摘要
提出一种基于支持向量机的渐近式半监督式学习算法,它以少量的有标记数据来训练初始学习器,通过选择性取样规则和核参数来调节无标记样本的选择范围和控制学习器决策面的动态调节方向,并通过删除非支持向量来降低学习代价。仿真实验表明,只要能够选择适当的选择性取样的阈值和核参数,这种学习算法就能够以较少的学习代价获得较好的学习效果。
A Semi-supervised Learning algorithm based on Support Vector Machine and by gradual approach has been put forward,which trains early learner by a spot of labeled-data,adjusts the scope selected unlabeled-data and controls the direction adjusting the decision-function of the learner by means of a rule selective-sampling and kernel parameter and reduces learning cost by deleting non-support vector.Simulative experiments have shown that the algorithm may get good learning effect at less learning cost if only opportune threshold for selective sampling and kernel parameter are selected.
出处
《计算机工程与应用》
CSCD
北大核心
2006年第25期19-22,共4页
Computer Engineering and Applications
基金
国家自然科学基金重点资助项目(编号:60234030)
关键词
支持向量机
半监督式学习
算法
Support Vector Machine,semi-supervised learning,algorithm