基于一致性正则化与熵最小化的半监督学习算法  

Semi-supervised Learning Algorithm Based on the Consistency Regularization and Entropy Minimization

作　　者：邵伟志 潘丽丽[1] 雷前慧 黄诗祺 马骏勇 SHAO Weizhi;PAN Lili;LEI Qianhui;HUANG Shiqi;MA Junyong(School of Computer and Information Engineering,Central South University of Forestry and Technology,Changsha 410004,China)

机构地区：[1]中南林业科技大学计算机与信息工程学院,湖南长沙410004

出　　处：《郑州大学学报（理学版）》2021年第3期79-84,共6页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学基金项目(61772561);湖南省重点研发计划项目(2018NK2012)。

摘　　要：在一致性正则化与熵最小化的基础上提出一种新的半监督学习算法Mean Mixup,集成数据的互补信息,然后使用熵最小化给未标记数据生成可靠的伪标签,在一致性正则化下进一步优化模型分类结果。在常用数据集SVHN和CIFAR10上对Mean Mixup算法进行了评估,实验结果表明,所提出的方法在分类准确率上优于一些已有的半监督学习算法。A new semi-supervised learning algorithm Mean Mixup was proposed,which was based on the consistency regularization and entropy minimization.Complementary information of data was fused,and then entropy minimization was used to generate reliable pseudo labels for unlabeled data.The classification results of the model were further optimized by consistency regularization.The Mean Mixup algorithm was evaluated on the commonly used datasets SVHN and CIFAR10.The experimental results showed that the proposed method was superior to some existing semi-supervised algorithms in classification accuracy.

关 键 词：半监督学习 熵最小化 一致性正则化 伪标签 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]