Discriminative information preservation: A general framework for unsupervised visual Domain Adaptation
作者：
Highlights：
• A novel feature learning method DIPDA is proposed for unsupervised domain adaptation.
• DIPDA is extended to kernelized version with linear and RBF kernel functions.
• manifold subspaces are considered to reduce time complexity and to improve accuracy.
• Both domain discriminative information is preserved to obtain well separated clusters.
• The validity of the proposed methods is verified by extensive comparison experiments.
摘要
•A novel feature learning method DIPDA is proposed for unsupervised domain adaptation.•DIPDA is extended to kernelized version with linear and RBF kernel functions.•manifold subspaces are considered to reduce time complexity and to improve accuracy.•Both domain discriminative information is preserved to obtain well separated clusters.•The validity of the proposed methods is verified by extensive comparison experiments.
论文关键词：Feature learning,Classification,Unsupervised discriminant analysis,Manifold learning,Transfer learning,Domain adaptation
论文评审过程：Received 7 January 2021, Revised 11 April 2021, Accepted 14 May 2021, Available online 20 May 2021, Version of Record 31 May 2021.
论文官网地址：https://doi.org/10.1016/j.knosys.2021.107158
