影响图模型选择中存在数据依赖性、计算复杂性和非概率关系问题.通过对影响图结构进行分解，提出PSEM算法对影响图的概率结构部分进行模型选择.给出一种BP神经网络，通过对局部效用函数的学习实现效用结构部分的模型选择，并引入权重阈值来避免过拟合.PSEM算法是在SEM算法中引入一种融合先验知识的MDL评分标准来降低传统MDL评分对数据的依赖性，并通过将参数学习和结构评分分开计算提高计算效率.算法比较的结果显示PSEM比标准SEM的时间性能好、对数据依赖性小，且效用部分的结构选择易于实现.
Abstract
In the model selection of influence diagrams(IDs), the problems of the data dependency, the computation complexity and nonprobability relation are discussed. Based on the structure decomposition of IDs, a PSEM algorithm is presented. A BP Neural Network is introduced by learning local utility function of each utility node, and the overfitting is avoided by inducing the threshold of weights. To reduce the data dependency, a new MDL scoring is presented which includes the prior knowledge of network structures. Based on SEM algorithm, PSEM algorithm induces the new MDL scoring, and separates parameters learning from structures scoring to improve the computation efficiency. Compared with SEM algorithm, the performances of both the computation complexity and the data dependency of PSEM algorithm are improved, and the model selection of the utility part is easy to achieve.
关键词
影响图(IDs) /
结构期望最大值(SEM)算法 /
后向神经网络 /
最小描述长度(MDL)评分
{{custom_keyword}} /
Key words
Influence Diagrams (IDs) /
Structural Expectation Maximization (SEM) Algorithm /
Back Propagation (BP) Neural Network /
Minimum Description Length (MDL) Scoring
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] Howard R A, Matheson J E. Influence Diagrams // Howard R A, Matheson J E, eds. Readings on the Principles and Applications of Decision Analysis. Menlo Park, USA: Strategic Decision Group, 1984, Ⅱ: 719792
[2] Charnes J, Shenoy P. Multistage Monte Carlo Method for Solving Influence Diagrams. Management Science, 2004, 50(3): 405418
[3] Diehl M, Haimes Y Y. Influence Diagrams with Multiple Objectives and Tradeoff Analysis. IEEE Trans on Systems, Man, and Cybernetics, 2004, 34(3): 293304
[4] Pettersson J, Wahde M. Application of the Utility Function Method for Behavioral Organization in a Locomotion Task. IEEE Trans on Evolutionary Computation, 2005, 9(5): 506521
[5] Nielseni T D, Jensen F V. Learning a Decision Maker’s Utility Function from (Possibly) Inconsistent Behavior. Artificial Intelligence, 2004, 160(1/2): 5378
[6] Heckerman D, Geiger D. Learning Bayesian Networks: The Combination of Knowledge and Statistical Data. Machine Learning, 1995, 20(3): 197243
[7] Chajewska U, Koller D. Utilities as Random Variables: Density Estimation and Structure Discovery // Proc of the 16th Annual Conference on Uncertainty in Artificial Intelligence. Stanford, USA, 2000: 6371
[8] Wang Shuangcheng, Yuan Senmiao. Research on Learning Bayesian Networks Structure with Missing Data. Journal of Software, 2004, 15(7): 10421048 (in Chinese)
(王双成,苑森淼.具有丢失数据的贝叶斯网络结构学习研究.软件学报, 2004, 15(7): 10421048)
[9] Ji Junzhong, Yan Jing, Liu Chunnian, et al. An Improved Bayesian Networks Learning Algorithm Based on Independence Test and MDL Scoring // Proc of the International Conference on Active Media Technology. Takamatsu, Japan, 2005: 315320
[10] Dodge Y, Zoppe A. Adjusting the EM Algorithm for Design of Experiments with Missing Data // Proc of the 26th International Conference on Information Technology Interfaces. Covtat, Croatia, 2004, Ⅰ: 912
[11] Friedman N. The Bayesian Structural EM Algorithm // Proc of the 14th International Conference on Uncertainty in Artificial Intelligence. Madison, USA, 1998: 129138
[12]Liu Dayou, Wang Fei, Lu Yinan, et al. Research on Learning Bayesian Network Structure Based on Genetic Algorithm. Journal of Computer Research and Development, 2001,38(8):916922 (in Chinese)
(刘大有,王 飞,卢奕南,等.基于遗传算法的Bayesian网络结构学习研究.计算机研究与发展, 2001, 38(8): 916922)
[13] Pernkopf E F, Bouchaffra D. GeneticBased EM Algorithm for Learning Gaussian Mixture Models. IEEE Trans on Pattern Analysis and Machine Intelligence, 2005, 27(8):13441348
[14] Lampinen J, Vehtari A. Bayesian Approach for Neural Networks-Review and Case Studies. Neural Networks, 2001, 14(3): 257274
[15] Ng S K, McLachlan G J. Using the EM Algorithm to Train Neural Networks: Misconceptions and a New Algorithm for Multiclass Classification. IEEE Trans on Neural Networks, 15(3): 738749
[16] Murphy K. The Bayes Net Toolbox for Matlab [DB/OL]. [20010101]. http://bnt.sourceforge.net
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
国家自然科学基金(No.60575023)、教育部博士点基金(No.20050359012)资助项目
{{custom_fund}}
PDF(454 KB)
