Artificial intelligence designer for optical Fibers: Inverse design of a Hollow-Core Anti-Resonant fiber based on a tandem neural network

Abstract

In this work, artificial intelligence (AI) is trained to “study” optical fibers as an AI optical fiber scientist. The dataset is constructed on the structural parameters and confinement loss of hollow-core anti-resonant fibers and is divided into different datasets. An effective approach to overcome the non-uniqueness challenge of inverse designs is the tandem neural network (T-NN), which consists of a forward prediction model and an inverse design model, which are trained by different datasets. The impact of the hyperparameter on the performance of the proposed model is also studied. For the forward prediction model, the mean square error reaches the lowest value of 0.0007 with 6 hidden layers; the nodes of each hidden layer are 800, and the corresponding coefficient of determination R2 is 0.9154. Moreover, the R2CL of the T-NN can reach the maximum value of 0.9881 for the dataset-1-0. The comparison between the target, prediction result, and calculation result proves that the proposed approach is an effective way to design hollow-core anti-resonant fibers intelligently. Compared with current numerical simulation methods, the presented AI model based on the T-NN is more “intelligent”. This study can both accelerate the design of hollow-core anti-resonant fibers and provide guidance on the development of AI scientists in optics.

Keywords

• Hollow-core anti-resonant fibers
• Confinement loss
• Inverse design
• Tandem neural network