Colloquium - J.M. Burgerscentrum Tour 2025 - Prof. Luca Biferale University of Rome Tor Vergata Data driven tools for Multiphase Turbulence

We present a stochastic method for generating and reconstructing complex signals along the trajectories of small objects passively advected by turbulent  lows. Our approach makes use of generative Diffusion Models, a recently proposed data-driven machine learning technique. We show applications to data augmentation and data-assimilation for 3D tracers in Turbulence, 2D trajectories from NOAA’s Global Drifter Program and extensions to the case of Eulerian multiphase  lows. Supremacy against linear decomposition and Gaussian Regression Processes is analyzed in terms of statistical and point-wise metrics concerning non-Gaussian components and multi-scale properties. Preliminary results concerning generalizability and model collapse will also be discussed, as well as a personal point of view on long term goals and potentialities of black-box data-driven approaches for turbulence.