[1]林 磊,孙 涵.基于自纠错伪标签的无监督域自适应[J].计算机技术与发展,2023,33(01):193-199.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 029]
　LIN Lei,SUN Han.Self-correcting Pseudo Label for Unsupervised Domain Adaptation[J].,2023,33(01):193-199.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 029]
点击复制

基于自纠错伪标签的无监督域自适应()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
33
期数:
2023年01期
页码:
193-199
栏目:
新型计算应用系统
出版日期:
2023-01-10

文章信息/Info

Title:
Self-correcting Pseudo Label for Unsupervised Domain Adaptation
文章编号:
1673-629X(2023)01-0193-07
作者:
林 磊; 孙 涵
南京航空航天大学 计算机科学与技术学院 / 人工智能学院,江苏 南京 211106
Author(s):
LIN Lei; SUN Han
School of Computer Science and Technology / Artificial Intelligence,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China
关键词:
域自适应; 伪标签; 学生教师模型; 数据降维; 子空间变换
Keywords:
domain adaptation; pseudo-label; student-teacher model; dimensionality reduction; subspace transformation
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 01. 029
摘要:
当前不少域自适应方法采用为目标域生成伪标签的思想,但是由于源域数据的不充足以及源域和目标域之间的域差异,模型生成的伪标签往往含有大量错误信息,这些噪声会导致模型出现很严重的负迁移现象。 针对伪标签可能带有噪声的情况,一个具有自纠错能力的双网络伪标签模型被提出,该模型拥有一个学生网络和一个教师网络。 教师网络利用源域标注数据进行数据降维和子空间变换为目标域无标注数据生成伪标签,该伪标签基于源域类别信息与目标域结构信息。 学生网络利用伪标签进行学习,并且将学习结果反馈给教师网络,教师网络根据反馈更新伪标签。 通过这种循环自纠错的过程,目标域的伪标签会更加贴合目标域的真实空间,最终达到迁移的效果。 所提方法在多个数据集下表现优异,实验结果证明了其有效性。
Abstract:
Many current domain adaptation methods use the idea of generating pseudo - labels for the target domain, but due to theinsufficient data in the source domain and the domain differences between the source and target domains,the pseudo-labels generated bythe model often contain a lot of noises,which can lead to a seriously negative migration phenomenon. To address 　the situation that pseudo-labels may be noisy,a two-network pseudo-label model with self-correcting capability is proposed,which has a student network and ateacher network. The teacher network uses the source domain labeled data to generate pseudo - labels by performing dimensionalityreduction and subspace transformation to the target domain unlabeled data,which is based on the source domain category information andthe target domain structure information. The student network uses the pseudo-labels to learn and feeds the learning results to the teacher　network,which updates the pseudo-labels based on the feedback. Through this circular self-correcting process,the pseudo-labels will bemore closely matched with the real label of the target domain after each update,and finally achieve the effect of migration. The proposedmethod achieves excellent results under the experiments of several datasets and proves its effectiveness.

相似文献/References:

[1]郑雄风,汪云云.基于参数字典的多源域自适应学习算法[J].计算机技术与发展,2020,30(11):7.[doi:10. 3969 / j. issn. 1673-629X. 2020. 11. 002]
　ZHENG Xiong-feng,WANG Yun-yun.Multi-source Domain Adaptation Learning Based on Parameter Dictionary[J].,2020,30(01):7.[doi:10. 3969 / j. issn. 1673-629X. 2020. 11. 002]
[2]刘佳美,孙 涵,林 磊.基于伪标签的可防御稳定网络[J].计算机技术与发展,2022,32(06):34.[doi:10. 3969 / j. issn. 1673-629X. 2022. 06. 006]
　LIU Jia-mei,SUN Han,LIN Lei.Pseudo-label Based Defensible Stable Network[J].,2022,32(01):34.[doi:10. 3969 / j. issn. 1673-629X. 2022. 06. 006]
[3]关新宇,孙 涵.基于不确定性加权混合训练的无源域自适应[J].计算机技术与发展,2023,33(11):135.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 020]
　GUAN Xin-yu,SUN Han.Uncertainty-guided Weighted Hybrid Training Based Source-free Domain Adaptation[J].,2023,33(01):135.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 020]

更新日期/Last Update: 2023-01-10