Robust multi-source co-adaptation with adaptive loss minimization

作者：

Highlights：

• We present a unified framework of multi-source adaptation learning by jointly exploiting the correlation.

• Multiple feature selection functions for different source adaptation objects are simultaneously learned in a joint framework.

• Due to local graph Laplacian regularization and adaptive regression scheme, our framework can preserve the geometrical structure and be robust to noises.

• We further present an effective extension to domain generalization.

摘要

•We present a unified framework of multi-source adaptation learning by jointly exploiting the correlation.•Multiple feature selection functions for different source adaptation objects are simultaneously learned in a joint framework.•Due to local graph Laplacian regularization and adaptive regression scheme, our framework can preserve the geometrical structure and be robust to noises.•We further present an effective extension to domain generalization.

论文关键词：Multi-source adaptation,Domain generalization,Adaptive loss,Maximum mean discrepancy

论文评审过程：Received 3 February 2021, Revised 16 July 2021, Accepted 23 August 2021, Available online 1 September 2021, Version of Record 20 September 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116455