Machine Learning Control of Modulation Instability in Optical Fiber

The fields of ultrafast and nonlinear photonics have recently experienced remarkable progress, with machine learning emerging as a key driver in many research efforts. These advances particularly benefit from the development of efficient characterization techniques combined with advanced optical pulse shaping methods designed to control the nonlinear dynamics involved in guided pulse propagation. A notable example is modulation instability (MI), a well-known nonlinear process of guided optics [1]. Based on four-wave mixing, MI is characterized by the emergence of new spectral components, originating from coherent signals (induced MI) or appearing spontaneously through noise amplification (spontaneous MI). In realistic conditions, MI intrinsically involves incoherent and complex processes, ultimately making its experimental control difficult. Here, we present an overview of our recent results focusing on machine learning strategies for the characterization, control, and optimization of incoherent signals generated via modulation instability in optical fibers. Specifically, we numerically and experimentally study the control of spontaneous MI by optical seeding, using weak coherent signals whose parameters are adjusted by machine-learning optimization techniques.