMeta] The phrase 'superior spectral prediction accuracy' in the abstract and introduction should be moderated unless the statistical analysis recommended above supports it; the current difference from VAEMeta is small and not shown to be significant.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: DiffMeta's inverse-design accuracy is validated by held-out RCWA forward simulation and one fabricated sample; the only author-overlapping citation is background and not load-bearing.

full rationale

The paper's derivation chain is not circular in the sense the review targets. Training spectra are generated by RCWA forward simulation, and the model is then evaluated by taking newly generated pattern-parameter pairs and simulating their spectra with the same RCWA solver, e.g., Table 1's 'spectrum error' is 'the difference between the input spectrum and the spectrum simulated from the generated pair.' This is a closed-loop check on held-out outputs, not a refit of the training labels or a prediction forced by construction. The one fabricated thermal-camouflage emitter, with FTIR-measured emissivity suppressing the 3-5 and 8-13 um windows and enhancing 5-8 um, provides an external anchor beyond the simulator. The only self-citation with present-author overlap is ref. 34 (Xi et al. 2023, including Shiomi and Hu), cited in the introductory list of thin-film camouflage options; it is background and not load-bearing for the central diffusion-model claim. The main residual risk, as the paper itself notes, is that the simulation uses p-polarized normal-incidence illumination and that experimental deviations are attributed to 'fabrication imperfections and material impurities' without quantifying simulator error; this is a correctness/validation concern, not a circular reduction. No equation or fitted parameter is equivalent to the claimed prediction by definition.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No new physical entities are introduced. The framework rests on RCWA as the forward model, the expressiveness of the polygon-based design space, the fabrication process approximating design geometry, and standard results in generative modeling. The main hand-set parameters are the geometric bounds defining the design space; the network weights are learned from data rather than introduced as free physical parameters.

free parameters (4)
  • Pitch size range = 3-8 µm
    Hand-set fabrication bound; defines the design space for generated patterns and affects achievable resonance positions.
  • Top pattern height range = 0-0.8 µm
    Hand-set; determines aspect ratios and resonance characteristics of the top metal pattern.
  • Spacer and reflector thickness ranges = 0-1 µm and 0-0.2 µm
    Hand-set fabrication bounds; the tolerance analysis in Figure 5B is over these ranges.
  • Vertex count range = 5-20
    Hand-set complexity bound for free-form patterns; constrains the expressiveness of the shape space.
assumptions (5)
  • domain assumption RCWA simulation provides accurate emissivity for periodic MIM metamaterials in the 3-15 µm range.
    All training labels and evaluation spectra are computed by RCWA; the experimental FTIR check on one sample shows deviations, so this assumption is load-bearing.
  • standard math Kirchhoff's law: emissivity equals absorptivity for reciprocal materials.
    Used in Methods to derive emissivity from absorption computed as (1-R-T).
  • domain assumption The free-form pattern class (smoothed polygons with 5-20 vertices, limited sizes) spans structures that can realize the target spectra, including the thermal camouflage step function.
    If the target spectrum is not reachable within this parameterized class, the inverse design cannot succeed; the paper demonstrates one example but does not prove coverage.
  • domain assumption The fabricated DWL-written structure, with rounding, approximates the designed pattern closely enough that optical performance is preserved.
    The paper compares design versus rounded shape in Figure S4 and claims negligible effect; this is checked indirectly through a single experimental sample.
  • domain assumption The conditional diffusion model with cross-attention can learn the one-to-many spectrum-to-structure mapping from the available dataset.
    The method's success depends on the capacity of the neural network and the sufficiency of the dataset; no theorem guarantees this mapping is learnable to the required accuracy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inverse Design of Metamaterials with Manufacturing-Guiding Spectrum-to-Structure Conditional Diffusion Model." pith.science (2026). https://pith.science/paper/3QV7UTQH

@misc{pith2026250607083,
  author       = {Pith},
  title        = {Pith review of: Inverse Design of Metamaterials with Manufacturing-Guiding Spectrum-to-Structure Conditional Diffusion Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3QV7UTQH}},
  note         = {Machine review of arXiv:2506.07083}
}
read the original abstract

Metamaterials are artificially engineered structures that manipulate electromagnetic waves, having optical properties absent in natural materials. Recently, machine learning for the inverse design of metamaterials has drawn attention. However, the highly nonlinear relationship between the metamaterial structures and optical behaviour, coupled with fabrication difficulties, poses challenges for using machine learning to design and manufacture complex metamaterials. Herein, we propose a general framework that implements customised spectrum-to-shape and size parameters to address one-to-many metamaterial inverse design problems using conditional diffusion models. Our method exhibits superior spectral prediction accuracy, generates a diverse range of patterns compared to other typical generative models, and offers valuable prior knowledge for manufacturing through the subsequent analysis of the diverse generated results, thereby facilitating the experimental fabrication of metamaterial designs. We demonstrate the efficacy of the proposed method by successfully designing and fabricating a free-form metamaterial with a tailored selective emission spectrum for thermal camouflage applications.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 50 canonical work pages

  1. [1]

    Yu, N., and Capasso, F. (2014). Flat optics with designer metasurfaces. Nature Mater 13, 139–150. https://doi.org/10.1038/nmat3839

    doi:10.1038/nmat3839 2014

  2. [2]

    Cui, T.-J., Liu, S., and Li, L.-L. (2016). Information entropy of coding metasurface. Light Sci Appl 5, e16172–e16172. https://doi.org/10.1038/lsa.2016.172

    work page doi:10.1038/lsa.2016.172 2016

  3. [3]

    Chen, S., Liu, W., Li, Z., Cheng, H., and Tian, J. (2020). Metasurface-Empowered Optical Multiplexing and Multifunction. Advanced Materials 32, 1805912. https://doi.org/10.1002/adma.201805912

    work page doi:10.1002/adma.201805912 2020

  4. [4]

    Chu, Q., Zhong, F., Shang, X., Zhang, Y., Zhu, S., and Liu, H. (2024). Controlling thermal emission with metasurfaces and its applications. Nanophotonics 13, 1279–1301. https://doi.org/10.1515/nanoph-2023- 0754

    work page doi:10.1515/nanoph-2023- 2024

  5. [5]

    Pendry, J.B., Schurig, D., and Smith, D.R. (2006). Controlling Electromagnetic Fields. Science 312, 1780–

    work page 2006

  6. [6]

    Li, Z., Pestourie, R., Lin, Z., Johnson, S.G., and Capasso, F. (2022). Empowering Metasurfaces with Inverse Design: Principles and Applications. ACS Photonics 9, 2178–2192. https://doi.org/10.1021/acsphotonics.1c01850

    doi:10.1021/acsphotonics.1c01850 2022

  7. [7]

    Khaireh-Walieh, A., Langevin, D., Bennet, P., Teytaud, O., Moreau, A., and Wiecha, P.R. (2023). A newcomer’s guide to deep learning for inverse design in nano-photonics. Nanophotonics 12, 4387–4414. https://doi.org/10.1515/nanoph-2023-0527

    doi:10.1515/nanoph-2023-0527 2023

  8. [8]

    Molesky, S., Lin, Z., Piggott, A.Y., Jin, W., Vucković, J., and Rodriguez, A.W. (2018). Inverse design in nanophotonics. Nature Photon 12, 659–670. https://doi.org/10.1038/s41566-018-0246-9

    doi:10.1038/s41566-018-0246-9 2018

Show all 71 references
  1. [9]

    Ma, W., Cheng, F., Xu, Y., Wen, Q., and Liu, Y. (2019). Probabilistic Representation and Inverse Design of Metamaterials Based on a Deep Generative Model with Semi-Supervised Learning Strategy. Advanced Materials 31, 1901111. https://doi.org/10.1002/adma.201901111

    2019 doi

  2. [10]

    Wang, N., Yan, W., Qu, Y., Ma, S., Li, S.Z., and Qiu, M. (2021). Intelligent designs in nanophotonics: from optimization towards inverse creation. PhotoniX 2, 22. https://doi.org/10.1186/s43074-021-00044-y

    2021 doi

  3. [11]

    Ma, W., Liu, Z., Kudyshev, Z.A., Boltasseva, A., Cai, W., and Liu, Y. (2021). Deep learning for the design of photonic structures. Nat. Photonics 15, 77–90. https://doi.org/10.1038/s41566-020-0685-y

    2021 doi

  4. [12]

    Jiang, J., Chen, M., and Fan, J.A. (2021). Deep neural networks for the evaluation and design of photonic devices. Nat Rev Mater 6, 679–700. https://doi.org/10.1038/s41578-020-00260-1

    2021 doi

  5. [13]

    Liu, D., Tan, Y., Khoram, E., and Yu, Z. (2018). Training Deep Neural Networks for the Inverse Design of Nanophotonic Structures. ACS Photonics 5, 1365–1369. https://doi.org/10.1021/acsphotonics.7b01377

    2018 doi

  6. [14]

    Gao, L., Li, X., Liu, D., Wang, L., and Yu, Z. (2019). A Bidirectional Deep Neural Network for Accurate Silicon Color Design. Advanced Materials 31, 1905467. https://doi.org/10.1002/adma.201905467

    2019 doi

  7. [15]

    Inverse Design of Photonic Surfaces via High throughput Femtosecond Laser Processing and Tandem Neural Networks

    Park, M., Grbčić, L., Motameni, P., Song, S., Singh, A., Malagrino, D., Elzouka, M., Vahabi, P.H., Todeschini, A., de Jong, W.A., et al. Inverse Design of Photonic Surfaces via High throughput Femtosecond Laser Processing and Tandem Neural Networks. Advanced Science n/a, 24019...

    doi

  8. [16]

    Kiarashinejad, Y., Abdollahramezani, S., and Adibi, A. (2020). Deep learning approach based on dimensionality reduction for designing electromagnetic nanostructures. npj Comput Mater 6, 1–12. https://doi.org/10.1038/s41524-020-0276-y

    2020 doi

  9. [17]

    Kingma, D.P., and Welling, M. (2013). Auto-Encoding Variational Bayes. Preprint at arXiv, https://doi.org/10.48550/ARXIV.1312.6114 https://doi.org/10.48550/ARXIV.1312.6114

    2013 doi arXiv

  10. [18]

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. (2020). Generative adversarial networks. Commun. ACM 63, 139–144. https://doi.org/10.1145/3422622

    2020 doi

  11. [19]

    Liu, Z., Raju, L., Zhu, D., and Cai, W. (2020). A Hybrid Strategy for the Discovery and Design of Photonic Structures. IEEE Journal on Emerging and Selected Topics in Circuits and Systems 10, 126–135. https://doi.org/10.1109/JETCAS.2020.2970080

    2020

  12. [20]

    Liu, Z., Zhu, D., Rodrigues, S.P., Lee, K.-T., and Cai, W. (2018). Generative Model for the Inverse Design of Metasurfaces. Nano Lett. 18, 6570–6576. https://doi.org/10.1021/acs.nanolett.8b03171

    2018 doi

  13. [21]

    Jiang, J., Sell, D., Hoyer, S., Hickey, J., Yang, J., and Fan, J.A. (2019). Free-Form Diffractive Metagrating Design Based on Generative Adversarial Networks. ACS Nano 13, 8872–8878. https://doi.org/10.1021/acsnano.9b02371

    2019 doi

  14. [22]

    Dosovitskiy, A., and Brox, T. (2016). Generating Images with Perceptual Similarity Metrics based on Deep Networks. In Advances in Neural Information Processing Systems (Curran Associates, Inc.)

    2016

  15. [23]

    Shekhovtsov, A., Schlesinger, D., and Flach, B. (2021). VAE Approximation Error: ELBO and Exponential Families. Preprint at arXiv, https://doi.org/10.48550/ARXIV.2102.09310 https://doi.org/10.48550/ARXIV.2102.09310

    2021 doi arXiv

  16. [24]

    Thanh-Tung, H., and Tran, T. (2020). Catastrophic forgetting and mode collapse in GANs. In 2020 International Joint Conference on Neural Networks (IJCNN), pp. 1–10. https://doi.org/10.1109/IJCNN48605.2020.9207181

    2020

  17. [25]

    Sohl-Dickstein, J., Weiss, E., Maheswaranathan, N., and Ganguli, S. (2015). Deep Unsupervised Learning using Nonequilibrium Thermodynamics. In Proceedings of the 32nd International Conference on Machine Learning (PMLR), pp. 2256–2265

    2015

  18. [26]

    Ho, J., Jain, A., and Abbeel, P. (2020). Denoising Diffusion Probabilistic Models. In Advances in Neural Information Processing Systems (Curran Associates, Inc.), pp. 6840–6851

    2020

  19. [27]

    Saharia, C., Chan, W., Chang, H., Lee, C., Ho, J., Salimans, T., Fleet, D., and Norouzi, M. (2022). Palette: Image-to-Image Diffusion Models. In ACM SIGGRAPH 2022 Conference Proceedings SIGGRAPH ’22. (Association for Computing Machinery), pp. 1–10. https://doi.org/10.1145/3528...

    2022

  20. [28]

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. (2022). High-Resolution Image Synthesis With Latent Diffusion Models. In, pp. 10684–10695

    2022

  21. [29]

    Saharia, C., Chan, W., Saxena, S., Li, L., Whang, J., Denton, E.L., Ghasemipour, K., Gontijo Lopes, R., Karagol Ayan, B., Salimans, T., et al. (2022). Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding. Advances in Neural Information Processing Syst...

    2022

  22. [30]

    Zhang, Z., Yang, C., Qin, Y., Feng, H., Feng, J., and Li, H. (2023). Diffusion probabilistic model based accurate and high-degree-of-freedom metasurface inverse design. Nanophotonics 12, 3871–3881. https://doi.org/10.1515/nanoph-2023-0292

    2023 doi

  23. [31]

    Shen, D., Song, G., Xue, Z., Wang, F.-Y., and Liu, Y. (2024). Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance. Preprint at arXiv, https://doi.org/10.48550/ARXIV.2404.05384 https://doi.org/10.48550/ARXIV.2404.05384

    2024 doi arXiv

  24. [32]

    Kim, J., Neseli, B., Yoon, J., Kim, J., Hong, S., Park, H., and Kurt, H. (2024). Semi‐Supervised Learning Leveraging Denoising Diffusion Probabilistic Models for the Characterization of Nanophotonic Devices. Laser & Photonics Reviews, 2300998. https://doi.org/10.1002/lpor....

    2024 doi

  25. [33]

    Jiang, X., Yuan, H., He, X., Du, T., Ma, H., Li, X., Luo, M., Zhang, Z., Chen, H., Yu, Y., et al. (2023). Implementing of infrared camouflage with thermal management based on inverse design and hierarchical metamaterial. Nanophotonics 12, 1891–1902. https://doi.org/10.1515/nan...

    2023 doi

  26. [34]

    Xi, W., Lee, Y.-J., Yu, S., Chen, Z., Shiomi, J., Kim, S.-K., and Hu, R. (2023). Ultrahigh-efficient material informatics inverse design of thermal metamaterials for visible-infrared-compatible camouflage. Nat Commun 14, 4694. https://doi.org/10.1038/s41467-023-40350-6

    2023 doi

  27. [35]

    Zhu, H., Li, Q., Zheng, C., Hong, Y., Xu, Z., Wang, H., Shen, W., Kaur, S., Ghosh, P., and Qiu, M. (2020). High-temperature infrared camouflage with efficient thermal management. Light Sci Appl 9, 60. https://doi.org/10.1038/s41377-020-0300-5

    2020 doi

  28. [36]

    Zhu, H., Li, Q., Tao, C., Hong, Y., Xu, Z., Shen, W., Kaur, S., Ghosh, P., and Qiu, M. (2021). Multispectral camouflage for infrared, visible, lasers and microwave with radiative cooling. Nat Commun 12, 1805. https://doi.org/10.1038/s41467-021-22051-0

    2021 doi

  29. [37]

    Qu, Y., Li, Q., Cai, L., Pan, M., Ghosh, P., Du, K., and Qiu, M. (2018). Thermal camouflage based on the phase-changing material GST. Light Sci Appl 7, 26. https://doi.org/10.1038/s41377-018-0038-5

    2018 doi

  30. [38]

    Hu, R., Xi, W., Liu, Y., Tang, K., Song, J., Luo, X., Wu, J., and Qiu, C.-W. (2021). Thermal camouflaging metamaterials. Materials Today 45, 120–141. https://doi.org/10.1016/j.mattod.2020.11.013

    2021 doi

  31. [39]

    Lee, N., Lim, J.-S., Chang, I., Lee, D., and Cho, H.H. (2021). Transparent Metamaterials for Multispectral Camouflage with Thermal Management. International Journal of Heat and Mass Transfer 173, 121173. https://doi.org/10.1016/j.ijheatmasstransfer.2021.121173

    2021

  32. [40]

    Lee, N., Lim, J.-S., Nam, J., Bae, H.M., and Cho, H.H. (2023). Durable camouflage materials by polyimide nanofilm with thermal management. Applied Surface Science 608, 155107. https://doi.org/10.1016/j.apsusc.2022.155107

    2023

  33. [41]

    Kim, T., Bae, J.-Y., Lee, N., and Cho, H.H. (2019). Hierarchical Metamaterials for Multispectral Camouflage of Infrared and Microwaves. Advanced Functional Materials 29, 1807319. https://doi.org/10.1002/adfm.201807319

    2019 doi

  34. [42]

    Dang, S., and Ye, H. (2021). A visible-infrared-compatible camouflage photonic crystal with heat dissipation by radiation in 5–8 μm. Cell Reports Physical Science 2, 100617. https://doi.org/10.1016/j.xcrp.2021.100617

    2021

  35. [43]

    Xiong, Y., Zhang, G., Tian, Y., Wang, J.-L., Wang, Y., Zhuo, Z., and Zhao, X. (2024). Optimization of a Ge2Sb2Te5-Based Electrically Tunable Phase-Change Thermal Emitter for Dynamic Thermal Camouflage. Materials 17, 1641. https://doi.org/10.3390/ma17071641

    2024 doi

  36. [44]

    Kang, Q., Li, D., Guo, K., Gao, J., and Guo, Z. (2021). Tunable Thermal Camouflage Based on GST Plasmonic Metamaterial. Nanomaterials 11, 260. https://doi.org/10.3390/nano11020260

    2021 doi

  37. [45]

    Lim, J.-S., Lee, N., Kim, T., Chang, I., Nam, J., and Cho, H.H. (2024). Multiresonant Selective Emitter with Enhanced Thermal Management for Infrared Camouflage. ACS Appl. Mater. Interfaces 16, 15416–15425. https://doi.org/10.1021/acsami.3c15504

    2024 doi

  38. [46]

    Li, W., Wu, S., Tian, C., Li, J., Liu, S., and Wu, G. (2024). Multi-resonance coupled metal pattern metamaterial for selective thermal emission. JNP 18, 016012. https://doi.org/10.1117/1.JNP.18.016012

    2024 doi

  39. [47]

    Ronneberger, O., Fischer, P., and Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. In Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 Lecture Notes in Computer Science., N. Navab, J. Hornegger, W. M. Wells, and A. F. Fr...

    2015 doi

  40. [48]

    Goodfellow, I., Bengio, Y., and Courville, A. (2016). Deep learning (MIT press)

    2016

  41. [49]

    Ma, T., Tobah, M., Wang, H., and Guo, L.J. (2022). Benchmarking deep learning-based models on nanophotonic inverse design problems. OES 1, 210012–210015. https://doi.org/10.29026/oes.2022.210012

    2022

  42. [50]

    Li, Y., Kind, T., Folz, J., Vaniya, A., Mehta, S.S., and Fiehn, O. (2021). Spectral entropy outperforms MS/MS dot product similarity for small-molecule compound identification. Nat Methods 18, 1524–1531. https://doi.org/10.1038/s41592-021-01331-z

    2021 doi

  43. [51]

    Shahriari, B., Swersky, K., Wang, Z., Adams, R.P., and de Freitas, N. (2016). Taking the Human Out of the Loop: A Review of Bayesian Optimization. Proceedings of the IEEE 104, 148–175. https://doi.org/10.1109/JPROC.2015.2494218

    2016

  44. [52]

    Snoek, J., Larochelle, H., and Adams, R.P. (2012). Practical Bayesian Optimization of Machine Learning Algorithms. In Advances in Neural Information Processing Systems (Curran Associates, Inc.)

    2012

  45. [53]

    LeCun, Y., Bengio, Y., and Hinton, G. (2015). Deep learning. Nature 521, 436–444. https://doi.org/10.1038/nature14539

    2015 doi

  46. [54]

    Greenhill, S., Rana, S., Gupta, S., Vellanki, P., and Venkatesh, S. (2020). Bayesian Optimization for Adaptive Experimental Design: A Review. IEEE Access 8, 13937–13948. https://doi.org/10.1109/ACCESS.2020.2966228

    2020

  47. [55]

    Hickman, R.J., Aldeghi, M., Häse, F., and Aspuru-Guzik, A. (2022). Bayesian optimization with known experimental and design constraints for chemistry applications. Digital Discovery 1, 732–744. https://doi.org/10.1039/D2DD00028H

    2022 doi

  48. [56]

    Gelbart, M.A., Snoek, J., and Adams, R.P. (2014). Bayesian optimization with unknown constraints. In Proceedings of the Thirtieth Conference on Uncertainty in Artificial Intelligence UAI’14. (AUAI Press), pp. 250–259

    2014

  49. [57]

    Huang, C.-W., Lim, J.H., and Courville, A. (2024). A variational perspective on diffusion-based generative models and score matching. In Proceedings of the 35th International Conference on Neural Information Processing Systems NIPS ’21. (Curran Associates Inc.), pp. 22863–22876

    2024

  50. [58]

    Nichol, A.Q., and Dhariwal, P. (2021). Improved Denoising Diffusion Probabilistic Models. In Proceedings of the 38th International Conference on Machine Learning (PMLR), pp. 8162–8171

    2021

  51. [59]

    Kawar, B., Elad, M., Ermon, S., and Song, J. (2024). Denoising diffusion restoration models. In Proceedings of the 36th International Conference on Neural Information Processing Systems NIPS ’22. (Curran Associates Inc.), pp. 23593–23606

    2024

  52. [60]

    Ajay, A., Du, Y., Gupta, A., Tenenbaum, J., Jaakkola, T., and Agrawal, P. (2023). Is Conditional Generative Modeling all you need for Decision-Making? Preprint at arXiv

    2023

  53. [61]

    Janner, M., Du, Y., Tenenbaum, J.B., and Levine, S. (2022). Planning with Diffusion for Flexible Behavior Synthesis. Preprint at arXiv

    2022

  54. [62]

    Zhu, Z., Zhao, H., He, H., Zhong, Y., Zhang, S., Guo, H., Chen, T., and Zhang, W. (2024). Diffusion Models for Reinforcement Learning: A Survey. Preprint at arXiv

    2024

  55. [63]

    Aydin, K., Ferry, V.E., Briggs, R.M., and Atwater, H.A. (2011). Broadband polarization-independent resonant light absorption using ultrathin plasmonic super absorbers. Nat Commun 2, 517. https://doi.org/10.1038/ncomms1528

    2011 doi

  56. [64]

    Chen, Y.-B., and Chen, C.-J. (2013). Interaction between the magnetic polariton and surface plasmon polariton. Optics Communications 297, 169–175. https://doi.org/10.1016/j.optcom.2013.01.049

    2013 doi

  57. [65]

    Xuan, Y., and Zhang, Y. (2014). Investigation on the physical mechanism of magnetic plasmons polaritons. Journal of Quantitative Spectroscopy and Radiative Transfer 132, 43–51. https://doi.org/10.1016/j.jqsrt.2012.10.020

    2014 doi

  58. [66]

    Wang, L.P., and Zhang, Z.M. (2009). Resonance transmission or absorption in deep gratings explained by magnetic polaritons. Applied Physics Letters 95, 111904. https://doi.org/10.1063/1.3226661

    2009 doi

  59. [67]

    Wang, L.P., and Zhang, Z.M. (2011). Phonon-mediated magnetic polaritons in the infrared region. Opt. Express, OE 19, A126–A135. https://doi.org/10.1364/OE.19.00A126

    2011 doi

  60. [68]

    Lee, B.J., Wang, L.P., and Zhang, Z.M. (2008). Coherent thermal emission by excitation of magnetic polaritons between periodic strips and a metallic film. Opt. Express, OE 16, 11328–11336. https://doi.org/10.1364/OE.16.011328

    2008 doi

  61. [69]

    Lee, N., Kim, T., Lim, J.-S., Chang, I., and Cho, H.H. (2019). Metamaterial-Selective Emitter for Maximizing Infrared Camouflage Performance with Energy Dissipation. ACS Appl. Mater. Interfaces 11, 21250–21257. https://doi.org/10.1021/acsami.9b04478

    2019 doi

  62. [70]

    Lee, N., Lim, J.-S., Chang, I., Bae, H.M., Nam, J., and Cho, H.H. (2022). Flexible Assembled Metamaterials for Infrared and Microwave Camouflage. Advanced Optical Materials 10, 2200448. https://doi.org/10.1002/adom.202200448. PROGRESS AND POTENTIAL Machine learning is reshapin...

    2022 doi

  63. [1782]

    https://doi.org/10.1126/science.1125907

    doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.