Machine learning for ultrafast nonlinear fibre photonics

Abstract

We provide an overview of our latest advances in the application of machine learning methods to ultrafast nonlinear fibre optics. We establish that neural networks are capable of accurately forecasting the temporal and spectral properties of optical signals that are obtained after propagation in the focusing or defocusing regimes of nonlinearity. Machine learning is also efficient in addressing the related inverse problem as well as providing insights into the underlying physical process. In addition, we illustrate the use of evolutionary algorithms to access and optimise complex nonlinear dynamics of ultrafast fibre lasers.