Data-driven and equation informed tools for turbulent reconstruction and classification

Data-driven and equation-informed tools to model small-scales, high-frequencies fluctuations in complex flows and to reconstruct large-scale features in gappy-data are presented. Recent implementations of Generative Adversarial Networks to assimilate turbulent data of rotating flows [1] are compared against Principal Orthogonal Decompositions and Nudging [2-3]. My personal understanding of open challenges towards a quantitative-AI for fluid dynamics applications is also presented.

Orateur : Luca Biferale Dept. of Physics, University of Rome, Tor Vergata