基于最优传输距离正则化和近邻聚类方法的开放集域适应

doi:10.19678/j.issn.1000-3428.0070761

摘要/Abstract

摘要： 无监督域适应(UDA)旨在将知识从标记的源域迁移到未标记的目标域,从而提高目标域模型的性能。然而,传统的UDA方法假设源域和目标域的类别空间完全一致,导致无法处理目标域中存在的未知类别,这限制了其在实际场景中的应用。开放集域适应(OSDA)通过引入对未知类别的识别来解决这一问题,但如何有效减少域间差异和类别不平衡对模型性能的负面影响,仍是一个重要挑战。现有的OSDA方法往往忽略了域特定特征,并简单地对域差异直接进行最小化,这可能导致类别之间的边界不清晰并削弱模型的泛化能力。为了解决这一问题,提出基于最优传输距离正则化(OTR)和近邻聚类方法的开放集域适应(OTRNC)方法。该方法通过OTR来最大化高、低置信度样本组之间的分布距离,减少未知类别对域适应过程的干扰。然后利用动态近邻检索和不变特征学习,减少目标域的类内变化,增强特征的泛化能力。实验结果表明,OTRNC在多个基准数据集上均表现出色,具有一定的有效性。

关键词: 开放集域适应, 最优传输距离正则化, 动态近邻检索, 不变特征学习, 近邻聚类

Abstract: Unsupervised Domain Adaptation (UDA) aims to transfer knowledge from a labeled source domain to an unlabeled target domain, thereby improving the performance of the target domain model. However, traditional UDA methods assume that the category spaces of the source and target domains are identical; hence, they cannot handle unknown categories present in the target domain, which limits their applicability in real-world scenarios. Open-Set Domain Adaptation (OSDA) addresses this issue by introducing the recognition of unknown categories. Nevertheless, effectively mitigating the negative impact of inter-domain discrepancies and class imbalances on the model performance remains a significant challenge. Existing OSDA methods tend to overlook domain-specific features and directly minimize domain discrepancies, which may lead to ambiguous decision boundaries between classes and weaken the generalization ability of the model. To address this problem, this paper proposes an open-set domain adaptation method based on Optimal Transport Distance Regularization (OTR) and neighborhood clustering, called OTRNC. Specifically, the method employs OTR to maximize the distributional distance between high-confidence and low-confidence sample groups, thereby reducing the interference of unknown categories in the domain adaptation process. Subsequently, it leverages dynamic neighbor retrieval and invariant feature learning to reduce intra-class variations in the target domain and enhance the generalization capability of the learned features. Experimental results demonstrate that OTRNC achieves superior performance across multiple benchmark datasets, thereby validating its effectiveness.

Key words: Open-Set Domain Adaptation (OSDA), Optimal Transport Distance Regularization (OTR), dynamic neighbor retrieval, invariant feature learning, neighbor clustering

中图分类号:

TP391

田青, 申珺妤, 郁江森. 基于最优传输距离正则化和近邻聚类方法的开放集域适应[J]. 计算机工程, 2026, 52(10): 116-128.

TIAN Qing, SHEN Junyu, YU Jiangsen. Open-Set Domain Adaptation Based on Optimal Transport Distance Regularization and Neighbor Clustering Method[J]. Computer Engineering, 2026, 52(10): 116-128.

https://www.ecice06.com/CN/Y2026/V52/I10/116

参考文献

[1] TIAN Q, SUN H Y, MA C, et al. Heterogeneous domain adaptation with structure and classification space alignment[J]. IEEE Transactions on Cybernetics, 2022, 52(10): 10328-10338.
[2] FARAHANI A, VOGHOEI S, RASHEED K, et al. A brief review of domain adaptation[EB/OL].[2024-10-05]. https://arxiv.org/abs/2010.03978.
[3] BUSTO P P, GALL J. Open set domain adaptation[C]//Proceedings of the IEEE International Conference on Computer Vision (ICCV). Washington D.C.,USA:IEEE Press,2017: 754-763.
[4] 刘星宏, 周毅, 周涛, 等. 基于自步学习的开放集领域自适应[J]. 计算机研究与发展, 2023, 60(8): 1711-1726. LIU X H, ZHOU Y, ZHOU T, et al. Self-paced learning for open-set domain adaptation[J]. Journal of Computer Research and Development, 2023, 60(8): 1711-1726. (in Chinese)
[5] ZHANG R T, WEI J G, LU X G, et al. Self-supervised domain exploration with an optimal transport regularization for open set cross-domain speech emotion recognition[C]//Proceedings of the 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Washington D.C.,USA:IEEE Press, 2024: 10466-10470.
[6] 罗浩, 王彦捷, 牛明航, 等. 动态区间的加权模糊聚类算法[J]. 计算机科学与探索, 2020, 14(7): 1142-1153. LUO H, WANG Y J, NIU M H, et al. Weighted fuzzy clustering algorithm based on dynamic interval[J]. Journal of Frontiers of Computer Science and Technology, 2020, 14(7): 1142-1153. (in Chinese)
[7] 王少将, 刘佳, 郑锋, 等. 机器学习层谱聚类综述[J]. 计算机科学, 2023, 50(1): 9-17. WANG S J, LIU J, ZHENG F, et al. Survey on hierarchical clustering for machine learning[J]. Computer Science, 2023, 50(1): 9-17. (in Chinese)
[8] 张永, 夏天琦, 黄丹. 基于无监督域适应的可区分联合匹配算法[J]. 模式识别与人工智能, 2021, 34(10): 932-940. ZHANG Y, XIA T Q, HUANG D. Discriminative joint matching for unsupervised domain adaptation[J]. Pattern Recognition and Artificial Intelligence, 2021, 34(10): 932-940. (in Chinese)
[9] TZENG E, HOFFMAN J, SAENKO K, et al. Adversarial discriminative domain adaptation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2017: 2962-2971.
[10] KANG G L, ZHENG L, YAN Y, et al. Deep adversarial attention alignment for unsupervised domain adaptation: the benefit of target expectation maximization[EB/OL].[2024-10-05]. https://arxiv.org/abs/1801.10068.
[11] CHOE J, SHIM H. Attention-based dropout layer for weakly supervised object localization[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2020: 2214-2223.
[12] FENG Y L, ZHU H Y, PENG D Z, et al. ROAD: robust unsupervised domain adaptation with noisy labels[C]//Proceedings of the 31st ACM International Conference on Multimedia. New York,USA:ACM Press,2023: 7264-7273.
[13] SCHEIRER W J, DE REZENDE ROCHA A, SAPKOTA A, et al. Toward open set recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(7): 1757-1772.
[14] SAITO K, YAMAMOTO S, USHIKU Y, et al. Open set domain adaptation by backpropagation[EB/OL].[2024-10-05]. https://arxiv.org/abs/1804.10427.
[15] ZOU Y, YU Z D, LIU X F, et al. Confidence regularized self-training[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Washington D.C.,USA:IEEE Press,2020: 5981-5990.
[16] PAN Y W, YAO T, LI Y H, et al. Transferrable prototypical networks for unsupervised domain adaptation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press, 2020: 2234-2242.
[17] GAO F, PI D C, CHEN J F. Balanced and robust unsupervised open set domain adaptation via joint adversarial alignment and unknown class isolation[J]. Expert Systems with Applications, 2024, 238: 122127.
[18] COURTY N, FLAMARY R, TUIA D, et al. Optimal transport for domain adaptation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(9): 1853-1865.
[19] LI H Q, KIM Y, KUO C H, et al. Acted vs. improvised: domain adaptation for elicitation approaches in audio-visual emotion recognition[EB/OL].[2024-10-05]. https://arxiv.org/abs/2104.01978.
[20] COURTY N, FLAMARY R, HABRARD A, et al. Joint distribution optimal transportation for domain adaptation[EB/OL].[2024-10-05]. https://arxiv.org/abs/1705.08848.
[21] ZHANG R T, WEI J G, LU X G, et al. Optimal transport with a diversified memory bank for cross-domain speaker verification[C]//Proceedings of the 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Washington D.C.,USA:IEEE Press,2023: 1-5.
[22] SHAMEEM M U S, FERDOUS R. An efficient k-means algorithm integrated with Jaccard distance measure for document clustering[C]//Proceedings of the 1st Asian Himalayas International Conference on Internet. Washington D.C.,USA:IEEE Press,2009: 1-6.
[23] SAENKO K, KULIS B, FRITZ M, et al. Adapting visual category models to new domains[EB/OL].[2024-10-05]. https://people.bu.edu/bkulis/pubs/saenko_eccv_2010.pdf.
[24] VENKATESWARA H, EUSEBIO J, CHAKRABORTY S, et al. Deep hashing network for unsupervised domain adaptation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2017: 5385-5394.
[25] PENG X C, USMAN B, KAUSHIK N, et al. VisDA: a synthetic-to-real benchmark for visual domain adaptation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Washington D.C.,USA:IEEE Press,2018: 2102-2105.
[26] DENG J, DONG W, SOCHER R, et al. ImageNet: a large-scale hierarchical image database[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Washington D.C.,USA:IEEE Press,2009: 248-255.
[27] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2016: 770-778.
[28] LU X G, SHEN P, TSAO Y, et al. Unsupervised neural adaptation model based on optimal transport for spoken language identification[C]//Proceedings of the 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Washington D.C.,USA:IEEE Press,2021: 7213-7217.
[29] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL].[2024-10-05]. https://arxiv.org/abs/1409.1556.
[30] GANIN Y, USTINOVA E, AJAKAN H, et al. Domain-adversarial training of neural networks[EB/OL].[2024-10-05]. https://arxiv.org/abs/1505.07818.
[31] LIU H, CAO Z J, LONG M S, et al. Separate to adapt: open set domain adaptation via progressive separation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press, 2020: 2922-2931.
[32] GRETTON A, BORGWARDT K M, RASCH M,et al. A kernel two-sample test[J]. Journal of Machine Learning Research, 2012, 13: 723-773.
[33] 但雨芳, 陶剑文. 可能性分布距离度量: 一种鲁棒的域适应学习方法[J]. 计算机科学与探索, 2024, 18(3): 674-692. DAN Y F, TAO J W. Possibilistic distribution distance measure: robust domain adaptation learning method[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(3): 674-692. (in Chinese)
[34] FANG Z, LU J, LIU F, et al. Open set domain adaptation: theoretical bound and algorithm[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(10): 4309-4322.
[35] LUO Y D, WANG Z J, HUANG Z, et al. Progressive graph learning for open-set domain adaptation[EB/OL].[2024-10-05]. https://arxiv.org/abs/2006.12087.
[36] WANG Q, MENG F L, BRECKON T P. Progressively select and reject pseudo-labelled samples for open-set domain adaptation[EB/OL].[2024-10-05]. https://arxiv.org/abs/2110.12635.
[37] BUCCI S, LOGHMANI M R, TOMMASI T. On the effectiveness of image rotation for open set domain adaptation[EB/OL].[2024-10-05]. https://arxiv.org/abs/2007.12360.
[38] LIU Z F, CHEN G, LI Z J, et al. PSDC: a prototype-based shared-dummy classifier model for open-set domain adaptation[J]. IEEE Transactions on Cybernetics, 2023, 53(11): 7353-7366.
[39] JANG J, JI M, MOON I C, et al. Unknown-aware domain adversarial learning for open-set domain adaptation[EB/OL].[2024-10-05]. https://papers.nips.cc/paper_files/paper/2022/hash/6a934325ec64639ba83b492b9c317085-Abstract-Conference.html.
[40] LI J, YANG L, WANG Q L, et al. WDAN: a weighted discriminative adversarial network with dual classifiers for fine-grained open-set domain adaptation[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2023, 33(9): 5133-5147.
[41] YOU K C, LONG M S, CAO Z J, et al. Universal domain adaptation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2020: 2715-2724.
[42] LI W Y, LIU J, HAN B, et al. Adjustment and alignment for unbiased open set domain adaptation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2023: 24110-24119.
[43] LI G R, KANG G L, ZHU Y, et al. Domain consensus clustering for universal domain adaptation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C.,USA:IEEE Press,2021: 9752-9761.
[44] XU Y M, CHEN L, DUAN L X, et al. Open set domain adaptation with soft unknown-class rejection[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023, 34(3): 1601-1612.
[45] CHANG D L, SAIN A, MA Z Y, et al. Mind the gap: open set domain adaptation via mutual-to-separate framework[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2024, 34(6): 4159-4174.
[46] 汪云云, 孙顾威, 赵国祥, 等. 基于自监督知识的无监督新集域适应学习[J]. 软件学报, 2022, 33(4): 1170-1182. WANG Y Y, SUN G W, ZHAO G X, et al. Unsupervised new-set domain adaptation with self-supervised knowledge[J]. Journal of Software, 2022, 33(4): 1170-1182. (in Chinese)
[47] 田青, 孙灿宇, 储奕. 基于自适应权重的多源部分域适应[J]. 软件学报, 2024, 35(4): 1703-1716. TIAN Q, SUN C Y, CHU Y. Adaptive weight-induced multi-source partial domain adaptation[J]. Journal of Software, 2024, 35(4): 1703-1716. (in Chinese)
[48] WANG Y F, HUANG H Y, RUDIN C, et al. Understanding how dimension reduction tools work: an empirical approach to deciphering t-SNE, UMAP, TriMAP, and PaCMAP for data visualization[J]. Journal of Machine Learning Research, 2020, 22: 73.
[49] LU Y Z, SHEN M, MA A J, et al. MLNet: mutual learning network with neighborhood invariance for universal domain adaptation[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2024, 38(4): 3900-3908.
[50] TIAN Q, ZHU Y N, SUN H Y, et al. Unsupervised domain adaptation through dynamically aligning both the feature and label spaces[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(12): 8562-8573.

选择文件类型/文献管理软件名称

选择包含的内容