ProxyMix: Proxy-based Mixup training with label refinery for source-free domain adaptation

doi:10.1016/j.neunet.2023.08.005

CORC > 自动化研究所 > 中国科学院自动化研究所 > 多模态人工智能系统全国重点实验室

	ProxyMix: Proxy-based Mixup training with label refinery for source-free domain adaptation
	Ding, Yuhe 5; Sheng, Lijun 1,2,3; Liang, Jian 1,2,6; Zheng, Aihua 4; He, Ran 1,2,6
刊名	NEURAL NETWORKS
	2023-10-01
卷号	167 页码:92-103
关键词	Source-free unsupervised domain adaptation Pseudo labeling
ISSN号	0893-6080
DOI	10.1016/j.neunet.2023.08.005
通讯作者	Liang, Jian(liangjian92@gmail.com)
英文摘要	Due to privacy concerns and data transmission issues, Source-free Unsupervised Domain Adaptation (SFDA) has gained popularity. It exploits pre-trained source models, rather than raw source data for target learning, to transfer knowledge from a labeled source domain to an unlabeled target domain. Existing methods solve this problem typically with additional parameters or noisy pseudo labels, and we propose an effective method named Proxy-based Mixup training with label refinery (ProxyMix) to avoid these drawbacks. To avoid additional parameters and leverages information in the source model, ProxyMix defines classifier weights as class prototypes and creates a class-balanced proxy source domain using nearest neighbors of the prototypes. To improve the reliability of pseudo labels, we further propose the frequency-weighted aggregation strategy to generate soft pseudo labels for unlabeled target data. Our strategy utilizes target features' internal structure, increases weights of low frequency class samples, and aligns the proxy and target domains using inter-and intra-domain mixup regularization. This mitigates the negative impact of noisy labels. Experiments on three 2D image and 3D point cloud object recognition benchmarks demonstrate that ProxyMix yields state-of-the-art performance for source-free UDA tasks. (c) 2023 Elsevier Ltd. All rights reserved.
资助项目	National Natural Science Foundation of China[62276256] ; Beijing Nova Program, China[Z211100002121108] ; University Synergy Innovation Program of Anhui Province[GXXT-2022-036]
WOS研究方向	Computer Science ; Neurosciences & Neurology
语种	英语
出版者	PERGAMON-ELSEVIER SCIENCE LTD
WOS记录号	WOS:001068310300001
资助机构	National Natural Science Foundation of China ; Beijing Nova Program, China ; University Synergy Innovation Program of Anhui Province
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/53089]
专题	多模态人工智能系统全国重点实验室
通讯作者	Liang, Jian
作者单位	1.Chinese Acad Sci CASIA, Inst Automat, Ctr Res Intelligent Percept & Comp, Beijing, Peoples R China 2.Chinese Acad Sci CASIA, State Key Lab Multimodal Artificial Intelligence S, Beijing, Peoples R China 3.Univ Sci & Technol China, Hefei, Peoples R China 4.Anhui Univ, Sch Artificial Intelligence, Hefei, Peoples R China 5.Anhui Univ, Sch Comp Sci & Technol, Hefei, Peoples R China 6.Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing, Peoples R China
推荐引用方式 GB/T 7714	Ding, Yuhe,Sheng, Lijun,Liang, Jian,et al. ProxyMix: Proxy-based Mixup training with label refinery for source-free domain adaptation[J]. NEURAL NETWORKS,2023,167:92-103.
APA	Ding, Yuhe,Sheng, Lijun,Liang, Jian,Zheng, Aihua,&He, Ran.(2023).ProxyMix: Proxy-based Mixup training with label refinery for source-free domain adaptation.NEURAL NETWORKS,167,92-103.
MLA	Ding, Yuhe,et al."ProxyMix: Proxy-based Mixup training with label refinery for source-free domain adaptation".NEURAL NETWORKS 167(2023):92-103.