A study of the effects of negative transfer on deep unsupervised domain adaptation methods

作者：

Highlights：

• A study of the effects of negative transfer is performed over D-UDA methods.

• The study covers the effect of dataset shifts and distribution shapes.

• A discussion based on the results of the evaluated methods is presented.

• Some insights to select and design robust D-UDA methods is provided.

摘要

•A study of the effects of negative transfer is performed over D-UDA methods.•The study covers the effect of dataset shifts and distribution shapes.•A discussion based on the results of the evaluated methods is presented.•Some insights to select and design robust D-UDA methods is provided.

论文关键词：Unsupervised domain adaptation,Deep learning,Negative transfer,Dataset shift

论文评审过程：Received 11 June 2019, Revised 12 September 2020, Accepted 1 October 2020, Available online 17 October 2020, Version of Record 10 February 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.114088