摘要
针对半监督学习中基于线性嵌入的回归正则项难以捕获数据流形结构的问题,提出基于L21范数和回归模型的半监督聚类算法。一方面充分利用监督信息,指导初始相似矩阵的构造,并利用L21正则项构造标签矩阵F的弹性嵌入回归模型;另一方面借助L21范数的鲁棒性学习合理的相似矩阵,从而改善聚类效果。通过实验表明,所提出的聚类算法在人工数据集和真实数据集上的聚类结果较其他聚类算法更加有效。
Linear embedding based regression model was difficult to capture data manifold structure in semi-supervised learning.To solve the problem a semi-supervised clustering algorithm based on L21 norm and regression regular term was proposed.On one hand,the supervision information was fully used to guide the construction of initial similarity matrix,and the elastic embedded regression model of label matrix F was constructed by using L21 regular terms.On the other hand,by virtue of the robustness of L21 norm,a reasonable similarity matrix was learned to improve the clustering effect.Experiments showed that the proposed clustering algorithm was more effective than other clustering algorithms on artificial datasets and real datasets.
作者
朱恒东
马盈仓
张要
张宁
ZHU Hengdong;MA Yingcang;ZHANG Yao;ZHANG Ning(School of Science, Xi′an Polytechnic University,Xi′an 710048, China)
出处
《郑州大学学报(理学版)》
CAS
北大核心
2020年第4期67-74,共8页
Journal of Zhengzhou University:Natural Science Edition
基金
国家自然科学基金项目(61976130)
陕西省重点研发计划项目(2018KW-021)
陕西省自然科学基金项目(2020JQ-923)。
关键词
L21范数
半监督学习
监督信息
回归
L21 norm
semi-supervised learning
label information
regression