Robustness criterion for diffractive deep neural networks using normalized cutoff frequency
Optics Express · 13.07.2026 · 34 (14) · 25423
DOI: 10.1364/oe.601000 ↗Kurz gefasst
Fachartikel in der Zeitschrift Optics Express, online erschienen am 30.06.2026. Verfasst von Ning Wang und 6 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Robustness of diffractive deep neural networks (D 2 NNs) as a key performance for the deployment in real-world remains insufficiently elucidated. We propose a criterion to evaluate the robustness of D 2 NN using the normalized cutoff frequency (NCF), and reveal that the robustness is determined by the spatial frequency propagation characteristics of the light field. We demonstrate that despite differences in D 2 NN architectures, the same robustness of D 2 NNs is attributed to equal NCF values in classification and regression tasks. This principle is physically interpretable and widely applicable to linear and nonlinear D 2 NNs. The relationship between the robustness of D 2 NN and the spatial frequency bandwidth of the optical field is discovered, enabling efficient and accurate prediction of the robustness of large-scale D 2 NN models.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Wang, N., Wang, S., Guo, Z., Zhao, Y., Jia, Q., Zhang, J., & Wang, J. (2026). Robustness criterion for diffractive deep neural networks using normalized cutoff frequency. Optics Express, 34(14), 25423. https://doi.org/10.1364/oe.601000
Vancouver
Wang N, Wang S, Guo Z, Zhao Y, Jia Q, Zhang J, et al. Robustness criterion for diffractive deep neural networks using normalized cutoff frequency. Optics Express. 2026;34(14):25423. doi:10.1364/oe.601000
BibTeX
@article{Wang2026,
author = {Wang, Ning and Wang, Siyuan and Guo, Zonglin and Zhao, Yu and Jia, Qi and Zhang, Jingbo and Wang, Jian},
title = {Robustness criterion for diffractive deep neural networks using normalized cutoff frequency},
journal = {Optics Express},
year = {2026},
volume = {34},
number = {14},
pages = {25423},
doi = {10.1364/oe.601000}
}
EUVIMED