[1]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):135-142.[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(11):135-142.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 020]
点击复制

基于不确定性加权混合训练的无源域自适应()
分享到：

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

卷:
33
期数:
2023年11期
页码:
135-142
栏目:
人工智能
出版日期:
2023-11-10

文章信息/Info

Title:
Uncertainty-guided Weighted Hybrid Training Based Source-free Domain Adaptation
文章编号:
1673-629X(2023)11-0135-08
作者:
关新宇; 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):
GUAN Xin-yu; 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; 无源域数据; 不确定性; 伪标签; 混合训练
Keywords:
domain adaptation; source-free data; uncertainty; pseudo label; hybrid training
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 11. 020
摘要:
针对无源域数据的域自适应问题中源模型中的源知识得不到充分利用以及目标数据内在结构信息会被忽略的问题,提出了一种基于不确定性指导的加权混合训练的无源域数据的域自适应框架。 首先在通道级对源模型预测的不确定性进行量化,将目标数据同时输入到源特征提取器和目标特征提取器,在通道级计算目标特征与源模型之间的不确定性距离,衡量源知识对目标模型的可迁移的不确定性,并利用它来指导对目标数据的适应过程。 同时考虑到目标数据的内在结构信息,从源知识和目标数据两个方面考虑伪标签的置信度,以降低噪声伪标签的影响。 最后对目标样本进行混合训练,并根据置信度得分对目标样本进行加权,以提高目标模型的鲁棒性。 在 Office-31,Office-Home 和 VisDA-C 3 个域自适应基准数据集上分别取得了 90. 5% ,72. 8% ,85. 9% 的分类效果,大量的对比实验及消融分析都证明了所提出的基于不确定性指导加权混合训练的框架具备良好的迁移能力。
Abstract:
Aiming at the problem that the knowledge in the source model cannot be fully utilized and the intrinsic structure information ofthe target data is ignored in the source - free domain adaptation, which results in the performance degradation of the target model, anetwork based on uncertainty - guided weighted hybrid training is proposed. Firstly, the uncertainty of source model prediction isquantified at the channel level,the target data is fed into the source feature extractor and the target feature extractor at the same time,tocalculate the uncertainty distance between the target feature and the source model at the channel level,and the transferability uncertainty ofsource knowledge to the target model is measured,　which is used to guide the process of adapting target data. Considering the intrinsicstructure of target data,the joint confidence score of the pseudo-label is considered from both source knowledge and target data to reducethe influence of the noise label. Finally,the target samples are conduced hybrid training,and weighted according to the confidence scoreto improve the robustness of the target model. The classification results of 90. 5% ,72. 8% , and 85. 9% on three domain adaptivebenchmark datasets of Office-31,Office-Home,and VisDA-C,show that the proposed framework for source - free domain adaptationbased on uncertainty-guided weighted hybrid training has a good transfer ability.

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(11):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(11):34.[doi:10. 3969 / j. issn. 1673-629X. 2022. 06. 006]
[3]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.[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(11):193.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 029]

更新日期/Last Update: 2023-11-10