CAJ | 학술논문

提出在学习贝叶斯网下的一种行之有效的爬山算法,HCBest算法.该算法在学习网络结构形成环时,选择删除能提高打分值最多的边,直到没有环为止.实验证明,HCBest既可以作为一种独立的贝叶斯网学习方法,又可以作为其它复杂元启发方法的局部搜索算法.HCBest学出的网络在打分质量和结构上都比较好.在算法的简洁性和稳定性方面,HCBest的表现也令人满意.
제출재학습패협사망하적일충행지유효적파산산법,HCBest산법.해산법재학습망락결구형성배시,선택산제능제고타분치최다적변,직도몰유배위지.실험증명,HCBest기가이작위일충독립적패협사망학습방법,우가이작위기타복잡원계발방법적국부수색산법.HCBest학출적망락재타분질량화결구상도비교호.재산법적간길성화은정성방면,HCBest적표현야령인만의.
This paper presents an efficient hill-climbing approach for learning Bayesian Networks, HCBest algorithm. When a circle emerges in the process of evolving networks, the algorithm delete the edge that can best improve the score of the learned network until there is no circle. The experiments show that, HCBest not only can be used as an independent approach for learning Bayesian networks, but also can be applied in other complicated meta heuristic approach as a local search algorithm. HCBest can achieve good quality network structure with higher metric score. In the aspects of simplicity and robustness, the algorithm also performs well.