双向协同的去中心化多源域自适应

魏义康; 韩亚洪

doi:10.1631/FITEE.2200284

Your Location：

Home >

Browse articles >

双向协同的去中心化多源域自适应

常规文章 | Updated：2022-12-16

- 双向协同的去中心化多源域自适应
  Enhanced Publication
- Dual collaboration for decentralized multi-source domain adaptation
- 信息与电子工程前沿（英文） 2022年23卷第12期页码：1780-1794
- Affiliations：
  
  1.College of Intelligence and Computing, Tianjin University, Tianjin 300350, China
  2.Tianjin Key Lab of Machine Learning, Tianjin University, Tianjin 300350, China
- Author bio：
  
  E-mail: yikang@tju.edu.cn;
  ‡Corresponding author
- Funds：
  
  National Nature Science Foundation of China(61876130;61932009);Starry Night Science Fund of Zhejiang University Shanghai Institute for Advanced Study
- DOI：10.1631/FITEE.2200284
  中图分类号： TP181
- 纸质出版日期：2022-12-0 ，
  
  收稿日期：2022-06-29，
  
  录用日期：2022-09-22
- Accepted：
Scan QR Code
魏义康, 韩亚洪. 双向协同的去中心化多源域自适应[J]. 信息与电子工程前沿（英文）, 2022,23(12):1780-1794.

YIKANG WEI, YAHONG HAN. Dual collaboration for decentralized multi-source domain adaptation. [J]. Frontiers of information technology & electronic engineering, 2022, 23(12): 1780-1794.
魏义康, 韩亚洪. 双向协同的去中心化多源域自适应[J]. 信息与电子工程前沿（英文）, 2022,23(12):1780-1794. DOI： 10.1631/FITEE.2200284.

YIKANG WEI, YAHONG HAN. Dual collaboration for decentralized multi-source domain adaptation. [J]. Frontiers of information technology & electronic engineering, 2022, 23(12): 1780-1794. DOI： 10.1631/FITEE.2200284.

摘要

去中心化多源域自适应是指在数据去中心化场景下执行无监督多源域自适应。数据去中心化的挑战是源域与目标域在训练中缺乏跨域协同。对于无标签的目标域，目标域模型需要在源域模型的协助下迁移监督知识，而域差距会导致源域模型的适应性能有限。对于有标签的源域，源域模型在数据去中心化场景下倾向于过拟合本地数据，从而导致负迁移问题。对于以上挑战，提出双向协同的去中心化多源域自适应方法，通过其它域模型的协助进行局部源域模型与局部目标域模型的协同训练与聚合。对于目标域，我们在源域模型的协助下蒸馏监督知识，同时完全利用无标签目标域的数据来缓解域偏移问题。对于源域，我们在目标域模型的协助下正则化源域模型来避免负迁移问题。以上过程在去中心化的源域和目标域之间形成一种双向协同，以便在数据去中心化场景下提升域自适应性能。在标准多源域自适应数据集上的实验表明，我们的方法以较大优势优于现有的多源域自适应方法。

Abstract

The goal of decentralized multi-source domain adaptation is to conduct unsupervised multi-source domain adaptation in a data decentralization scenario. The challenge of data decentralization is that the source domains and target domain lack cross-domain collaboration during training. On the unlabeled target domain

the target model needs to transfer supervision knowledge with the collaboration of source models

while the domain gap will lead to limited adaptation performance from source models. On the labeled source domain

the source model tends to overfit its domain data in the data decentralization scenario

which leads to the negative transfer problem. For these challenges

we propose dual collaboration for decentralized multi-source domain adaptation by training and aggregating the local source models and local target model in collaboration with each other. On the target domain

we train the local target model by distilling supervision knowledge and fully using the unlabeled target domain data to alleviate the domain shift problem with the collaboration of local source models. On the source domain

we regularize the local source models in collaboration with the local target model to overcome the negative transfer problem. This forms a dual collaboration between the decentralized source domains and target domain

which improves the domain adaptation performance under the data decentralization scenario. Extensive experiments indicate that our method outperforms the state-of-the-art methods by a large margin on standard multi-source domain adaptation datasets.

关键词

多源域自适应数据去中心化域偏移负迁移

Keywords

Multi-source domain adaptationData decentralizationDomain shiftNegative transfer

references

Ahmed SM, Raychaudhuri DS, Paul S, et al., 2021. Unsupervised multi-source domain adaptation without access to source data. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.10103-10112. doi: 10.1109/CVPR46437.2021.00997http://doi.org/10.1109/CVPR46437.2021.00997

Cubuk ED, Zoph B, Shlens J, et al., 2020. Randaugment: practical automated data augmentation with a reduced search space. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition Workshops, p.702-703. doi: 10.1109/CVPRW50498.2020.00359http://doi.org/10.1109/CVPRW50498.2020.00359

Feng HZ, You ZY, Chen MH, et al., 2021. KD3A: unsupervised multi-source decentralized domain adaptation via knowledge distillation. Pro 38th Int Conf on Machine Learning, p.3274-3283. doi: 10.48550/arXiv.2011.09757http://doi.org/10.48550/arXiv.2011.09757

Ganin Y, Lempitsky V, 2015. Unsupervised domain adaptation by backpropagation. Pro 32nd Int Conf on Machine Learning, p.1180-1189.

Gao Z, Guo LM, Ren TW, et al., 2022. Pairwise two-stream ConvNets for cross-domain action recognition with small data. IEEE Trans Neur Netw Learn Syst, 33(3):1147-1161. doi: 10.1109/TNNLS.2020.3041018http://doi.org/10.1109/TNNLS.2020.3041018

Han YH, Wu AM, Zhu LC, et al., 2021. Visual commonsense reasoning with directional visual connections. Front Inform Technol Electron Eng, 22(5):625-637. doi: 10.1631/FITEE.2000722http://doi.org/10.1631/FITEE.2000722

Kundu JN, Venkat N, Rahul MV, et al., 2020. Universal source-free domain adaptation. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.4544-4553. doi: 10.1109/CVPR42600.2020.00460http://doi.org/10.1109/CVPR42600.2020.00460

Kundu JN, Kulkarni A, Singh A, et al., 2021. Generalize then adapt: source-free domain adaptive semantic segmentation. Proc IEEE/CVF Int Conf on Computer Vision, p.7046-7056. doi: 10.1109/ICCV48922.2021.00696http://doi.org/10.1109/ICCV48922.2021.00696

Li R, Jiao QF, Cao WM, et al., 2020. Model adaptation: unsupervised domain adaptation without source data. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.9641-9650. doi: 10.1109/CVPR42600.2020.00966http://doi.org/10.1109/CVPR42600.2020.00966

Li RH, Jia X, He JZ, et al., 2021. T-SVDNet: exploring high-order prototypical correlations for multi-source domain adaptation. Proc IEEE/CVF Int Conf on Computer Vision, p.9991-10000. doi: 10.1109/ICCV48922.2021.00984http://doi.org/10.1109/ICCV48922.2021.00984

Li S, Song SJ, Wu C, 2018. Layer-wise domain correction for unsupervised domain adaptation. Front Inform Technol Electron Eng, 19(1):91-103. doi: 10.1631/FITEE.1700774http://doi.org/10.1631/FITEE.1700774

Liang J, Hu DP, Feng JS, 2020. Do we really need to access the source data? Source hypothesis transfer for unsupervised domain adaptation. Proc 37th Int Conf on Machine Learning, p.6028-6039.

Liu YA, Zhang W, Wang J, 2021. Source-free domain adaptation for semantic segmentation. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.1215-1224. doi: 10.1109/CVPR46437.2021.00127http://doi.org/10.1109/CVPR46437.2021.00127

Long MS, Cao Y, Wang JM, et al., 2015. Learning transferable features with deep adaptation networks. Proc 32nd Int Conf on Machine Learning, p.97-105.

McMahan B, Moore E, Ramage D, et al., 2017. Communication-efficient learning of deep networks from decentralized data. Proc 20th Int Conf on Artificial Intelligence and Statistics, p.1273-1282.

Miyato T, Maeda Si, Koyama M, et al., 2018. Virtual adversarial training: a regularization method for supervised and semi-supervised learning. IEEE Trans Patt Anal Mach Intell, 41(8):1979-1993. doi: 10.1109/tpami.2018.2858821http://doi.org/10.1109/tpami.2018.2858821

Nguyen VA, Nguyen T, Le T, et al., 2021. STEM: an approach to multi-source domain adaptation with guarantees. Proc IEEE/CVF Int Conf on Computer Vision, p.9352-9363. doi: 10.1109/ICCV48922.2021.00922http://doi.org/10.1109/ICCV48922.2021.00922

Peng XC, Bai QX, Xia XD, et al., 2019. Moment matching for multi-source domain adaptation. Proc IEEE/CVF Int Conf on Computer Vision, p.1406-1415. doi: 10.1109/ICCV.2019.00149http://doi.org/10.1109/ICCV.2019.00149

Peng XC, Huang ZJ, Zhu YZ, et al., 2020. Federated adversarial domain adaptation. Proc 8th Int Conf on Learning Representations, p.1-19. doi: 10.48550/arXiv.1911.02054http://doi.org/10.48550/arXiv.1911.02054

Pu SL, Zhao W, Chen WJ, et al., 2021. Unsupervised object detection with scene-adaptive concept learning. Front Inform Technol Electron Eng, 22(5):638-651. doi: 10.1631/FITEE.2000567http://doi.org/10.1631/FITEE.2000567

Qiu Z, Zhang YF, Lin HB, et al., 2021. Source-free domain adaptation via avatar prototype generation and adaptation. Proc 30th Int Joint Conf on Artificial Intelligence, p.2921-2927. doi: 10.24963/ijcai.2021/402http://doi.org/10.24963/ijcai.2021/402

Saito K, Kim D, Sclaroff S, et al., 2019. Semi-supervised domain adaptation via minimax entropy. Proc IEEE/CVF Int Conf on Computer Vision, p.8050-8058. doi: 10.1109/ICCV.2019.00814http://doi.org/10.1109/ICCV.2019.00814

van der Maaten L, Hinton G, 2008. Visualizing data using t-SNE. J Mach Learn Res, 9(86):2579-2605.

Venkat N, Kundu JN, Singh DK, et al., 2020. Your classifier can secretly suffice multi-source domain adaptation. Proc 34th Int Conf on Neural Information Processing Systems, 33:4647-4659. doi: 10.48550/arXiv.2103.11169http://doi.org/10.48550/arXiv.2103.11169

Wang B, Li G, Wu C, et al., 2022. A framework for self-supervised federated domain adaptation. EURASIP J Wirel Commun Network, 2022(1):37. doi: 10.1186/s13638-022-02104-8http://doi.org/10.1186/s13638-022-02104-8

Wang H, Xu MH, Ni BB, et al., 2020. Learning to combine: knowledge aggregation for multi-source domain adaptation. Proc 16th European Conf on Computer Vision, p.727-744.

Wu GL, Gong SG, 2021. Collaborative optimization and aggregation for decentralized domain generalization and adaptation. Proc IEEE/CVF Int Conf on Computer Vision, p.6484-6493. doi: 10.1109/iccv48922.2021.00642http://doi.org/10.1109/iccv48922.2021.00642

Xia HF, Zhao HD, Ding ZM, 2021. Adaptive adversarial network for source-free domain adaptation. Proc IEEE/CVF Int Conf on Computer Vision, p.9010-9019. doi: 10.1109/ICCV48922.2021.00888http://doi.org/10.1109/ICCV48922.2021.00888

Xu RJ, Chen ZL, Zuo WM, et al., 2018. Deep cocktail network: multi-source unsupervised domain adaptation with category shift. Proc IEEE Conf on Computer Vision and Pattern Recognition, p.3964-3973. doi: 10.1109/CVPR.2018.00417http://doi.org/10.1109/CVPR.2018.00417

Yang LY, Balaji Y, Lim SN, et al., 2020. Curriculum manager for source selection in multi-source domain adaptation. Proc 16th European Conf on Computer Vision, p.608-624. doi: 10.1007/978-3-030-58568-6_36http://doi.org/10.1007/978-3-030-58568-6_36

Yang Q, Liu Y, Cheng Y, et al., 2020. Federated Learning. Springer, Cham, Germany.

Yang Y, Zhuang YT, Pan YH, 2021. Multiple knowledge representation for big data artificial intelligence: framework, applications, and case studies. Front Inform Technol Electron Eng, 22(12):1551-1558. doi: 10.1631/FITEE.2100463http://doi.org/10.1631/FITEE.2100463

Zhao H, Zhang SH, Wu GH, et al., 2018. Adversarial multiple source domain adaptation. Proc 32nd Int Conf on Neural Information Processing Systems, p.8568-8579.

Zhao SC, Li B, Yue XY, et al., 2019. Multi-source domain adaptation for semantic segmentation. Proc 33rd Int Conf on Neural Information Processing Systems, p.655.

Zhao SC, Wang GZ, Zhang SH, et al., 2020. Multi-source distilling domain adaptation. Proc AAAI on Conf Artificial Intelligence, 34(7):12975-12983. doi: 10.1609/aaai.v34i07.6997http://doi.org/10.1609/aaai.v34i07.6997

Zhao SC, Li B, Xu PF, et al., 2021. Madan: multi-source adversarial domain aggregation network for domain adaptation. Int J Comput Vis, 129(8):2399-2424. doi: 10.1007/s11263-021-01479-3http://doi.org/10.1007/s11263-021-01479-3

Zhao YB, Zhang H, Gao Z, et al., 2022. A temporal-aware relation and attention network for temporal action localization. IEEE Trans Image Process, 31:4746-4760. doi: 10.1109/TIP.2022.3182866http://doi.org/10.1109/TIP.2022.3182866

Zhou KY, Yang YX, Qiao Y, et al., 2021. Domain adaptive ensemble learning. IEEE Trans Image Process, 30:8008-8018. doi: 10.1109/TIP.2021.3112012http://doi.org/10.1109/TIP.2021.3112012

Zuo YK, Yao HT, Xu CS, 2021. Attention-based multi-source domain adaptation. IEEE Trans Image Process, 30:3793-3803. doi: 10.1109/TIP.2021.3065254http://doi.org/10.1109/TIP.2021.3065254

浏览量

Downloads

CSCD

文章被引用时，请邮件提醒。

Submit

工具集

关联资源

暂无数据