本体学习算法的两类LOO一致稳定性和广义界

Two classes of LOO uniform stability and generalization bounds of ontology learning algorithm

近年来,随着本体研究的深入,各类机器学习方法被尝试应用于本体相似度计算和本体映射获取。稳定性是本体学习算法的必要条件,它从本质上体现了算法的可用性,即要求本体学习算法的最优解不会受到本体样本的小幅度调整而发生大的变化。本文研究了删除一个本体样本点的条件下,对本体学习算法的期望误差与经验误差的差值产生的影响。分别在本体学习算法一致稳定和假设空间一致稳定两种不同的框架下,利用统计学习理论的技巧,得到对应广义界的上界估计。

Recently,with deepening ontology research,efforts have been made to apply various machine-learning methods to ontology similarity calculations and mapping acquisitions.Stability is a necessary condition for ontology-learning algorithms.It requires that the optimal solution of the algorithm does not undergo major changes due to small adjustments to the ontology samples;thus,it essentially reflects the usability of the algorithm.In this study,we investigated the effect of deleting an ontology sample point on the difference between the expected and empirical errors of the ontology-learning algorithm.In two settings of the ontology-learning algorithm—uniform stability and hypothetical space uniform stability—obtained using statistical learning theory,the corresponding upper bound estimates of generalized bounds are determined.