Can machine learning methods predict chaos? (M-level project)

In a series of recently published results, it was claimed that “model-free” machine learning methods can provide better predictions of chaotic dynamical systems than the underlying physical law of the system itself. In this project, the student will study these claims in detail, with the aim of replicating the original results, and critically reflecting on them. The student will gain an understanding of modern machine learning methods, as well as simulation and prediction of spatiotemporal chaotic systems.

Reading: Pathak et al (2018)