Causal discovery based on hierarchical reinforcement learning

Jingchi Jiang, Rujia Shen, Chao Zhao, Yi Guan, Xuehui Yu, Xuelian Fu

Abstract: Highlights•A causal discovery method based on hierarchical reinforcement learning.•Interdependent high-/low-level policies divide causal discovery into two subtasks.•The high-level policy iteratively generates causal skeletons as subgoals.•The low-level policy identifies causal directions of pairs of variables.•A curiosity module with a graph comparator is introduced for exploration efficiency.