Evaluating a neural network and a convolutional neural network for predicting soliton properties in a quantum noise environment

Abstract

With its applications in science and engineering, supercontinuum (SC) generation is a phenomenon widely studied in nonlinear fiber optics. The SC spectral properties are not difficult to measure, except those related to time. Fortunately, machine learning can help predict the time behavior of various nonlinear optics phenomena using spectral characteristics. In this study, supervised machine learning tools are used to evaluate the prediction accuracy of the soliton properties in a noisy environment. A neural network (NN) and a convolutional neural network (CNN) are implemented to assess the performance of these techniques in relation to predicting soliton properties when noise is included in a laser that pumps a nonlinear fiber optics. We conclude that the CNN shows better performance compared with NN, as it involves more data with the same quantity of simulations conducted in both cases, whereas NN can better predict the target in the absence of noise.