Single-wavelength continual learning diffractive deep neural networks
Applied Optics · 01.08.2026 · 65 (22) · 7306
DOI: 10.1364/ao.597691 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Optics, online erschienen am 23.07.2026. Verfasst von Jiaxin Yan und 2 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Electronic deep neural networks suffer from “catastrophic forgetting,” which means they forget previously learned tasks rapidly after being trained with a new task. Recently, optical neural networks like diffractive deep neural networks (D2NNs) have emerged as a potential solution to replace electronic neural networks. D2NN was proposed in 2018, and it shows great potential in many fields such as saliency detection and high-precision object classification. Although D2NNs have a lot of advantages such as light speed signal processing and low power consumption, they have the same issue of “catastrophic forgetting.” In this work, we propose a single-wavelength continual learning diffractive deep neural network (CL-D2NN). As a result, the test accuracy of the MNIST dataset remains 90.05% after switching to the Fashion MNIST dataset, while the test accuracy of the Fashion MNIST dataset is 85.73%. The proposed network can handle over five tasks and has demonstrated its continual learning capability.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Yan, J., Wang, Q., & Zhou, J. (2026). Single-wavelength continual learning diffractive deep neural networks. Applied Optics, 65(22), 7306. https://doi.org/10.1364/ao.597691
Vancouver
Yan J, Wang Q, Zhou J. Single-wavelength continual learning diffractive deep neural networks. Applied Optics. 2026;65(22):7306. doi:10.1364/ao.597691
BibTeX
@article{Yan2026,
author = {Yan, Jiaxin and Wang, Qiqi and Zhou, Junhe},
title = {Single-wavelength continual learning diffractive deep neural networks},
journal = {Applied Optics},
year = {2026},
volume = {65},
number = {22},
pages = {7306},
doi = {10.1364/ao.597691}
}
EUVIMED