Author(s)

Jie Bai, Jianfei Li, Zihao Luo, Yaobing Wang, Li Liu

Corresponding Author

Jie Bai

ABSTRACT

More and more deeper reinforcement learning algorithms have been proposed and demonstrated on a series of decision-making domains. However, little research has been hammered at algorithm extraction. With duality in deep reinforcement learning substantially summarized, we propose a conceptually simple framework for deep reinforcement learning based on duality. Then, we propose the dual method of prioritized sampling: prioritized learning. Finally, we give the formula and analysis for the duality with priority. The algorithm implementation and experiment will be put on Part II-Implementation.

KEYWORDS

Deep Reinforcement Learning, Duality, Prioritized Sampling, Prioritized Learning