Application of Machine Learning Techniques and Neural Networks in the Development of Nanostructured Optical Fibers

This paper explores the application of machine learning techniques, including standard algorithms, ensemble methods, neural networks, and generative inverse design networks (GIDNs), for the rapid design and prediction of optical properties in nanostructured optical fibers. We focus on fibers with thousands of subwavelength elements, such as dispersion-modifying fibers and weakly coupled few-mode fibers, where spatial distribution governs optical parameters like zero-dispersion wavelength and effective refractive index differences between modes. We demonstrate training set construction and optimization of complex refractive index profiles, achieving mode separation exceeding 1.2×10^-3 with a sixfold reduction in dataset size compared to traditional methods. Our approach accelerates design, bypasses time-intensive simulations, and enables the fabrication of advanced fibers, paving the way for rapid prototyping and innovation in photonics.