【摘要】 提出了一种贝叶斯网络增量学习方法———ILBN.ILBN将EM算法和遗传算法引入到了贝叶斯网络的增量学习过程中,用EM算法从不完整数据计算充分统计量的期望,用遗传算法进化贝叶斯网络的结构,在一定程度上缓解了确定性搜索算法的局部极值问题.通过定义新变异算子和扩展传统的交叉算子,ILBN能够增量学习包含隐变量的贝叶斯网络结构.最后,ILBN改进了Friedman等人的增量学习过程.实验结果表明,ILBN和Friedman等人的增量学习方法存储开销相当,但在相同条件下,学到的网络更精确;实验结果也证实了存在不完整数据和隐变量时,ILBN的增量学习能力.更多还原

【Abstract】 An incremental approach to learning Bayesian networks based on genetic algorithm,namely ILBN,is put forward in this paper.ILBN introduces the EM algorithm and genetic algorithm into the incremental process of Bayesian network learning,calculates the expectation of the sufficient statistics with incomplete data using EM algorithm and evolves network structures using genetic algorithm,that could avoid getting into local maxima to some extent.Furthermore,by defining a new mutation operator and extending the traditional crossover operator,ILBN could incrementally learn and evolve Bayesian networks containing hidden variables.Finally,ILBN improves the incremental process by Friedman et al.The experimental results show that,in terms of storage cost,ILBN is comparable with the method by Friedman et al,while under the same experimental conditions,ILBN could learn more accurate networks than that of Friedman et al.The experimental results also verify the validity of ILBN in presence of incomplete data and hidden variables.更多还原