Implicit Class-Conditioned Domain Alignment for Unsupervised Domain Adaptation

Xiang Jiang
Xiang Jiang
Qicheng Lao
Qicheng Lao
Mohammad Havaei
Mohammad Havaei

ICML, pp. 4816-4827, 2020.

被引用0|浏览21
EI
微博一下
We present an approach for unsupervised domain adaptation---with a strong focus on practical considerations of within-domain class imbalance and between-domain class distribution shift---from a class-conditioned domain alignment perspective

摘要

We present an approach for unsupervised domain adaptation---with a strong focus on practical considerations of within-domain class imbalance and between-domain class distribution shift---from a class-conditioned domain alignment perspective. Current methods for class-conditioned domain alignment aim to explicitly minimize a loss functio...更多

代码

数据

0
ZH
下载 PDF 全文
引用
微博一下
简介
  • Supervised learning aims to extract statistical patterns from data by learning to approximate the conditional density p(y|x).
  • Unsupervised Domain Adaptation (UDA) aims to address domain shift with access to labeled data in the source domain and unlabeled data in the target domain (Pan & Yang, 2009).
  • In real-world applications, it is very common to have class imbalance within each domain and class distribution shift between different domains, necessitating the incorporation of label space distribution into adaptation.
  • While explicit alignment has the advantage of directly minimizing class-conditioned misalignment, it presents critical vulnerabilities to error accumulation (Chen et al, 2019a) and ill-calibrated probabilities (Guo et al, 2017) due to its dependence on explicit supervision from pseudo-labels provided by model predictions
重点内容
  • Supervised learning aims to extract statistical patterns from data by learning to approximate the conditional density p(y|x)
  • The contributions of this paper are as follows: (i) We propose implicit class-conditioned domain alignment to address the challenge of within-domain class imbalance and between-domain class distribution shift, which overcomes the limitation of error accumulation in explicit domain alignment; We provide theoretical analysis on the empirical domain divergence and reveal the existence of a shortcut function that interferes with domain-invariant learning, which is addressed by the proposed approach; We show that the proposed approach is orthogonal to the choice of domain adaptation algorithms and offers consistent improvements to two adversarial domain adaptation algorithms; We report state-of-the-art Unsupervised Domain Adaptation performance under extreme withindomain class imbalance and between-domain class distribution shift, and competitive results on standard Unsupervised Domain Adaptation tasks
  • We introduce an approach for unsupervised domain adaptation—with a strong focus on practical considerations of within-domain class imbalance and between-domain class distribution shift—from a class-conditioned domain alignment perspective
  • We show theoretically that the proposed implicit alignment provides a more reliable measure of empirical domain divergence which facilitates adversarial domaininvariant representation learning, that would otherwise be hampered by the class-misaligned domain divergence
  • We show that our proposed approach leads to superior Unsupervised Domain Adaptation performance under extreme within-domain class imbalance and between-domain class distribution shift, as well as competitive results on standard Unsupervised Domain Adaptation tasks
  • We show that the proposed approach is orthogonal to the choice of domain adaptation algorithms and offers consistent improvements to featurebased and classifier-based domain adaptation algorithms
方法
  • 2019b) for MDD-based domain discrepancy measure.
  • The authors' main explicit alignment baselines are COAL (Tan et al, 2019), PACET (Liang et al, 2019b) and MCS (Liang et al, 2019a), state-of-the-art explicit alignment methods based on domain discriminator discrepancy.
  • As the domain discrepancy measure is MDD, the authors re-implement various MDD-based explicit alignment for fair comparison.
  • The authors show in the supplementary materials that the method does not require more frequent pseudo-label updates
结果
  • The proposed method outperforms the state-of-the-art explicit alignment methods—PACET and MCS—across all domain pairs.
结论
  • Conclusion and Future Work

    The authors introduce an approach for unsupervised domain adaptation—with a strong focus on practical considerations of within-domain class imbalance and between-domain class distribution shift—from a class-conditioned domain alignment perspective.
  • The authors show theoretically that the proposed implicit alignment provides a more reliable measure of empirical domain divergence which facilitates adversarial domaininvariant representation learning, that would otherwise be hampered by the class-misaligned domain divergence.
  • The authors show that the proposed approach leads to superior UDA performance under extreme within-domain class imbalance and between-domain class distribution shift, as well as competitive results on standard UDA tasks.
  • The authors emphasize that the proposed method is robust to pseudo-label bias, simple to implement, has a unified training objective, and does not require additional parameter tuning.
  • It is important to analyze the probability calibration of different domain adaptation models and develop well-calibrated methods for more effective use of pseudo-labels
总结
  • Introduction:

    Supervised learning aims to extract statistical patterns from data by learning to approximate the conditional density p(y|x).
  • Unsupervised Domain Adaptation (UDA) aims to address domain shift with access to labeled data in the source domain and unlabeled data in the target domain (Pan & Yang, 2009).
  • In real-world applications, it is very common to have class imbalance within each domain and class distribution shift between different domains, necessitating the incorporation of label space distribution into adaptation.
  • While explicit alignment has the advantage of directly minimizing class-conditioned misalignment, it presents critical vulnerabilities to error accumulation (Chen et al, 2019a) and ill-calibrated probabilities (Guo et al, 2017) due to its dependence on explicit supervision from pseudo-labels provided by model predictions
  • Methods:

    2019b) for MDD-based domain discrepancy measure.
  • The authors' main explicit alignment baselines are COAL (Tan et al, 2019), PACET (Liang et al, 2019b) and MCS (Liang et al, 2019a), state-of-the-art explicit alignment methods based on domain discriminator discrepancy.
  • As the domain discrepancy measure is MDD, the authors re-implement various MDD-based explicit alignment for fair comparison.
  • The authors show in the supplementary materials that the method does not require more frequent pseudo-label updates
  • Results:

    The proposed method outperforms the state-of-the-art explicit alignment methods—PACET and MCS—across all domain pairs.
  • Conclusion:

    Conclusion and Future Work

    The authors introduce an approach for unsupervised domain adaptation—with a strong focus on practical considerations of within-domain class imbalance and between-domain class distribution shift—from a class-conditioned domain alignment perspective.
  • The authors show theoretically that the proposed implicit alignment provides a more reliable measure of empirical domain divergence which facilitates adversarial domaininvariant representation learning, that would otherwise be hampered by the class-misaligned domain divergence.
  • The authors show that the proposed approach leads to superior UDA performance under extreme within-domain class imbalance and between-domain class distribution shift, as well as competitive results on standard UDA tasks.
  • The authors emphasize that the proposed method is robust to pseudo-label bias, simple to implement, has a unified training objective, and does not require additional parameter tuning.
  • It is important to analyze the probability calibration of different domain adaptation models and develop well-calibrated methods for more effective use of pseudo-labels
表格
  • Table1: Per-class average accuracy on Office-Home dataset with RS-UT label shifts (ResNet-50)
  • Table2: Accuracy (%) on Office-31 (standard) for unsupervised domain adaptation (ResNet-50). We repeated each experiment 5 times with different random seeds and report the average and the standard error of the accuracy
  • Table3: Accuracy (%) on Office-Home (standard) for unsupervised domain adaptation (ResNet-50)
  • Table4: VisDA2017 target accuracy (ResNet-50)
  • Table5: The impact of different implicit alignment options, i.e., masking in the MDD estimation and sampling class-aligned minibatches, on Office-Home (RS-UT)
  • Table6: Per-class average accuracy (%) with mismatched prior where the source domain is balanced while the target domain is imbalanced
  • Table7: Per-class average accuracy (%) with mismatched prior where the source domain is imbalanced while the target domain is balanced
  • Table8: Per-class average accuracy (%) with mismatched prior where both domains are imbalanced
  • Table9: Evaluation on Office-Home (%) with ResNet-50
  • Table10: The impact of different implicit alignment options, i.e., masking the classifier-based domain discrepancy measure and sampling examples from the source and target domains, on Ar→Cl and Cl→Pr, Office-Home (standard)
  • Table11: The impact of pseudo-label update frequency on Ar→Cl, Office-Home (standard)
  • Table12: Impact of batch size on target domain accuracy (%), Ar→Cl, Office-Home (standard). The MDD results are based on our re-implementation
Download tables as Excel
相关工作
  • We review related work on unsupervised domain adaptation and discuss their relations with our proposed method.

    Instance-based importance-weighting (Chawla et al, 2002; Kouw & Loog, 2019) aims to minimize the target error directly from the source domain data, weighted at the example level or class level. Unlike our approach, importance-weighting only uses the source data to train the classifier without learning domain invariant representations.

    Feature-based distribution adaptation is the prevailing approach to domain adaptation that aims to minimize the distribution discrepancy between the source and target domains. The domain difference can be measured in various ways, such as Maximum Mean Discrepancy (MMD) (Borgwardt et al, 2006), which is further minimized to achieve domain invariance. The minimization of such discrepancy can be carried out by directly minimizing the distance (Tzeng et al, 2014) or with the help of adversarial learning (Ganin et al, 2016).
基金
  • Xiang Jiang acknowledges the support of NVIDIA Corporation with the donation of the Titan X GPU used for this research
引用论文
  • Ben-David, S., Blitzer, J., Crammer, K., Kulesza, A., Pereira, F., and Vaughan, J. W. A theory of learning from different domains. Machine learning, 79(1-2):151–175, 2010.
    Google ScholarLocate open access versionFindings
  • Blum, A. and Mitchell, T. Combining labeled and unlabeled data with co-training. In Proceedings of the eleventh annual conference on Computational learning theory, pp. 92–100. ACM, 1998.
    Google ScholarLocate open access versionFindings
  • Borgwardt, K. M., Gretton, A., Rasch, M. J., Kriegel, H.-P., Scholkopf, B., and Smola, A. J. Integrating structured biological data by kernel maximum mean discrepancy. Bioinformatics, 22(14):e49–e57, 2006.
    Google ScholarLocate open access versionFindings
  • Cao, Z., Ma, L., Long, M., and Wang, J. Partial adversarial domain adaptation. In Proceedings of the European Conference on Computer Vision (ECCV), pp. 135–150, 2018.
    Google ScholarLocate open access versionFindings
  • Chawla, N. V., Bowyer, K. W., Hall, L. O., and Kegelmeyer, W. P. Smote: synthetic minority over-sampling technique. Journal of artificial intelligence research, 16:321–357, 2002.
    Google ScholarLocate open access versionFindings
  • Chen, C., Xie, W., Huang, W., Rong, Y., Ding, X., Huang, Y., Xu, T., and Huang, J. Progressive feature alignment for unsupervised domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 627–636, 2019a.
    Google ScholarLocate open access versionFindings
  • Chen, M., Weinberger, K. Q., and Blitzer, J. Co-training for domain adaptation. In Advances in neural information processing systems, pp. 2456–2464, 2011.
    Google ScholarLocate open access versionFindings
  • Chen, M., Xue, H., and Cai, D. Domain adaptation for semantic segmentation with maximum squares loss. In Proceedings of the IEEE International Conference on Computer Vision, pp. 2090–2099, 2019b.
    Google ScholarLocate open access versionFindings
  • Chen, X., Wang, S., Long, M., and Wang, J. Transferability vs. discriminability: Batch spectral penalization for adversarial domain adaptation. In International Conference on Machine Learning, pp. 1081–1090, 2019c.
    Google ScholarLocate open access versionFindings
  • Cicek, S. and Soatto, S. Unsupervised domain adaptation via regularized conditional alignment. In The IEEE International Conference on Computer Vision (ICCV), October 2019.
    Google ScholarLocate open access versionFindings
  • Deng, Z., Luo, Y., and Zhu, J. Cluster alignment with a teacher for unsupervised domain adaptation. In Proceedings of the IEEE International Conference on Computer Vision, pp. 9944–9953, 2019.
    Google ScholarLocate open access versionFindings
  • Ganin, Y., Ustinova, E., Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., Marchand, M., and Lempitsky, V. Domain-adversarial training of neural networks. The Journal of Machine Learning Research, 17(1):2096–2030, 2016.
    Google ScholarLocate open access versionFindings
  • Grandvalet, Y. and Bengio, Y. Semi-supervised learning by entropy minimization. In Advances in neural information processing systems, pp. 529–536, 2005.
    Google ScholarLocate open access versionFindings
  • Guo, C., Pleiss, G., Sun, Y., and Weinberger, K. Q. On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pp. 1321–1330. JMLR. org, 2017.
    Google ScholarLocate open access versionFindings
  • He, K., Zhang, X., Ren, S., and Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016.
    Google ScholarLocate open access versionFindings
  • Heckman, J. J. Sample selection bias as a specification error. Econometrica: Journal of the econometric society, pp. 153–161, 1979.
    Google ScholarLocate open access versionFindings
  • Kouw, W. M. and Loog, M. A review of domain adaptation without target labels. IEEE transactions on pattern analysis and machine intelligence, 2019.
    Google ScholarLocate open access versionFindings
  • Kumar, A., Sattigeri, P., Wadhawan, K., Karlinsky, L., Feris, R., Freeman, B., and Wornell, G. Co-regularized alignment for unsupervised domain adaptation. In Advances in Neural Information Processing Systems, pp. 9345–9356, 2018.
    Google ScholarLocate open access versionFindings
  • Liang, J., He, R., Sun, Z., and Tan, T. Distant supervised centroid shift: A simple and efficient approach to visual domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2975–2984, 2019a.
    Google ScholarLocate open access versionFindings
  • Liang, J., He, R., Sun, Z., and Tan, T. Exploring uncertainty in pseudo-label guided unsupervised domain adaptation. Pattern Recognition, 96:106996, 2019b.
    Google ScholarLocate open access versionFindings
  • Lifshitz, O. and Wolf, L. A sample selection approach for universal domain adaptation. arXiv preprint arXiv:2001.05071, 2020.
    Findings
  • Lipton, Z. C., Wang, Y.-X., and Smola, A. Detecting and correcting for label shift with black box predictors. arXiv preprint arXiv:1802.03916, 2018.
    Findings
  • Long, M., Wang, J., Ding, G., Sun, J., and Yu, P. S. Transfer feature learning with joint distribution adaptation. In Proceedings of the IEEE international conference on computer vision, pp. 2200–2207, 2013.
    Google ScholarLocate open access versionFindings
  • Long, M., Cao, Y., Wang, J., and Jordan, M. I. Learning transferable features with deep adaptation networks. In Proceedings of the 32nd International Conference on International Conference on Machine Learning-Volume 37, pp. 97–105. JMLR. org, 2015.
    Google ScholarLocate open access versionFindings
  • Long, M., Zhu, H., Wang, J., and Jordan, M. I. Deep transfer learning with joint adaptation networks. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pp. 2208–2217. JMLR. org, 2017.
    Google ScholarLocate open access versionFindings
  • Long, M., Cao, Z., Wang, J., and Jordan, M. I. Conditional adversarial domain adaptation. In Advances in Neural Information Processing Systems, pp. 1640–1650, 2018.
    Google ScholarLocate open access versionFindings
  • Luo, Z., Zou, Y., Hoffman, J., and Fei-Fei, L. F. Label efficient learning of transferable representations acrosss domains and tasks. In Advances in Neural Information Processing Systems, pp. 165–177, 2017.
    Google ScholarLocate open access versionFindings
  • Nigam, K. and Ghani, R. Analyzing the effectiveness and applicability of co-training. In Cikm, volume 5, pp. 3, 2000.
    Google ScholarLocate open access versionFindings
  • Pan, S. J. and Yang, Q. A survey on transfer learning. IEEE Transactions on knowledge and data engineering, 22(10): 1345–1359, 2009.
    Google ScholarLocate open access versionFindings
  • Pan, Y., Yao, T., Li, Y., Wang, Y., Ngo, C.-W., and Mei, T. Transferrable prototypical networks for unsupervised domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2239–2247, 2019.
    Google ScholarLocate open access versionFindings
  • Pei, Z., Cao, Z., Long, M., and Wang, J. Multi-adversarial domain adaptation. In Thirty-Second AAAI Conference on Artificial Intelligence, 2018.
    Google ScholarLocate open access versionFindings
  • Peng, X., Usman, B., Kaushik, N., Hoffman, J., Wang, D., and Saenko, K. Visda: The visual domain adaptation challenge. arXiv preprint arXiv:1710.06924, 2017.
    Findings
  • Pinheiro, P. O. Unsupervised domain adaptation with similarity learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 8004–8013, 2018.
    Google ScholarLocate open access versionFindings
  • Quionero-Candela, J., Sugiyama, M., Schwaighofer, A., and Lawrence, N. D. Dataset shift in machine learning. The MIT Press, 2009.
    Google ScholarFindings
  • Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. Imagenet large scale visual recognition challenge. International journal of computer vision, 115 (3):211–252, 2015.
    Google ScholarLocate open access versionFindings
  • Saito, K., Ushiku, Y., and Harada, T. Asymmetric tri-training for unsupervised domain adaptation. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pp. 2988–2997. JMLR. org, 2017.
    Google ScholarLocate open access versionFindings
  • Saito, K., Watanabe, K., Ushiku, Y., and Harada, T. Maximum classifier discrepancy for unsupervised domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3723–3732, 2018.
    Google ScholarLocate open access versionFindings
  • Sankaranarayanan, S., Balaji, Y., Castillo, C. D., and Chellappa, R. Generate to adapt: Aligning domains using generative adversarial networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 8503–8512, 2018.
    Google ScholarLocate open access versionFindings
  • Shimodaira, H. Improving predictive inference under covariate shift by weighting the log-likelihood function. Journal of statistical planning and inference, 90(2): 227–244, 2000.
    Google ScholarLocate open access versionFindings
  • Shu, R., Bui, H. H., Narui, H., and Ermon, S. A dirt-t approach to unsupervised domain adaptation. arXiv preprint arXiv:1802.08735, 2018.
    Findings
  • Snell, J., Swersky, K., and Zemel, R. Prototypical networks for few-shot learning. In Advances in Neural Information Processing Systems, pp. 4077–4087, 2017.
    Google ScholarLocate open access versionFindings
  • Tan, S., Peng, X., and Saenko, K. Generalized domain adaptation with covariate and label shift co-alignment. arXiv preprint arXiv:1910.10320, 2019.
    Findings
  • Tsai, Y.-H., Hung, W.-C., Schulter, S., Sohn, K., Yang, M.-H., and Chandraker, M. Learning to adapt structured output space for semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 7472–7481, 2018.
    Google ScholarLocate open access versionFindings
  • Tzeng, E., Hoffman, J., Zhang, N., Saenko, K., and Darrell, T. Deep domain confusion: Maximizing for domain invariance. arXiv preprint arXiv:1412.3474, 2014.
    Findings
  • Tzeng, E., Hoffman, J., Saenko, K., and Darrell, T. Adversarial discriminative domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 7167–7176, 2017.
    Google ScholarLocate open access versionFindings
  • Venkateswara, H., Eusebio, J., Chakraborty, S., and Panchanathan, S. Deep hashing network for unsupervised domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5018–5027, 2017.
    Google ScholarLocate open access versionFindings
  • Webb, G. I. and Ting, K. M. On the application of roc analysis to predict classification performance under varying class distributions. Machine learning, 58(1):25–32, 2005.
    Google ScholarLocate open access versionFindings
  • Wen, J., Zheng, N., Yuan, J., Gong, Z., and Chen, C. Bayesian uncertainty matching for unsupervised domain adaptation. arXiv preprint arXiv:1906.09693, 2019.
    Findings
  • Wu, Y., Winston, E., Kaushik, D., and Lipton, Z. Domain adaptation with asymmetrically-relaxed distribution alignment. arXiv preprint arXiv:1903.01689, 2019.
    Findings
  • Xie, S., Zheng, Z., Chen, L., and Chen, C. Learning semantic representations for unsupervised domain adaptation. In International Conference on Machine Learning, pp. 5419–5428, 2018.
    Google ScholarLocate open access versionFindings
  • Zhang, Q., Zhang, J., Liu, W., and Tao, D. Category anchor-guided unsupervised domain adaptation for semantic segmentation. In Advances in Neural Information Processing Systems, pp. 433–443, 2019a.
    Google ScholarLocate open access versionFindings
  • Zhang, W., Ouyang, W., Li, W., and Xu, D. Collaborative and adversarial network for unsupervised domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3801–3809, 2018.
    Google ScholarLocate open access versionFindings
  • Zhang, Y., Liu, T., Long, M., and Jordan, M. Bridging theory and algorithm for domain adaptation. In Chaudhuri, K. and Salakhutdinov, R. (eds.), Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pp. 7404–7413, Long Beach, California, USA, 09–15 Jun 2019b. PMLR.
    Google ScholarLocate open access versionFindings
  • Zou, Y., Yu, Z., Vijaya Kumar, B., and Wang, J. Unsupervised domain adaptation for semantic segmentation via class-balanced self-training. In Proceedings of the European Conference on Computer Vision (ECCV), pp. 289–305, 2018.
    Google ScholarLocate open access versionFindings
  • Zou, Y., Yu, Z., Liu, X., Kumar, B. V., and Wang, J. Confidence regularized self-training. In The IEEE International Conference on Computer Vision (ICCV), October 2019.
    Google ScholarLocate open access versionFindings
您的评分 :
0

 

标签
评论