Existing Unsupervised Domain Adaptation (UDA) literature adopts the covariateshift and conditional shift assumptions, which essentially encourage models to learn common features across domains . However, due to the lack of supervision in the target domain, they suffer from the semantic loss . We use a causal view — transportabilitytheory — to identify that such loss is in fact a confounding effect, which can only be removed by causal intervention . We propose a practical solution:Transporting Causal Mechanisms (TCM) to identify the confounder stratum andrepresentations by using domain-invariant disentangled causal mechanisms . Extensive experiments show that TCM achieves state-of-the-art performance on three challenging UDA benchmarks: ImageCLEF-DA,Office-Home, and VisDA-2017

Author(s) : Zhongqi Yue, Hanwang Zhang, Qianru Sun, Xian-Sheng Hua

Links : PDF - Abstract

Code :
Coursera

Keywords : causal - domain - mechanisms - transporting - tcm -

Leave a Reply

Your email address will not be published. Required fields are marked *