Taming nonlinear dynamics using Deep Reinforcement Learning

Title: Taming nonlinear dynamics using Deep Reinforcement Learning

by: Jean Rabault, Miroslav Kuchta, Ulysse Réglade and Nicolas Cerardi

Abstract: Machine Learning (ML) methods are a promising way to perform optimal control. In a recent book ('Machine Learning Control - Taming Nonlinear Dynamics and Turbulence', Duriez et. al., 2017), several ML methods were presented as well as a couple of benchmarks. One particular benchmark is a small system of ODEs that present features, such as multimodality and cross-talks, that are representative of more complex systems found in Fluid Mechanics.

In this seminar, we present ongoing work about active control of this system of ODEs.