摘要
从大量复杂的数据中学习贝叶斯网络(Bayesian network,BN)一直是一个难点问题,本文借鉴课程学习的思想,引入了一种适合于BN中节点之间互相影响程度的测量,然后划分课程阶段,分阶段构造无向图骨架,并利用优化函数对骨架进行优化;通过集成策略,将各个集成学习结果所得到的课程权重进行集合,并通过边过滤来减少错误边的出现;最后,通过爬山搜索构建BN结构。实验结果表明,在4个标准数据集上,本文所提方法具有较高的精确度和稳定性。与多种传统贝叶斯结构学习(Bayesian network structure learning,BNSL)方法相比,本文所提方法性能平均提高了37.18%。本文分析结果可为BNSL的增量学习过程进一步提供参考。
Learning Bayesian network(BN)from a large number of complex data has always been a difficult problem.Based on the idea of course learning,this paper introduces a measurement suitable for the degree of mutual influence between nodes in BN structure,then divides the course stage,constructs the undirected graph skeleton in stages,and uses the optimization function to optimize the skeleton.Through the integration strategy,the course weights obtained from each integrated learning result are aggregated,and the error edges are reduced by edge filtering.Finally,the BN structure is constructed by hill-climbing search.The experimental results indicate that the method proposed in this paper exhibits high precision and stability on four standard datasets.Compared with various traditional bayesian network structure learning(BNSL)methods,the method proposed in this paper shows an average performance improvement of 37.18%.The analytical results presented in this paper can further provide insights into the incremental learning process of BNSL.
作者
刘凯越
周鋆
LIU Kaiyue;ZHOU Yun(Science and Technology on Information Systems and Engineering Laboratory,National University of Defense Technology,Changsha 410073,China)
出处
《应用科技》
CAS
2024年第1期1-9,共9页
Applied Science and Technology
基金
国家自然科学基金项目(62276262)
湖南省科技创新计划(2021RC3076)
长沙市优秀青年创新者培训班项目(KQ2009009)。
关键词
贝叶斯网络
结构学习
课程学习
权重
边约束
权重互信息
集成学习
无向图骨架
Bayesian network
structure learning
curriculum learning
weight
edge constraint
weighted mutual information
ensemble learning
undirected graph skeleton
