异步策略的强化因果发现方法  

A reinforcement causal discovery approach for asynchronous strategy

作　　者：张英 郭辉 ZHANG Ying;GUO Hui(College of Information Engineering,Ningxia University,Yinchuan,Ningxia 750021,China)

机构地区：[1]宁夏大学信息工程学院,宁夏银川750021

出　　处：《燕山大学学报》2024年第4期356-368,共13页Journal of Yanshan University

基　　金：宁夏自然科学基金资助项目(2021AAC03117)。

摘　　要：研究和发掘事物之间的因果关系是数据科学的核心问题之一。针对因果发现面临着搜索空间超指数量级增长、评价指标低、收敛速度慢且效果差等问题,本文提出一种基于异步策略的强化因果发现方法。首先采用自注意力机制的编码器和单层解码器模型探索数据之间的因果关系;其次,改进强化学习模型中的结构约束,并基于异步优势算法更新网络模型参数;最后,搜索、输出最大奖励的有向无环图。通过实验对比验证了该方法的良好性能。The research and discovery of causality between things is one of the core issues in data science.Causal discovery usually faces problems such as super exponential growth of the search space,low evaluation index,slow rate of convergence and poor effect.To solve them,a reinforcement causal discovery method is proposed for asynchronous strategy.Firstly,a self⁃attentional encoder and a single⁃layer decoder model are used to explore the causal relationship between the data.Secondly,the structural constraints in the reinforcement learning model are improved,and the parameters of the network model are updated based on the asynchronous dominance algorithm.Finally,the directed acyclic graph with the maximum reward is given by searching.The good performance of this method has been verified through experimental comparison.

关 键 词：因果关系 有向无环图 强化因果发现 结构约束 异步优势算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]