MURR Georges : Machine Learning-assisted spatiotemporal chaos forecasting

Résumé de thèse : 

Spatiotemporal chaos may be understood as the exponential destruction of information in both time and space, making the dynamics require many spatially distributed chaotic elements to be described. As a consequence of this quick vanishing of information, its forecasting holds immense value due to the occurrence of extreme events in these spatiotemporal chaotic regimes, making it's forecasting a challenging and hot topic. Extreme events are large and short time deviation of an observable, that can have profound implications in contexts such as climate change and natural disaster preparedness like ocean rogue waves, heat waves, floods, earthquakes and strong winds. Optics provide interesting media to study these extreme events owing to the reproductibility of experiments and the large amount of data that can be recorded in a short observation window. In this context, Kerr-based passive cavities are systems of particular interest. Indeed, frequency combs generated by these cavities have significant technological applications and also serve as a fascinating platform to explore complex dynamics such as spatiotemporal chaos.

Here, we have characterized spatiotemporal chaos in Kerr optical frequency combs observed in all fiber cavities. Using a statistical approach based on Kolmogorov structure functions, together with measurement tools of spatiotemporal chaos, correlation function analysis, alongside information theory concepts, we were able to have a very precise description of the dynamics inside the chaos which paves the way to train neural networks aimed at forecasting these complex dynamical systems.

Doctorant : MURR Georges

Directeur(s) de thèse : Saliya COULIBALY