DCL: Dipolar Confidence Learning for Source-Free Unsupervised Domain Adaptation

被引:0
|
作者
Tian, Qing [1 ,2 ,3 ]
Sun, Heyang [4 ,5 ]
Peng, Shun [6 ]
Zheng, Yuhui [7 ]
Wan, Jun [8 ]
Lei, Zhen [6 ]
机构
[1] Nanjing Univ Informat Sci & Technol, Sch Software, Wuxi 214000, Peoples R China
[2] Nanjing Univ Informat Sci & Technol, Wuxi Inst Technol, Wuxi 214000, Peoples R China
[3] Nanjing Univ, State Key Lab Novel Software Technol, Nanjing 210023, Peoples R China
[4] Nanjing Univ Informat Sci & Technol, Sch Software, Nanjing 210044, Peoples R China
[5] Nanjing Univ Aeronaut & Astronaut, Coll Comp Sci & Technol, Nanjing 210016, Peoples R China
[6] Nanjing Univ Informat Sci & Technol, Sch Software, Nanjing 210044, Peoples R China
[7] Qinghai Normal Univ, Coll Comp, Xining 810016, Peoples R China
[8] Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing 100190, Peoples R China
基金
中国国家自然科学基金;
关键词
Adaptation models; Data models; Task analysis; Predictive models; Generators; Feature extraction; Training; Source-free unsupervised domain adaptation (SFUDA); dipolar confidence learning (DCL); fuzzy mixup; rotation-based self-supervised learning;
D O I
10.1109/TCSVT.2023.3332353
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
Source-free unsupervised domain adaptation (SFUDA) aims to conduct prediction on the target domain by leveraging knowledge from the well-trained source model. Due to the absence of source data in the SFUDA setting, the existing methods mainly build the target classifier by fine-tuning the source model incorporated with empirical adaptation losses. Although these methods have achieved somewhat promising results, nearly all of them typically suffer from the closed-fitting dilemma that their models are dominantly affected by these easy-to-distinguish instances than those hard-to-distinguish ones, resulting from the absence of the labeled source data. To address aforementioned issues, we propose the Dipolar Confidence Learning (DCL) for SFUDA. Specifically, we conduct positive confidence learning on the samples with standard outputs to avoid overfitting of the model to these samples. In contrast, we perform negative confidence learning for the samples with abnormal outputs to optimize the complementary label, which forces the network to pay more attention to these confusing samples. Furthermore, to achieve more generalized domain alignment, both the confidence-based fuzzy mixup and rotation-based self-supervised learning are respectively constructed to boost the representation ability of the target model. Finally, extensive experiments are conducted to demonstrate the effectiveness and performance superiority of the proposed method.
引用
收藏
页码:4342 / 4353
页数:12
相关论文
共 50 条
  • [1] Confidence Score for Source-Free Unsupervised Domain Adaptation
    Lee, Jonghyun
    Jung, Dahuin
    Yim, Junho
    Yoon, Sungroh
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 162, 2022,
  • [2] Rethinking confidence scores for source-free unsupervised domain adaptation
    Tian Q.
    Sun C.
    Neural Computing and Applications, 2024, 36 (24) : 14951 - 14966
  • [3] Source-Free Unsupervised Domain Adaptation with Sample Transport Learning
    Qing Tian
    Chuang Ma
    Feng-Yuan Zhang
    Shun Peng
    Hui Xue
    Journal of Computer Science and Technology, 2021, 36 : 606 - 616
  • [4] Source-Free Unsupervised Domain Adaptation with Sample Transport Learning
    Tian, Qing
    Ma, Chuang
    Zhang, Feng-Yuan
    Peng, Shun
    Xue, Hui
    JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY, 2021, 36 (03) : 606 - 616
  • [5] Source-free unsupervised domain adaptation: A survey
    Fang, Yuqi
    Yap, Pew-Thian
    Lin, Weili
    Zhu, Hongtu
    Liu, Mingxia
    NEURAL NETWORKS, 2024, 174
  • [6] Continual Source-Free Unsupervised Domain Adaptation
    Ahmed, Waqar
    Morerio, Pietro
    Murino, Vittorio
    IMAGE ANALYSIS AND PROCESSING, ICIAP 2023, PT I, 2023, 14233 : 14 - 25
  • [7] Class Relationship Embedded Learning for Source-Free Unsupervised Domain Adaptation
    Zhang, Yixin
    Wang, Zilei
    He, Weinan
    2023 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR, 2023, : 7619 - 7629
  • [8] SOURCE-FREE UNSUPERVISED DOMAIN ADAPTATION VIA DENOISING MUTUAL LEARNING
    Hao, Zhang
    Liang, Tian
    2022 19TH INTERNATIONAL COMPUTER CONFERENCE ON WAVELET ACTIVE MEDIA TECHNOLOGY AND INFORMATION PROCESSING (ICCWAMTIP), 2022,
  • [9] Source-free unsupervised domain adaptation via bi-classifier confidence score weighting
    Tian, Qing
    Zhao, Mengna
    COMPUTERS & ELECTRICAL ENGINEERING, 2023, 111
  • [10] Source-free Unsupervised Domain Adaptation with Trusted Pseudo Samples
    Tian, Qing
    Peng, Shun
    Ma, Tinghuai
    ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY, 2023, 14 (02)