Challenging tough samples in unsupervised domain adaptation
作者：
Highlights：
• A novel domain adaptation method, named challenging tough sample networks (CTSN), is proposed to challenge tough samples in the target domain.
• We report that leveraging the labels of easy target samples can ideally convert an unsupervised domain adaptation problem to a semi-supervised one.
• An algorithm for tough sample identification is developed.
摘要
•A novel domain adaptation method, named challenging tough sample networks (CTSN), is proposed to challenge tough samples in the target domain.•We report that leveraging the labels of easy target samples can ideally convert an unsupervised domain adaptation problem to a semi-supervised one.•An algorithm for tough sample identification is developed.
论文关键词：Domain adaptation,transfer learning,adversarial learning
论文评审过程：Received 7 November 2019, Revised 26 May 2020, Accepted 3 July 2020, Available online 6 July 2020, Version of Record 1 November 2020.
论文官网地址：https://doi.org/10.1016/j.patcog.2020.107540
