pith. sign in

arxiv: 2512.08109 · v2 · submitted 2025-12-08 · ⚛️ physics.optics

Advantages of Broadband Metalenses for Generalizable Image Classification

Yubo Zhang , Johannes Fr\"och , Jinlin Xiang , Shane Colburn , Myunghoo Lee , Zhihao Zhou , Minho Choi , Eli Shlizerman
show 1 more author
Arka Majumdar
This is my paper

Pith reviewed 2026-05-16 23:46 UTC · model grok-4.3

classification ⚛️ physics.optics
keywords broadband metalensoptical neural networksimage classificationmetasurface encodermodulation transfer functiongeneralizable visionmeta-optics
0
0 comments X

The pith

A visible-spectrum broadband metalens achieves image classification accuracy comparable to high-end optics and outperforms the hyperboloid baseline across sensor sizes and backends.

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

The paper shows that a single-layer metalens can act as a static meta-optical encoder in optical neural networks for generalizable image classification. A broadband design in the visible spectrum reaches accuracy levels on par with high-end sensor-limited optics. It consistently surpasses the corresponding hyperboloid baseline over wide ranges of pixel sizes and digital processing methods. End-to-end optimization for ImageNet classification balances the modulation transfer function across wavelengths, which points to spatial-frequency preservation as a key performance driver. These results give concrete guidance on building energy-efficient ONNs that still generalize well.

Core claim

A visible-spectrum broadband metalens can achieve image classification accuracy comparable to high-end, sensor-limited optics and consistently outperforms the corresponding hyperboloid baseline across a wide range of sensor pixel sizes and digital backends. End-to-end optimization of a single-aperture metasurface for ImageNet classification balances the modulation transfer function within the sensor-detectable passband, indicating that preservation of spatial-frequency information is an important factor influencing ONN performance.

What carries the argument

The single-layer broadband metalens as a meta-optical encoder whose balanced modulation transfer function across wavelengths preserves spatial-frequency content for downstream classification.

If this is right

  • Optical neural networks can use simple single-layer broadband metalenses for low-energy classification with little accuracy penalty.
  • Meta-optical encoder design should prioritize broadband operation and uniform MTF rather than narrowband focusing.
  • Classification performance remains stable across different sensor pixel sizes and choices of digital backend.
  • Task-driven end-to-end optimization naturally produces metasurfaces that maintain spatial-frequency information across the visible band.

Where Pith is reading between the lines

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

  • Broadband metalens encoders may extend directly to other vision tasks such as detection or segmentation without retraining the optics.
  • Robustness to pixel-size variation suggests these designs could be paired with a range of commercial image sensors.
  • The observed MTF-balancing behavior offers a design heuristic for metasurface optimization in other optical computing settings.
  • Real deployment would next require quantifying how fabrication variance affects the observed accuracy advantage.

Load-bearing premise

The simulated or prototyped metalens performance accurately reflects real-world optical behavior without major losses from fabrication errors, unmodeled aberrations, or sensor-specific effects.

What would settle it

Fabricate the optimized broadband metalens and the hyperboloid baseline, then measure end-to-end classification accuracy on the same ImageNet subset using identical sensors and digital backends.

Figures

Figures reproduced from arXiv: 2512.08109 by Arka Majumdar, Eli Shlizerman, Jinlin Xiang, Johannes Fr\"och, Minho Choi, Myunghoo Lee, Shane Colburn, Yubo Zhang, Zhihao Zhou.

Figure 1
Figure 1. Figure 1: a. Schematic of the single aperture meta-optic encoder. An image of axolotl from ImageNet30 is displayed [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: a. Schematic of a single scatterer in a 2D [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: a. RGB PSFs of End-to-end lens (up) and hyperboloid lens (down). b. Log-scaled MTF of the end-to-end [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: a. Increasing the sensor pixel size spatially averages the incident intensity, which reduces the resolvable PSF [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: a. Radial MTF curves before (dashed RGB lines) and after (solid RGB lines) end-to-end optimization using [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

Optical neural networks (ONNs) are gaining increasing attention to accelerate machine learning tasks. In particular, static meta-optical encoders designed for task-specific pre-processing have demonstrated orders of magnitude smaller energy consumption over purely digital counterparts, albeit at the cost of a slight degradation in classification accuracy. However, a lack of generalizability poses serious challenges for wide deployment of static meta-optical front-ends. Here, we investigate the utility of a single-layer metalens as a meta-optical encoder in ONNs for generalizable image classification. Specifically, we show that a visible-spectrum broadband metalens can achieve image classification accuracy comparable to high-end, sensor-limited optics and consistently outperforms the corresponding hyperboloid baseline across a wide range of sensor pixel sizes and digital backends. We further design an end-to-end optimized single-aperture metasurface for ImageNet classification and observe that the optimization tends to balance the modulation transfer function (MTF) across wavelengths within the sensor-detectable passband. Together, these observations suggest that the preservation of spatial-frequency information is an important factor influencing the performance of ONNs. Our results provide physical insight into the process of task-driven optical optimization and offer practical guidance for the design of high-performance ONNs and meta-optical encoders for generalizable computer vision tasks.

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

3 major / 1 minor

Summary. The manuscript investigates single-layer broadband metalenses as static meta-optical encoders in optical neural networks for generalizable image classification. It claims that a visible-spectrum broadband metalens achieves classification accuracy comparable to high-end sensor-limited optics and consistently outperforms the corresponding hyperboloid baseline across a wide range of sensor pixel sizes and digital backends. An end-to-end optimized single-aperture metasurface for ImageNet classification is shown to balance the modulation transfer function (MTF) across wavelengths in the sensor-detectable passband, leading to the conclusion that spatial-frequency preservation is key to ONN performance.

Significance. If the simulation results hold under realistic conditions, the work supplies concrete physical insight into how task-driven optimization shapes meta-optic design and offers practical guidance for broadband meta-optical encoders in energy-efficient computer vision systems. The MTF-balancing observation is a useful design heuristic that could influence future ONN architectures.

major comments (3)
  1. [Abstract and Results] Abstract and Results section: The central performance claim (comparable accuracy to high-end optics and consistent outperformance of the hyperboloid baseline) is presented without any tabulated accuracy values, standard deviations, or error bars across the tested sensor pixel sizes and backends. This omission is load-bearing because the magnitude and statistical significance of the reported advantage cannot be assessed from the given information.
  2. [Methods] Methods section: No quantitative bounds or sensitivity analysis are provided for phase-error tolerance, material dispersion mismatch, or fabrication-induced wavefront aberrations. Because the entire claim rests on simulated MTF and classification accuracy translating to physical devices, the absence of these tolerance studies prevents evaluation of whether the reported margins survive realistic fabrication.
  3. [Results] Results section: The end-to-end optimization procedure that produces the balanced-MTF metasurface is described only at a high level; the loss function, wavelength sampling, and constraint set used during optimization are not specified. This detail is required to determine whether the observed MTF balancing is a robust outcome or an artifact of the particular optimization setup.
minor comments (1)
  1. [Figures] Figure captions should explicitly state the number of independent simulation runs or random seeds used to generate the plotted accuracy curves.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed review. We have addressed each major comment below and will revise the manuscript to enhance clarity, reproducibility, and quantitative support for our claims.

read point-by-point responses

  1. Referee: [Abstract and Results] Abstract and Results section: The central performance claim (comparable accuracy to high-end optics and consistent outperformance of the hyperboloid baseline) is presented without any tabulated accuracy values, standard deviations, or error bars across the tested sensor pixel sizes and backends. This omission is load-bearing because the magnitude and statistical significance of the reported advantage cannot be assessed from the given information.

    Authors: We agree that tabulated quantitative results are necessary to evaluate the performance claims rigorously. In the revised manuscript, we will add a table reporting classification accuracies (with standard deviations and error bars) for the broadband metalens, hyperboloid baseline, and high-end optics across all tested sensor pixel sizes and digital backends. This will enable direct assessment of the magnitude and statistical significance of the observed advantages. revision: yes

  2. Referee: [Methods] Methods section: No quantitative bounds or sensitivity analysis are provided for phase-error tolerance, material dispersion mismatch, or fabrication-induced wavefront aberrations. Because the entire claim rests on simulated MTF and classification accuracy translating to physical devices, the absence of these tolerance studies prevents evaluation of whether the reported margins survive realistic fabrication.

    Authors: We acknowledge that fabrication tolerance analysis is essential for assessing real-world viability. The current work emphasizes ideal simulations to establish fundamental advantages. In revision, we will add quantitative sensitivity bounds for phase errors and dispersion mismatch drawn from typical fabrication tolerances reported in the literature, along with a discussion of their expected impact on MTF and classification accuracy. A comprehensive Monte Carlo study of wavefront aberrations is beyond the present scope but will be noted as future work. revision: partial

  3. Referee: [Results] Results section: The end-to-end optimization procedure that produces the balanced-MTF metasurface is described only at a high level; the loss function, wavelength sampling, and constraint set used during optimization are not specified. This detail is required to determine whether the observed MTF balancing is a robust outcome or an artifact of the particular optimization setup.

    Authors: We thank the referee for highlighting the need for optimization details to ensure reproducibility. The revised manuscript will explicitly specify the loss function (including classification cross-entropy and MTF regularization terms), wavelength sampling strategy (discrete points across 400-700 nm), and constraint sets (phase range, aperture size, and material properties). These additions will demonstrate that the observed MTF balancing arises from the task-driven optimization rather than setup-specific artifacts. revision: yes

Circularity Check

0 steps flagged

No circularity detected; claims rest on independent simulations

full rationale

The paper's core results compare simulated classification accuracy of a broadband metalens against a hyperboloid baseline across sensor sizes and backends, plus an end-to-end optimization that balances MTF. These are presented as outputs of numerical modeling and optimization runs rather than definitions, fitted parameters renamed as predictions, or self-citation chains. No load-bearing equations or premises reduce to their own inputs by construction. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper relies on standard domain assumptions in electromagnetic simulation of metasurfaces and machine learning training pipelines without introducing new free parameters or invented entities beyond typical design choices.

axioms (1)
  • domain assumption Electromagnetic simulation accurately models light propagation and focusing through nanostructured metalenses across visible wavelengths.
    Performance comparisons and optimization results depend on this modeling fidelity.

pith-pipeline@v0.9.0 · 5551 in / 1219 out tokens · 34171 ms · 2026-05-16T23:46:42.687626+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

42 extracted references · 42 canonical work pages

  1. [1]

    Fröch, Luocheng Huang, Zhihao Zhou, Virat Tara, Zhuoran Fang, Shane Colburn, Alan Zhan, Minho Choi, Arnab Manna, Andrew Tang, Zheyi Han, Karl F

    Johannes E. Fröch, Luocheng Huang, Zhihao Zhou, Virat Tara, Zhuoran Fang, Shane Colburn, Alan Zhan, Minho Choi, Arnab Manna, Andrew Tang, Zheyi Han, Karl F. Böhringer, and Arka Majumdar. Full color visible imaging with crystalline silicon meta-optics.Light: Science & Applications, 14(1), June 2025

    work page 2025

  2. [2]

    Design of an infrared wide-angle metalens for medical endoscopic imaging systems.Optics Express, 33(14):29182–29196, 2025

    Xinjie Zhao, Xing Peng, Shaohui Xu, Shiqing Li, Hongbing Cao, and Feng Shi. Design of an infrared wide-angle metalens for medical endoscopic imaging systems.Optics Express, 33(14):29182–29196, 2025

    work page 2025

  3. [3]

    Extended depth of focus metalens toward the entire long-wave infrared spectrum through inverse design framework.Optics Express, 33(18):39081–39091, 2025

    Yongcan Zeng, Xiaoling Ge, Yuqing Zhang, Siyang Xiao, Fen Zhao, Chongchong Ran, Mingjie Wu, Fengyuan Gan, Jiagui Wu, and Junbo Yang. Extended depth of focus metalens toward the entire long-wave infrared spectrum through inverse design framework.Optics Express, 33(18):39081–39091, 2025

    work page 2025

  4. [4]

    Metasurface generation of paired accelerating and rotating optical beams for passive ranging and scene reconstruction.ACS Photonics, 7(6):1529–1536, May 2020

    Shane Colburn and Arka Majumdar. Metasurface generation of paired accelerating and rotating optical beams for passive ranging and scene reconstruction.ACS Photonics, 7(6):1529–1536, May 2020

    work page 2020

  5. [5]

    Flat optics with designer metasurfaces.Nature materials, 13(2):139–150, 2014

    Nanfang Yu and Federico Capasso. Flat optics with designer metasurfaces.Nature materials, 13(2):139–150, 2014

    work page 2014

  6. [6]

    Metasurface based on phase change materials for electrically reconfigurable thz beam steering in copolarized transmission mode.Scientific Reports, 15(1):40666, 2025

    Krishna Kumar, Borja Vidal, and Carlos Garcia-Meca. Metasurface based on phase change materials for electrically reconfigurable thz beam steering in copolarized transmission mode.Scientific Reports, 15(1):40666, 2025

    work page 2025

  7. [7]

    Incoherent pattern recognition with phase-only filters.Optics letters, 17(23):1703–1705, 1992

    Joseph van der Gracht and Joseph N Mait. Incoherent pattern recognition with phase-only filters.Optics letters, 17(23):1703–1705, 1992

    work page 1992

  8. [8]

    Computer-free, all-optical reconstruction of holograms using diffractive networks.ACS Photonics, 8(11):3375–3384, October 2021

    Md Sadman Sakib Rahman and Aydogan Ozcan. Computer-free, all-optical reconstruction of holograms using diffractive networks.ACS Photonics, 8(11):3375–3384, October 2021

    work page 2021

  9. [9]

    Reprogrammable metasurface holographic image encryption technology based on a three-dimensional discrete hyperchaotic system.Optics Express, 32(22):38703–38719, 2024

    Kaiyun Bi, Guanmao Zhang, Jilong Zhang, Guangchao Diao, Bochuan Xing, Mengjie Cui, Zhilin Ge, and Yuze Du. Reprogrammable metasurface holographic image encryption technology based on a three-dimensional discrete hyperchaotic system.Optics Express, 32(22):38703–38719, 2024

    work page 2024

  10. [10]

    Meta-optic accelerators for object classifiers.Sci

    Hanyu Zheng, Quan Liu, You Zhou, Ivan I Kravchenko, Yuankai Huo, and Jason Valentine. Meta-optic accelerators for object classifiers.Sci. Adv., 8(30):eabo6410, July 2022

    work page 2022

  11. [11]

    Analysis of diffractive optical neural networks and their integration with electronic neural networks.IEEE Journal of Selected Topics in Quantum Electronics, 26(1):1–14, 2020

    Deniz Mengu, Yi Luo, Yair Rivenson, and Aydogan Ozcan. Analysis of diffractive optical neural networks and their integration with electronic neural networks.IEEE Journal of Selected Topics in Quantum Electronics, 26(1):1–14, 2020

    work page 2020

  12. [12]

    Diffractive optical computing in free space.Nature Communications, 15(1):1525, 2024

    Jingtian Hu, Deniz Mengu, Dimitrios C Tzarouchis, Brian Edwards, Nader Engheta, and Aydogan Ozcan. Diffractive optical computing in free space.Nature Communications, 15(1):1525, 2024

    work page 2024

  13. [13]

    Programmable diffractive deep neural networks enabled by integrated rewritable metasurfaces

    Sanaz Zarei. Programmable diffractive deep neural networks enabled by integrated rewritable metasurfaces. Scientific Reports, 15(1):35624, 2025

    work page 2025

  14. [14]

    Optical generative models.Nature, 644(8078):903–911, 2025

    Shiqi Chen, Yuhang Li, Yuntian Wang, Hanlong Chen, and Aydogan Ozcan. Optical generative models.Nature, 644(8078):903–911, 2025

    work page 2025

  15. [15]

    Photonic advantage of optical encoders.Nanophotonics, 13(7):1191–1196, 2024

    Luocheng Huang, Quentin AA Tanguy, Johannes E Fröch, Saswata Mukherjee, Karl F Böhringer, and Arka Majumdar. Photonic advantage of optical encoders.Nanophotonics, 13(7):1191–1196, 2024

    work page 2024

  16. [16]

    Optical neural networks: progress and challenges.Light: Science & Applications, 13(1):263, 2024

    Tingzhao Fu, Jianfa Zhang, Run Sun, Yuyao Huang, Wei Xu, Sigang Yang, Zhihong Zhu, and Hongwei Chen. Optical neural networks: progress and challenges.Light: Science & Applications, 13(1):263, 2024

    work page 2024

  17. [17]

    All-optical neural network with nonlinear activation functions.Optica, 6(9):1132–1137, 2019

    Ying Zuo, Bohan Li, Yujun Zhao, Yue Jiang, You-Chiuan Chen, Peng Chen, Gyu-Boong Jo, Junwei Liu, and Shengwang Du. All-optical neural network with nonlinear activation functions.Optica, 6(9):1132–1137, 2019

    work page 2019

  18. [18]

    Programmable metasurfaces for future photonic artificial intelligence.Nature Reviews Physics, pages 1–17, 2025

    Loubnan Abou-Hamdan, Emil Marinov, Peter Wiecha, Philipp del Hougne, Tianyu Wang, and Patrice Genevet. Programmable metasurfaces for future photonic artificial intelligence.Nature Reviews Physics, pages 1–17, 2025

    work page 2025

  19. [19]

    Peter L. McMahon. The physics of optical computing.Nature Reviews Physics, 5(12):717–734, October 2023

    work page 2023

  20. [20]

    Free-space optical encoder for computer vision.npj Nanophotonics, 2(1):36, 2025

    Minho Choi and Arka Majumdar. Free-space optical encoder for computer vision.npj Nanophotonics, 2(1):36, 2025. 9

    work page 2025

  21. [21]

    Transfer- able polychromatic optical encoder for neural networks.Nat

    Minho Choi, Jinlin Xiang, Anna Wirth-Singh, Seung-Hwan Baek, Eli Shlizerman, and Arka Majumdar. Transfer- able polychromatic optical encoder for neural networks.Nat. Commun., 16(1):5623, July 2025

    work page 2025

  22. [22]

    The second optical metasurface revolution: moving from science to technology.Nature Reviews Electrical Engineering, 2(2):125–143, 2025

    Mark L Brongersma, Ragip A Pala, Hatice Altug, Federico Capasso, Wei Ting Chen, Arka Majumdar, and Harry A Atwater. The second optical metasurface revolution: moving from science to technology.Nature Reviews Electrical Engineering, 2(2):125–143, 2025

    work page 2025

  23. [23]

    Metasurface array for single-shot spectroscopic ellipsometry.Light: Science & Applications, 13(1):88, 2024

    Shun Wen, Xinyuan Xue, Shuai Wang, Yibo Ni, Liqun Sun, and Yuanmu Yang. Metasurface array for single-shot spectroscopic ellipsometry.Light: Science & Applications, 13(1):88, 2024

    work page 2024

  24. [24]

    Metasurface-enabled single-shot and complete mueller matrix imaging.Nature Photonics, 18(7):704– 712, 2024

    Aun Zaidi, Noah A Rubin, Maryna L Meretska, Lisa W Li, Ahmed H Dorrah, Joon-Suh Park, and Federico Capasso. Metasurface-enabled single-shot and complete mueller matrix imaging.Nature Photonics, 18(7):704– 712, 2024

    work page 2024

  25. [25]

    Spatially varying nanophotonic neural networks.Science Advances, 10(45):eadp0391, 2024

    Kaixuan Wei, Xiao Li, Johannes Froech, Praneeth Chakravarthula, James Whitehead, Ethan Tseng, Arka Ma- jumdar, and Felix Heide. Spatially varying nanophotonic neural networks.Science Advances, 10(45):eadp0391, 2024

    work page 2024

  26. [26]

    Fully parallel, high-speed incoherent optical method for performing discrete fourier transforms.Optics Letters, 2(1):1–3, 1978

    Joseph W Goodman, AR Dias, and LM Woody. Fully parallel, high-speed incoherent optical method for performing discrete fourier transforms.Optics Letters, 2(1):1–3, 1978

    work page 1978

  27. [27]

    Universal linear intensity transformations using spatially incoherent diffractive processors.Light: Science & Applications, 12(1):195, 2023

    Md Sadman Sakib Rahman, Xilin Yang, Jingxi Li, Bijie Bai, and Aydogan Ozcan. Universal linear intensity transformations using spatially incoherent diffractive processors.Light: Science & Applications, 12(1):195, 2023

    work page 2023

  28. [28]

    Knowledge distillation circumvents nonlinearity for optical convolutional neural networks.Applied Optics, 61(9):2173–2183, 2022

    Jinlin Xiang, Shane Colburn, Arka Majumdar, and Eli Shlizerman. Knowledge distillation circumvents nonlinearity for optical convolutional neural networks.Applied Optics, 61(9):2173–2183, 2022

    work page 2022

  29. [29]

    A knowledge-inherited learning for intelligent metasurface design and assembly.Light: Science & Applications, 12(1):82, 2023

    Yuetian Jia, Chao Qian, Zhixiang Fan, Tong Cai, Er-Ping Li, and Hongsheng Chen. A knowledge-inherited learning for intelligent metasurface design and assembly.Light: Science & Applications, 12(1):82, 2023

    work page 2023

  30. [30]

    Neural tangent knowledge distillation for optical convolutional networks.arXiv preprint arXiv:2508.08421, 2025

    Jinlin Xiang, Minho Choi, Yubo Zhang, Zhihao Zhou, Arka Majumdar, and Eli Shlizerman. Neural tangent knowledge distillation for optical convolutional networks.arXiv preprint arXiv:2508.08421, 2025

    work page arXiv 2025

  31. [31]

    Glaucoma identification using convolutional neural networks ensemble for optic disc and cup segmentation.IEEE Access, 12:82720–82729, 2024

    Sandra Virbukait˙e, Jolita Bernataviˇcien˙e, and Daiva Imbrasien˙e. Glaucoma identification using convolutional neural networks ensemble for optic disc and cup segmentation.IEEE Access, 12:82720–82729, 2024

    work page 2024

  32. [32]

    Tkil: tangent kernel approach for class balanced incremental learning.arXiv preprint arXiv:2206.08492, 2022

    Jinlin Xiang and Eli Shlizerman. Tkil: tangent kernel approach for class balanced incremental learning.arXiv preprint arXiv:2206.08492, 2022

    work page arXiv 2022

  33. [33]

    Eigencwd: a spatially varying deconvolution algorithm for single metalens imaging.Optics Express, 33(13):28481–28492, 2025

    Joel Yeo, N Duane Loh, Ramon Paniagua-Dominguez, and Arseniy I Kuznetsov. Eigencwd: a spatially varying deconvolution algorithm for single metalens imaging.Optics Express, 33(13):28481–28492, 2025

    work page 2025

  34. [34]

    Neural nano-optics for high-quality thin lens imaging.Nature communications, 12(1):6493, 2021

    Ethan Tseng, Shane Colburn, James Whitehead, Luocheng Huang, Seung-Hwan Baek, Arka Majumdar, and Felix Heide. Neural nano-optics for high-quality thin lens imaging.Nature communications, 12(1):6493, 2021

    work page 2021

  35. [35]

    Beating spectral bandwidth limits for large aperture broadband nano-optics.Nature communications, 16(1):3025, 2025

    Johannes E Fröch, Praneeth Chakravarthula, Jipeng Sun, Ethan Tseng, Shane Colburn, Alan Zhan, Forrest Miller, Anna Wirth-Singh, Quentin AA Tanguy, Zheyi Han, et al. Beating spectral bandwidth limits for large aperture broadband nano-optics.Nature communications, 16(1):3025, 2025

    work page 2025

  36. [36]

    End-to-end optimization of metasurfaces for imaging with compressed sensing.ACS Photonics, 11(5):2077–2087, 2024

    Gaurav Arya, William F Li, Charles Roques-Carmes, Marin Soljacic, Steven G Johnson, and Zin Lin. End-to-end optimization of metasurfaces for imaging with compressed sensing.ACS Photonics, 11(5):2077–2087, 2024

    work page 2077

  37. [37]

    Design framework for metasurface optics-based convolutional neural networks.Applied Optics, 60(15):4356–4365, 2021

    Carlos Mauricio Villegas Burgos, Tianqi Yang, Yuhao Zhu, and A Nickolas Vamivakas. Design framework for metasurface optics-based convolutional neural networks.Applied Optics, 60(15):4356–4365, 2021

    work page 2021

  38. [38]

    Hybrid optical-electronic convolutional neural networks with optimized diffractive optics for image classification.Scientific reports, 8(1):12324, 2018

    Julie Chang, Vincent Sitzmann, Xiong Dun, Wolfgang Heidrich, and Gordon Wetzstein. Hybrid optical-electronic convolutional neural networks with optimized diffractive optics for image classification.Scientific reports, 8(1):12324, 2018

    work page 2018

  39. [39]

    From performance to structure: a comprehensive survey of advanced metasurface design for next-generation imaging.npj Nanophotonics, 2(1):39, 2025

    Yunhui Zeng, Haopeng Zhong, Zhenwei Long, Hongkun Cao, and Xin Jin. From performance to structure: a comprehensive survey of advanced metasurface design for next-generation imaging.npj Nanophotonics, 2(1):39, 2025

    work page 2025

  40. [40]

    Designing lensless imaging systems to maximize information capture.arXiv preprint arXiv:2506.08513, 2025

    Leyla A Kabuli, Henry Pinkard, Eric Markley, Clara S Hung, and Laura Waller. Designing lensless imaging systems to maximize information capture.arXiv preprint arXiv:2506.08513, 2025

    work page arXiv 2025

  41. [41]

    Information-driven design of imaging systems , url =

    Henry Pinkard, Leyla Kabuli, Eric Markley, Tiffany Chien, Jiantao Jiao, and Laura Waller. Information-driven design of imaging systems.arXiv preprint arXiv:2405.20559, 2024

    work page arXiv 2024

  42. [42]

    Full-color metaoptical imaging in visible light.Advanced Photonics Research, 3(5):2100265, 2022

    Luocheng Huang, Shane Colburn, Alan Zhan, and Arka Majumdar. Full-color metaoptical imaging in visible light.Advanced Photonics Research, 3(5):2100265, 2022. 10 5 Supplement 5.0.1 Material and Fabrication The metaslenses (Fig. S1) were fabricated on a Si3N4–on–quartz platform using standard nanofabrication techniques. First, a thin Si3N4 film was deposite...

    work page 2022