摘要
BK算法是动态贝叶斯网络(DBNs)的一种主要近似推理方法,但对网络的人工分割会引入较大误差。首先通过将决策结点转换成随机结点,给出基于DBNs的Robocup协作问题的一种建模方法;然后,给出一种引入分割团的新BK算法,以减小网络分割产生的误差,并对Robocup中的两个球员配合射门问题进行推理。引入分割团的BK算法和1.5片联合树推理算法的比较实验结果表明,引入分割团使BK算法在精度损失较小的情况下,时间性能有显著提高。
Boyen-Koller（BK） algorithm is the primary algorithm of approximate inference for DBNs,however,more eror is introduced by the artificial division of the network. A new modeling means of cooperation problem in Robocup was given in this paper based on DBNs by converting the decision-making node into random node. Then a new BK algorithm introducing conditionally independent separators was presented for decreasing the error of inference, and was used to solve the problem of two Agent coordination shooting in Robocup. The results of inference experiment implemented on BK algorithm introducing separators and 1.5-slice junction tree algorithm show that time performance has improved significantly in the case that the accuracy loss is relatively low by introducing conditionally independent separators in BK algorithm.
出处
《计算机科学》
CSCD
北大核心
2009年第6期214-216,234,共4页
Computer Science
基金
国家自然科学基金(60705015)
安徽省自然科学基金(070412064)
安徽省教育厅自然科学重点项目(KJ2009A020Z)资助
关键词
动态贝叶斯网络
近似推理
BK算法
1.5片联合树
Dynamic Bayesian networks, Approximate inference, BK algorithm, 1.5-slice junction tree
