基于对比学习的标签带噪图像分类
作者:
基金项目:

国家自然科学基金(62201282);江苏省自然科学基金(BK20231456)


Label Noisy Image Classification Based on Contrastive Learning
Author:
  • 摘要
  • | |
  • 访问统计
  • |
  • 参考文献 [36]
  • |
  • 相似文献 [20]
  • |
  • 引证文献
  • | |
  • 文章评论
    摘要:

    标签噪声会极大地降低深度网络模型的性能. 针对这一问题, 本文提出了一种基于对比学习的标签带噪图像分类方法. 该方法包括自适应阈值、对比学习模块和基于类原型的标签去噪模块. 首先采用对比学习最大化一幅图像的两个增强视图的相似度来提取图像鲁棒特征; 接下来通过一种新颖的自适应阈值过滤训练样本, 在模型训练过程中根据各个类别的学习情况动态调整阈值; 然后创新性地引入基于类原型的标签去噪模块, 通过计算样本特征向量与原型向量的相似度更新伪标签, 从而避免标签中噪声的影响; 在公开数据集CIFAR-10、CIFAR-100和真实数据集ANIMAL10上进行对比实验, 实验结果表明, 在人工合成噪声的条件下, 本文方法实验结果均高于常规方法, 通过计算图像鲁棒的特征向量与各个原型向量的相似度更新伪标签的方式, 降低了噪声标签的负面影响, 在一定程度上提高模型的抗噪声能力, 验证了该模型的有效性.

    Abstract:

    Label noise can greatly reduce the performance of deep network models. To address this problem, this study proposes a contrastive learning-based label noisy image classification method. The method includes an adaptive threshold, contrastive learning module, and class prototype-based label denoising module. Firstly, the robust features of the image are extracted by maximizing the similarity between two augmented views of the same image using contrastive learning. Then, a novel adaptive threshold filtering training sample is used to dynamically adjust the threshold based on the learning status of each class during model training. Finally, a class prototype-based label denoising module is introduced to update pseudo-labels by calculating the similarity between sample feature vectors and prototype vectors, thus avoiding the influence of label noise. Comparative experiments are conducted on the publicly available datasets CIFAR-10 and CIFAR-100 and the real dataset ANIMAL10. The experimental results show that under the condition of artificially synthesized noise, the proposed method outperforms conventional methods. By updating pseudo-labels based on the similarity between the robust feature vector of the image and various prototype vectors, the negative impact of noisy labels is reduced, and the anti-noise ability of the model is improved to certain extent, verifying the effectiveness of the proposed model.

    参考文献
    [1] 王晓莉, 薛丽. 标签噪声学习算法综述. 计算机系统应用, 2021, 30(1): 10–18.
    [2] 张增辉, 姜高霞, 王文剑. 基于动态概率抽样的标签噪声过滤方法. 计算机应用, 2021, 41(12): 3485–3491.
    [3] 陈庆强, 王文剑, 姜高霞. 基于数据分布的标签噪声过滤. 清华大学学报(自然科学版), 2019, 59(4): 262–269.
    [4] 余孟池, 牟甲鹏, 蔡剑等. 噪声标签重标注方法. 计算机科学, 2020, 47(6): 79–84.
    [5] Cho JH, Hariharan B. On the efficacy of knowledge distillation. Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019. 4794–4802.
    [6] Xie QZ, Luong MT, Hovy E, et al. Self-training with noisy student improves ImageNet classification. Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020. 10687–10698.
    [7] Yang CL, Xie LX, Qiao SY, et al. Training deep neural networks in generations: A more tolerant teacher educates better students. Proceedings of the 33rd AAAI Conference on Artificial Intelligence. Honolulu: AAAI Press, 2019. 5628–5635.
    [8] Yim J, Joo D, Bae J, et al. A gift from knowledge distillation: Fast optimization, network minimization and transfer learning. Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017. 4133–4141.
    [9] Patrini G, Rozza A, Krishna Menon A, et al. Making deep neural networks robust to label noise: A loss correction approach. Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017. 1944–1952.
    [10] Gui XJ, Wang W, Tian ZH. Towards understanding deep learning from noisy labels with small-loss criterion. Proceedings of the 30th International Joint Conference on Artificial Intelligence. Montreal: IJCAI.org, 2021. 2469–2475.
    [11] Shu J, Xie Q, Yi LX, et al. Meta-weight-net: Learning an explicit mapping for sample weighting. Proceedings of the 33rd International Conference on Neural Information Processing Systems. Vancouver, 2019. 172.
    [12] Li M, Zhou ZH. SETRED: Self-training with editing. Proceedings of the 9th Pacific-Asia Conference on Knowledge Discovery and Data Mining. Hanoi: Springer, 2005. 611–621.
    [13] Jiang L, Zhou ZY, Leung T, et al. MentorNet: Learning data-driven curriculum for very deep neural networks on corrupted labels. Proceedings of the 35th International conference on machine learning. Stockholm: PMLR, 2018. 2309–2318.
    [14] Blum A, Mitchell TM. Combining labeled and unlabeled data with co-training. Proceedings of the 11th Annual Conference on Computational Learning Theory. Madison: ACM, 1998. 92–100.
    [15] Li JN, Socher R, Hoi SCH. DivideMix: Learning with noisy labels as semi-supervised learning. Proceedings of the 8th International Conference on Learning Representations. Addis Ababa: OpenReview.net, 2020. 1–14.
    [16] Sukhbaatar S, Fergus R. Learning from noisy labels with deep neural networks. Proceedings of the 3rd International Conference on Learning Representations. San Diego, 2015. 4–14.
    [17] Goldberger J, Ben-Reuven E. Training deep neural-networks using a noise adaptation layer. Proceedings of the 5th International Conference on Learning Representations. Toulon: OpenReview.net, 2017. 1–9.
    [18] Han B, Yao JC, Niu G, et al. Masking: A new perspective of noisy supervision. Proceedings of the 32nd International Conference on Neural Information Processing Systems. Montréal: Curran Associates Inc., 2018. 5841–5851.
    [19] Li JN, Wong YK, Zhao Q, et al. Learning to learn from noisy labeled data. Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019. 5051–5059.
    [20] Jindal I, Nokleby M, Chen XW. Learning deep networks from noisy labels with dropout regularization. Proceedings of the 16th IEEE International Conference on Data Mining. Barcelona: IEEE, 2016. 967–972.
    [21] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: A simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research, 2014, 15(1): 1929–1958.
    [22] Ciortan M, Dupuis R, Peel T. A framework using contrastive learning for classification with noisy labels. Data, 2021, 6(6): 61.
    [23] Ghosh A, Lan A. Contrastive learning improves model robustness under label noise. Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville: IEEE, 2021. 2703–2708.
    [24] Li JN, Xiong CM, Hoi SCH. Learning from noisy data with robust representation learning. Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2021. 9485–9494.
    [25] Li JN, Xiong CM, Hoi SCH. MoPro: Webly supervised learning with momentum prototypes. Proceedings of the 9th International Conference on Learning Representations. OpenReview.net, 2021. 1–15.
    [26] Wang HB, Xiao RX, Li YX, et al. PiCO: Contrastive label disambiguation for partial label learning. Proceedings of the 10th International Conference on Learning Representations. OpenReview.net, 2022. 1–18.
    [27] Yao YZ, Sun ZR, Zhang CY, et al. Jo-SRC: A contrastive approach for combating noisy labels. Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville: IEEE, 2021. 5192–5201.
    [28] Zhang X, Liu ZX, Xiao KW, et al. CoDiM: Learning with noisy labels via contrastive semi-supervised learning. arXiv:2111.11652, 2021.
    [29] Chen XL, He KM. Exploring simple siamese representation learning. Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville: IEEE, 2021. 15750–15758.
    [30] He KM, Fan HQ, Wu YX, et al. Momentum contrast for unsupervised visual representation learning. Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020. 9729–9738.
    [31] Chen T, Kornblith S, Norouzi M, et al. A simple framework for contrastive learning of visual representations. Proceedings of the 37th International Conference on Machine Learning. PMLR, 2020. 149.
    [32] 田浩江, 路娜, 崔二洋. 基于改进对比学习的道路裂缝图像分类. 计算机系统应用, 2023, 32(2): 310–315.
    [33] Wen HW, Cui JY, Hang HY, et al. Leveraged weighted loss for partial label learning. Proceedings of the 38th International Conference on Machine Learning. PMLR, 2021. 11091–11100.
    [34] Lv JQ, Xu M, Feng L, et al. Progressive identification of true labels for partial-label learning. Proceedings of the 37th International Conference on Machine Learning. PMLR, 2020. 603.
    [35] Feng L, Lv JQ, Han B, et al. Provably consistent partial-label learning. Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver: Curran Associates Inc., 2020. 919.
    [36] Feng L, Kaneko T, Han B, et al. Learning with multiple complementary labels. Proceedings of the 37th International Conference on Machine Learning. PMLR, 2020. 288.
    引证文献
    网友评论
    网友评论
    分享到微博
    发 布
引用本文

李俊哲,曹国.基于对比学习的标签带噪图像分类.计算机系统应用,2023,32(12):104-111

复制
分享
文章指标
  • 点击次数:641
  • 下载次数: 1628
  • HTML阅读次数: 1129
  • 引用次数: 0
历史
  • 收稿日期:2023-06-21
  • 最后修改日期:2023-07-19
  • 在线发布日期: 2023-10-20
文章二维码
您是第11200515位访问者
版权所有:中国科学院软件研究所 京ICP备05046678号-3
地址:北京海淀区中关村南四街4号 中科院软件园区 7号楼305房间,邮政编码:100190
电话:010-62661041 传真: Email:csa (a) iscas.ac.cn
技术支持:北京勤云科技发展有限公司

京公网安备 11040202500063号