A0164
Title: Reinforcement learning via nonparametric smoothing in a continuous-time stochastic setting
Authors: Yazhen Wang - University of Wisconsin (United States)
Shang Wu - Fudan University (China) [presenting]
Abstract: Reinforcement learning is mainly developed for discrete-time Markov decision processes. A method is proposed to establish a novel learning approach based on temporal difference and non-parametric smoothing to solve reinforcement learning problems in a continuous-time stochastic setting. Continuous-time temporal-difference learning developed for deterministic models is unstable and diverges when applied to data generated from stochastic models. It is shown that the proposed learning approach has a robust performance for data generated from stochastic models governed by stochastic differential equations. The asymptotic theory is established for the proposed approach, and a numerical study is carried out to solve a pendulum reinforcement learning problem and check the finite sample performance of the proposed method.
