Nonlinear Photonic Neural Network With Frequency Comb Lines
Authors:
- Daniel Alejandro Lopez MartinezDMDaniel Alejandro Lopez Martinez
- Usman KhanUKUsman Khan
- Misagh IzadiMIMisagh Izadi
- Yifan DuYDYifan Du
- Nirmal DangiNDNirmal Dangi
- Santosh KumarSKSantosh Kumar
- Yuping HuangYHYuping Huang
Preprint
- 28 February 2026
- preprint
- Published by Optica Publishing Group in Optica Open
Abstract
We propose and demonstrate a hybrid photonic–digital neural network using optical frequency comb lines and their second-order nonlinear interactions for classification and image generation. It codes the features in the amplitude of each line, and uses their relative phase as trainable parameters. For classification, it achieves ~98% accuracy on average on a 40-sample make-moons dataset, with the mean square error reduced from 0.7 to 0.05 post training. For image generation, it implements a conditional variational autoencoder trained with a small dataset of 18 MNIST digits over 3 categories, generating new handwritten digits with reconstruction loss of 0.75. These results demonstrate the feasibility of combining frequency comb lines and nonlinear optics for both discriminative and generative neural network tasks.Keywords
All Related Versions
- Published version: Optics Letters, 51 (9), 2404.