Unsupervised domain adaptation via discriminative feature learning and classifier adaptation from center-based distances

作者：

Highlights：

• Center-based distances (CS, DC, CNC) are designed on data-centers and classcenters.

• To learn the discriminative features, DCD minimizes CS+MMD and maximizes DC+CNC.

• To better measure the local structure, CS & CNC are used in Laplacian matrix of CCD.

• Sufficient experiments demonstrate that DCCD outperforms the advanced methods.

摘要

•Center-based distances (CS, DC, CNC) are designed on data-centers and classcenters.•To learn the discriminative features, DCD minimizes CS+MMD and maximizes DC+CNC.•To better measure the local structure, CS & CNC are used in Laplacian matrix of CCD.•Sufficient experiments demonstrate that DCCD outperforms the advanced methods.

论文关键词：Classifier adaptation,Center-based distances,Discriminative feature learning,Laplacian Regularization,Unsupervised domain adaptation

论文评审过程：Received 5 May 2021, Revised 30 August 2021, Accepted 9 May 2022, Available online 16 May 2022, Version of Record 25 May 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109022