Presentation Date:
Location:
Machine Learning for Fluid Dynamics
Petros Koumoutsakos
Recent advances in machine learning and an ever-increasing availability of data offer new perspectives (and hope) for solving long-standing fluid mechanics problems. Despite early connections dating back to Kolmogorov, the link between fluid mechanics and machine learning has not been fully explored. The situation is rapidly changing with machine learning algorithms entering in numerous efforts for modelling, optimising and controlling fluid flows. In this talk, Petros Koumoutsakos presents work from his group on the interface of fluid mechanics and machine learning ranging from low order models for turbulent flows to deep reinforcement learning algorithms and Bayesian experimental design for collective swimming, with the goal of demonstrating that machine learning has the potential to augment, and possibly even transform, current lines of fluid mechanics research.