Pairwise Adversarial Training for Unsupervised Class-imbalanced Domain Adaptation

Unsupervised domain adaptation (UDA) has become an appealing approach for knowledge transfer from a labeled source domain to an unlabeled target domain. However, when the classes in source and target domains are imbalanced, most existing UDA methods experience significant performance drop, as the decision boundary usually favors the majority classes. Some recent class-imbalanced domain adaptation (CDA) methods aim to tackle the challenge of biased label distribution by exploiting pseudo-labeled target data during training process. However, these methods may be challenged with the problem of unreliable pseudo labels and error accumulation during training. In this paper, we propose a pairwise adversarial training approach to augment training data for unsupervised class-imbalanced domain adaptation. Unlike conventional adversarial training in which the adversarial samples are obtained from the $\ell_p$ ball of the original data, we obtain the semantic adversarial samples from the interpolated line of the aligned pair-wise samples from source domain and target domain. Experimental results and ablation study show that our method can achieve considerable improvements on the CDA benchmarks compared with the state-of-art methods focusing on the same problem.