/GRL

A generalized reinforcement learning framework that adapts to time-varying and evolving systems.

Primary LanguageJupyter Notebook

Generalized Reinforcement Learning (GRL)

Introduction

Learning a control policy that involves time-varying and evolving system dynamics often poses a great challenge to standard reinforcement learning (RL). This repository will focus on building a GRL framework, an adaptive, concept-driven RL system, and crearting instances of GRL agent under such framework. On a high level perspective, a GRL agent continously and periodically fulfills three interleaving processes -- policy learning, strengthing the reinforcement field via "particle reinforcement" (what? see the paper), and decision concept learning ... (to be continued)