Scientific Multi-Agent Reinforcement Learning for Turbulence Closures and Flow Control

We will present results from developing and applying scientifiis multiagent reinforcement learning (SMARL)

as closures for Partial Differential Equations and in particular the Navier Stokes in the turbulence regime.

We will also show results from applications of SMARL to the control of unsteady separated flows. We

will discuss advantages and drawbacks in particular with resoect to other supervised learning algorithms.