用深度神经网络实现高精度纳米光子器件的光谱计算
Calculation of the Spectrum of High Precision Nanophotonics with Deep Neural Network
邱维阳 1何成 1郑根让 1易巧玲1
作者信息
- 1. 中山职业技术学院,广东中山 528400
- 折叠
摘要
作为一种数据驱动的研究方法,深度学习已改变了许多研究领域,如计算机视觉、自然语言处理,并已拓展至农业、航空航天、医疗保险等传统行业.本研究基于深度学习这一数据驱动研究方法,开发了一个深度神经网络模型,用于快速光谱计算.仅使用设计空间中一万亿分之一的样本来训练模型,但该模型以超高的精度预测了光谱,将完整数据集的均方根误差低至2.2％,对于清洗后的数据集,均方根误差仅为1.3％,显示了深度学习在光学器件设计中的巨大潜力.
Abstract
As a data-driven research method,deep learning has transformed many fields of research,such as computer vision,natural language processing,and extension to traditional industries such as agriculture,aerospace,and health insurance,etc.Based on the data-driven research method of deep learning,a deep neural network model is developed for fast spectral calculation.The model is trained with only one trillionth of a design space sample,but the model predicted the spectrum with ultra-high accuracy,with a RMS error of as low as 2.2％for the complete dataset and only 1.3％for the cleaned dataset,indicating the great potential of deep learning in optical device design.
关键词
数据驱动/机器学习/深度学习/纳米光子学/超材料Key words
Data driving/Machine learning/Deep learning/Nanophotonics/Metamaterial引用本文复制引用
基金项目
中山职业技术学院高层次人才科研启动项目(KYG2203)
广东省教育厅科研项目(2020KTSCX333)
中山市社会公益与基础研究项目(2020B2056)
出版年
2024
NETL
NSTL
万方数据