Thejani Gamage: Literature Survey on Stochastic Control Theory and Reinforcement Learning Algorithms

In this talk we would discuss main building blocks of the RL algorithms in general, namely Policy Update and Policy Evaluation Methods. Under the Policy Update methods, one of the most used algorithms are the ones that use a Policy Gradient method via a parameterized family of policies. These methods were originally designed for discrete time and space, but in many scenarios they are generalizable to the continuous time and space.  Our main focus is applying Reinforcement Learning to solve a continuous time stochastic optimal control problem, set up the Policy Update and Policy Evaluation methodologies for such a problem and analyze their convergence.