Disturbed Augmentation Invariance for Unsupervised Visual Representation Learning

被引:3
|
作者
Cheng, Haoyang [1 ]
Li, Hongliang [1 ]
Wu, Qingbo [1 ]
Qiu, Heqian [1 ]
Zhang, Xiaoliang [1 ]
Meng, Fanman [1 ]
Zhao, Taijin [1 ]
机构
[1] Univ Elect Sci & Technol China UESTC, Sch Informat & Commun Engn, Chengdu 611731, Peoples R China
基金
中国国家自然科学基金;
关键词
Unsupervised learning; self-supervised learning; representation learning; contrastive learning; convolutional neural network;
D O I
10.1109/TCSVT.2023.3272741
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
Contrastive learning has gained great prominence recently, which achieves excellent performance by simple augmentation invariance. However, the simple contrastive pairs suffer from lacking of diversity due to the mechanical augmentation strategies. In this paper, we propose Disturbed Augmentation Invariance (DAI for abbreviation), which constructs disturbed contrastive pairs by generating appropriate disturbed views for each augmented view in the feature space to increase the diversity. In practice, we establish a multivariate normal distribution for each augmented view, whose mean is corresponding augmented view and covariance matrix is estimated from its nearest neighbors in the dataset. Then we sample random vectors from this distribution as the disturbed views to construct disturbed contrastive pairs. In order to avoid extra computational cost with the increase of disturbed contrastive pairs, we utilize an upper bound of the trivial disturbed augmentation invariance loss to construct the DAI loss. In addition, we propose Bottleneck version of Disturbed Augmentation Invariance (BDAI for abbreviation) inspired by the Information Bottleneck principle, which further refines the extracted information and learns a compact representation by additionally increasing the variance of the original contrastive pair. In order to make BDAI work effectively, we design a statistical strategy to control the balance between the amount of the information shared by all disturbed contrastive pairs and the compactness of the representation. Our approach gets a consistent improvement over the popular contrastive learning methods on a variety of downstream tasks, e.g. image classification, object detection and instance segmentation.
引用
收藏
页码:6924 / 6938
页数:15
相关论文
共 50 条
  • [41] Delving into Inter-Image Invariance for Unsupervised Visual Representations
    Jiahao Xie
    Xiaohang Zhan
    Ziwei Liu
    Yew-Soon Ong
    Chen Change Loy
    International Journal of Computer Vision, 2022, 130 : 2994 - 3013
  • [42] Delving into Inter-Image Invariance for Unsupervised Visual Representations
    Xie, Jiahao
    Zhan, Xiaohang
    Liu, Ziwei
    Ong, Yew-Soon
    Loy, Chen Change
    INTERNATIONAL JOURNAL OF COMPUTER VISION, 2022, 130 (12) : 2994 - 3013
  • [43] Representation learning for unsupervised domain adaptation
    Xu Y.
    Yan H.
    Harbin Gongye Daxue Xuebao/Journal of Harbin Institute of Technology, 2021, 53 (02): : 40 - 46
  • [44] Simple Unsupervised Graph Representation Learning
    Mo, Yujie
    Peng, Liang
    Xu, Jie
    Shi, Xiaoshuang
    Zhu, Xiaofeng
    THIRTY-SIXTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE / THIRTY-FOURTH CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE / TWELVETH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2022, : 7797 - 7805
  • [45] Kalman contrastive unsupervised representation learning
    Yekta, Mohammad Mahdi Jahani
    SCIENTIFIC REPORTS, 2024, 14 (01):
  • [46] Automated Unsupervised Graph Representation Learning
    Hou, Zhenyu
    Cen, Yukuo
    Dong, Yuxiao
    Zhang, Jie
    Tang, Jie
    IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, 2023, 35 (03) : 2285 - 2298
  • [47] Unsupervised Representation Learning by Sorting Sequences
    Lee, Hsin-Ying
    Huang, Jia-Bin
    Singh, Maneesh
    Yang, Ming-Hsuan
    2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV), 2017, : 667 - 676
  • [48] Towards Federated Unsupervised Representation Learning
    van Berlo, Bram
    Saeed, Aaqib
    Ozcelebi, Tanir
    PROCEEDINGS OF THE THIRD ACM INTERNATIONAL WORKSHOP ON EDGE SYSTEMS, ANALYTICS AND NETWORKING (EDGESYS'20), 2020, : 31 - 36
  • [49] Consensus Clustering With Unsupervised Representation Learning
    Regatti, Jayanth Reddy
    Deshmukh, Aniket Anand
    Manavoglu, Eren
    Dogan, Urun
    2021 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), 2021,
  • [50] Unsupervised Representation Learning for Proteochemometric Modeling
    Kim, Paul T.
    Winter, Robin
    Clevert, Djork-Arne
    INTERNATIONAL JOURNAL OF MOLECULAR SCIENCES, 2021, 22 (23)