EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Optimized dimension-reduction algorithm on input layer for diffractive neural networks

Ruisi Li, Qian Ma, Ze Gu, Yu Ming Ning, Xinxin Gao, Tie Jun Cui

APL Photonics · 01.07.2025 · 10 (7)

DOI: 10.1063/5.0279686 ↗

Kurz gefasst

Fachartikel in der Zeitschrift APL Photonics, online erschienen am 16.07.2025. Verfasst von Ruisi Li und 5 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key Research and Development Program of China, Special Fund for Key Basic Research in Jiangsu Province, Jiangsu Joint Laboratory of Multidimensional Perceptual Information Technology.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Diffractive neural network (DNN) has provided a novel solution to address complex artificial intelligence tasks due to its ultrahigh processing speed and parallel computing capability. However, traditional DNNs lack fine design of input layers, which limits their performance in miniaturized applications. Here, we propose an optimized dimension-reduction algorithm based on an improved Fisher score to directly filter the diffractive information injected into the network, enabling DNNs to obtain inputs with stronger resolving capabilities. Numerical results demonstrate that our approach achieves accuracy improvements of 5% and 7.3% on the MNIST and Fashion-MNIST classification datasets, respectively, compared to the uniform grid sampling method that is commonly used in the existing dimension-reduction DNNs. In addition, when addressing the issue of complex input image distortion, our method holds the accuracy advantage of 4.8% in the pincushion distorted MNIST dataset and expands it to 6.7% in the tangential distortion scenario. Our optimized dimension-reduction algorithm has effectively utilized the information in diffractive fields, providing an efficient solution to the miniaturization of DNNs and promoting their application prospects in wearable devices, industrial inspection, and autonomous driving systems.

Förderung

  • National Key Research and Development Program of China
  • Special Fund for Key Basic Research in Jiangsu Province
  • Jiangsu Joint Laboratory of Multidimensional Perceptual Information Technology
  • National Natural Science Foundation of China
  • Natural Science Foundation of Jiangsu Province
  • Major Project of Natural Science Foundation of Jiangsu Province
  • Young Elite Scientists Sponsorship Program by CAST
  • State Key Laboratory of Millimeter Waves

Zitieren

APA

Li, R., Ma, Q., Gu, Z., Ning, Y. M., Gao, X., & Cui, T. J. (2025). Optimized dimension-reduction algorithm on input layer for diffractive neural networks. APL Photonics, 10(7). https://doi.org/10.1063/5.0279686

Vancouver

Li R, Ma Q, Gu Z, Ning YM, Gao X, Cui TJ. Optimized dimension-reduction algorithm on input layer for diffractive neural networks. APL Photonics. 2025;10(7). doi:10.1063/5.0279686

BibTeX
@article{Li2025,
  author = {Li, Ruisi and Ma, Qian and Gu, Ze and Ning, Yu Ming and Gao, Xinxin and Cui, Tie Jun},
  title = {Optimized dimension-reduction algorithm on input layer for diffractive neural networks},
  journal = {APL Photonics},
  year = {2025},
  volume = {10},
  number = {7},
  doi = {10.1063/5.0279686}
}
Kostenlos starten →