Jean Rabault: Using Deep Reinforcement Learning for the control of complex Fluid Mechanics systems
Jean Rabault from the Department of Mathematics who will talk about exciting master of science projects with machine learning applied to the solution of partial differential equations. This is a hot topic with tons of exciting applications and a wide range of applications spanning almost all study directions of the CS program. Everybody is welcome, and pizza is obviously served again.
Fluid Mechanics is a wide field of research. This is the consequence of its importance to many domains, from the aerospace industry to bio-mechanics and medicine. In addition, Fluid Mechanics is governed by the Navier-Stokes equations, which are famously challenging as a consequence of their non-linearity and of the high dimensionality that arises in their solutions, leading to complex problems such as turbulence.Therefore, performing for example Active Flow Control is a challenging problem, which is still largely unsolved.
However, following recent developments in the Machine Learning (ML) / Artificial Intelligence (AI) community, modern algorithms such as those produced from the field of Deep Reinforcement Learning (DRL) are starting to get results on the control of such complex dynamics. In this talk, I will discuss recent works performed at UiO / Mechanics, in which we successfully used modern DRL algorithms to control, for the first time, simple model problems arising from Fluid Mechanics. In addition, I will explain how one can envision to scale those algorithms to problems of increasing complexity within Fluid Mechanics. More generally, this talk should be relevant to the control of any strongly nonlinear, high-dimensionality problem on which classical techniques fail.