摘要
针对不完备弱标记数据的学习问题,提出基于粗糙集理论的半监督协同学习模型.首先定义不完备弱标记数据的半监督差别矩阵,提出充分、具有差异性的约简子空间获取算法.然后在有标记数据集上利用各约简子空间训练两个基分类器.在无标记数据上,各分类器基于协同学习的思想标注信度较大的无标记样本给另一分类器学习,迭代更新直至无可利用的无标记数据. UCI数据集实验对比分析表明,文中模型可以获得更好的不完备弱标记数据的分类学习性能,具有有效性.
To address the problem of learning from supervised co-training model based on rough set theory is incomplete weakly labeled data, a semi- proposed. A semi-supervised discernibility matrix is firstly defined and then used to generate two sufficient and diverse semi-supervised reducts. The base classifiers are trained on the labeled data with two reducts, and then the two classifiers are learned from each other on the unlabeled data by labeling the confident unlabeled examples to its concomitant until no eligible unlabeled example is available. Experimental results on selected UCI datasets show that the proposed model achieves better performance on incomplete weakly labeled data compared with other models, and the effectiveness of the proposed model is verified.
作者
高灿
周杰
高天宇
赖志辉
GAO Can1;2;ZHOU Jie1;2;GAO Tianyu3;LAI Zhihui1;2
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2018年第10期950-957,共8页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金项目(No.61573248,61672358,61703283,61773328)
中国博士后基金项目(No.2016M590812,2017M612736,2017T100645)
广东省自然科学基金项目(No.2017A030310067)
关键词
粗糙集
差别矩阵
半监督约简
粗糙协同训练
不完备数据
Rough Sets
Discernibility Matrix
Semi-supervised Reduction
Rough Co-training
Incomplete Data
