首页 | 本学科首页   官方微博 | 高级检索  
     

具有丢失数据的可分解马尔可夫网络结构学习
引用本文:王双成,苑森淼.具有丢失数据的可分解马尔可夫网络结构学习[J].计算机学报,2004,27(9):1221-1228.
作者姓名:王双成  苑森淼
作者单位:吉林大学计算机科学与技术学院,长春,130025
基金项目:国家自然科学基金 (60 2 750 2 6),吉林省自然科学基金 (2 0 0 30 51 7 1 )资助
摘    要:具有丢失数据的可分解马尔可夫网络结构学习是一个重要而困难的研究课题,数据的丢失使变量之间的依赖关系变得混乱,无法直接进行可靠的结构学习.文章结合最大似然树和Gibbs抽样,通过对随机初始化的丢失数据和最大似然树进行迭代修正一调整,得到修复后的完整数据集;在此基础上基于变量之间的基本依赖关系和依赖分析思想进行可分解马尔可夫网络结构学习,能够避免现有的丢失数据处理方法和可分解马尔可夫网络结构学习方法存在的效率和可靠性低等问题.试验结果显示,该方法能够有效地进行具有丢失数据的可分解马尔可夫网络结构学习.

关 键 词:可分解马尔可夫网络  结构学习  丢失数据  Gibbs抽样  最大似然树

Learning Decomposable Markov Network Structure with Missing Data
WANG Shuang-Cheng YUAN Sen-Miao.Learning Decomposable Markov Network Structure with Missing Data[J].Chinese Journal of Computers,2004,27(9):1221-1228.
Authors:WANG Shuang-Cheng YUAN Sen-Miao
Abstract:It is an important and difficult research project to learn decomposable Markov network structure with missing data. Missing data makes the dependency relationship between variables more disordered and it impossible to learn decomposable Markov network structure directly. In this paper, Gibbs sampling is combined with maximum likelihood tree to modify the missing data randomly initialized and regulate maximum likelihood tree iteratively so as to get complete data set. Using complete data set, the decomposable Markov network structure can be learned based on basic dependency relationship between variables and the idea of dependency analysis. The problems of low efficiency and reliability in existing methods of dealing with missing data and learning decomposable Markov network structure can be avoided. Experimental results show that this method can effectively learn decomposable Markov network structure with missing data.
Keywords:decomposable Markov network  structure learning  missing data  Gibbs sampling  maximum likelihood tree
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号