Source-Free Unsupervised Domain Adaptation with Sample Transport Learning

作     者：Qing Tian Chuang Ma Feng-Yuan Zhang Shun Peng Hui Xue Qing Tian;Chuang Ma;Feng-Yuan Zhang;Shun Peng;Hui Xue

作者机构：School of Computer and SoftwareNanjing University of Information Science and TechnologyNanjing 210044China Engineering Research Center of Digital ForensicsMinistry of EducationNanjing University of Information Science and TechnologyNanjing 210044China School of Computer Science and EngineeringSoutheast UniversityNanjing 211189China 

出 版 物：《Journal of Computer Science & Technology》 (计算机科学技术学报（英文版）)

年 卷 期：2021年第36卷第3期

页      面：606-616页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：This work was partially supported by the National Natural Science Foundation of China under Grant Nos.61702273 and 62076062 the Natural Science Foundation of Jinangsu Province of China under Grant No.BK20170956 the Open Projects Program of National Laboratory of Pattern Recognition under Grant No.20200007 was also sponsored by Qing Lan Project 

主　　题：unsupervised domain adaptation domain shift sample transport pseudo source domain 

摘      要：Unsupervised domain adaptation(UDA)has achieved great success in handling cross-domain machine learning *** typically benefits the model training of unlabeled target domain by leveraging knowledge from labeled source *** this purpose,the minimization of the marginal distribution divergence and conditional distribution divergence between the source and the target domain is widely adopted in existing ***,for the sake of privacy preservation,the source domain is usually not provided with training data but trained predictor(e.g.,classifier).This incurs the above studies infeasible because the marginal and conditional distributions of the source domain are *** this end,this article proposes a source-free UDA which jointly models domain adaptation and sample transport learning,namely Sample Transport Domain Adaptation(STDA).Specifically,STDA constructs the pseudo source domain according to the aggregated decision boundaries of multiple source classifiers made on the target ***,it refines the pseudo source domain by augmenting it through transporting those target samples with high confidence,and consequently generates labels for the target *** train the STDA model by performing domain adaptation with sample transport between the above steps in alternating manner,and eventually achieve knowledge adaptation to the target domain and attain confident labels for ***,evaluation results have validated effectiveness and superiority of the proposed method.