SPA Seminar: Learning to reflect -- Data-driven solutions to singular control problems

Abstract: Stochastic optimal control problems have a long tradition in applied probability, and the questions addressed are of high relevance in a wide variety of fields. Although theoretical solutions are well understood in many scenarios, their practicality suffers from the assumption of known dynamics of the underlying stochastic process, which raises the statistical challenge of developing purely data-driven controls in a nonparametric framework.

In this talk, we discuss how to bring together stochastic control and statistics, which we explore for ergodic singular control problems and underlying diffusions. We compare the results to those of deep Q-learning algorithms.

This is a seminar in our seminar series on Stochastic Pocesses and Applications (SPA).