Machine learning control and prediction of nonlinear instabilities in fibre optics

We review recent advances in the control and shaping of nonlinear spectral broadening in optical fibers. Focusing on the interplay between coherent seeding and noise-driven modulation instability, we demonstrate how weak coherent optical seeds, when combined with machine learning algorithms, enable precise control over spectral correlations and incoherent broadening dynamics. In particular, we highlight the use of artificial neural networks and evolutionary strategies to optimize seed parameters directly from noisy experimental data. Paired with real-time characterization techniques, these data-driven methods allow on-demand tailoring of nonlinear broadening in regimes that are otherwise highly sensitive to fluctuations and initial conditions. Additionally, we discuss complementary strategies for the spectro-temporal shaping of optical wavepackets, especially in the coherent supercontinuum regime. Overall, our work highlights the power of machine-learning strategies and advanced experimental techniques to probe, control, and exploit nonlinear processes in guided photonic systems.