Hierarchical Reinforcement Learning for Pedagogical Policy Induction

摘要: In interactive e-learning environments such as Intelligent Tutoring Systems, there are pedagogical decisions to make at two main levels of granularity: whole problems and single steps. Recent years have seen growing interest in data-driven techniques for decision making, which can dynamically tailor students’ learning experiences. Most existing approaches, however, treat these equally, or independently, disregarding the long-term impact that tutor may across granularity. this paper, we propose apply an offline, off-policy Gaussian Processes based Hierarchical Reinforcement Learning (HRL) framework induce a hierarchical policy makes both problem step levels. empirical classroom study with 180 students, our results show HRL is significantly more effective than Deep Q-Network (DQN) induced random yet reasonable baseline policy.