Reinforcement Learning in Non-Markovian Environments

11/03/2022

∙

by   Siddharth Chandak, et al.

∙

0

∙

share

Following the novel paradigm developed by Van Roy and coauthors for reinforcement learning in arbitrary non-Markovian environments, we propose a related formulation inspired by classical stochastic control that reduces the problem to recursive computation of approximate sufficient statistics.