Discriminative feature alignment: Improving transferability of unsupervised domain adaptation by Gaussian-guided latent alignment

• Published in: Pattern recognition Vol. 116; p. 107943
• Main Authors: Wang, Jing, Chen, Jiahong, Lin, Jianzhe, Sigal, Leonid, de Silva, Clarence W.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-08-2021
• Subjects: Computer vision; Domain adaptation; Information theory; Domain adaptation; Computer vision; Information theory
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2021.107943

•We propose a novel alignment method to construct a common feature space under the guidance of a Gaussian prior for UDA.•We introduce a new method to align two distributions by minimizing the direct L1-distance between the decoded samples.•The proposed work achieves state-of-the-art performance on both digit and object classification tasks.
In this paper, we focus on the unsupervised domain adaptation problem where an approximate inference model is to be learned from a labeled data domain and expected to generalize well to an unlabeled domain. The success of unsupervised domain adaptation largely relies on the cross-domain feature alignment. Previous work has attempted to directly align features by classifier-induced discrepancies. Nevertheless, a common feature space cannot always be learned via this direct feature alignment especially when large domain gaps exist. To solve this problem, we introduce a Gaussian-guided latent alignment approach to align the latent feature distributions of the two domains under the guidance of a prior. In such an indirect way, the distributions over the samples from the two domains will be constructed on a common feature space, i.e., the space of the prior, which promotes better feature alignment. To effectively align the target latent distribution with this prior distribution, we also propose a novel unpaired L1-distance by taking advantage of the formulation of the encoder-decoder. The extensive evaluations on nine benchmark datasets validate the superior knowledge transferability through outperforming state-of-the-art methods and the versatility of the proposed method by improving the existing work significantly.