Can we predict extreme events?

Predicting extreme events, rare and intense phenomena, is a complex challenge. In our latest study, we have made significant advances using machine learning techniques to classify these events in experimental data from a chaotic micro-cavity laser. Despite experimental constraints limiting accurate data collection, our method achieved 75% to 100% accuracy in predicting these events with a warning time of up to twice the theoretical time horizon. Surprisingly, we found that using non-local spatial data improved long-term prediction accuracy over local data.

This advance is of considerable importance, as it enables the prediction of extreme events in high-dimensional chaotic systems, even with incomplete experimental data and noise. The implications of this research are far-reaching, touching on many areas, both theoretically and practically.

We are convinced that our innovative approach will arouse the interest of a wide audience and open up new perspectives in the prediction of extreme events.

Link : https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.130.223801