BP-triplet net for unsupervised domain adaptation: A Bayesian perspective
作者：
Highlights：
• We design a BP-Triplet net to learn robust representations which are not only domain invariant but also class discriminative.
• Our method aims to deal with pair-wise importance imbalance and enhance domain confusion during classwise feature alignment.
• Our loss function is deduced from the perspective of Bayesian learning and MAP, which is a novel technical contribution.
摘要
•We design a BP-Triplet net to learn robust representations which are not only domain invariant but also class discriminative.•Our method aims to deal with pair-wise importance imbalance and enhance domain confusion during classwise feature alignment.•Our loss function is deduced from the perspective of Bayesian learning and MAP, which is a novel technical contribution.
论文关键词：Cross domain class alignment,Unsupervised domain adaptation,Metric learning,Bayesian perspective
论文评审过程：Received 19 October 2021, Revised 22 July 2022, Accepted 20 August 2022, Available online 28 August 2022, Version of Record 7 September 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.108993
