Nonparametric approaches to data-driven stochastic control

Abstract

One of the fundamental assumptions in stochastic control of continuous-time processes is that the dynamics of the underlying process are known. This is, however, usually obviously not fulfilled in practice. On the other hand, a rich theory for nonparametric estimation of the characteristics of continuous-time processes has been developed over the last decades. In this talk, we discuss how to bring together these two areas for developing purely data-driven strategies for stochastic control, which we explore for ergodic impulse and singular control problems. We compare the results with those of deep-Q learning algorithms.