Distributed Lifelong Reinforcement Learning With Sub-Linear Regret

In this paper we propose a distributed second-order method for lifelong reinforcement learning (LRL). Upon observing a new task, our algorithm scales state-of-the-art LRL by approximating the Newton direction up-to-any arbitrary precision epsilon > 0, while guaranteeing accurate solutions. We analyze the theoretical properties of this new method and derive, for the first time to the best of our knowledge, sub-linear regret under this setting