Stochastic Control and Reinforcement Learning

Stochastic control and reinforcement learning have developed over the last several decades at breakneck speed. On the one hand in stochastic control and filtering are well developed mathematical theories that have not found much use in practice beyond the Âlinear-quadratical case. On the other hand reinforcement learning has focused on algorithms and applications and can claim notable successes there. However theoretical understanding of reinforcement learning is still limited. This project will further develop mathematical analysis of convergence problems in reinforcement learning using tools from stochastic analysis, PDE theory and optimal transport.