Sparsely-labeled source assisted domain adaptation. (April 2021)

• Record Type: Journal Article
• Authors: Wang, Wei; Chen, Shenglun; Xiang, Yuankai; Sun, Jing; Li, Haojie; Wang, Zhihui; Sun, Fuming; Ding, Zhengming; Li, Baopu
• Is Part Of: Pattern recognition. Volume 112(2021)
• Publication Date: 2021-04
• Subjects: Domain adaptation -- Sparsely-labeled source -- Semi-supervised clustering -- Label propagation; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2020.107803
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Firstly, we consider a new yet practical DA scenario, called sparsely-labeled source assisted domain adaptation. Secondly, we propose a unified framework to jointly seek cluster centroids, source and target labels, and domain-invariant features. Thirdly, we construct an optimization strategy to solve the objective function efficiently. Abstract: Domain Adaptation (DA) aims to generalize the classifier learned from a well-labeled source domain to an unlabeled target domain. Existing DA methods usually assume that rich labels could be available in the source domain. However, we usually confront the source domain with a large number of unlabeled data but only a few labeled data, and thus, how to transfer knowledge from this sparsely-labeled source domain to the target domain is still a challenge, which greatly limits its application in the wild. This paper proposes a novel Sparsely-Labeled Source Assisted Domain Adaptation (SLSA-DA) algorithm to address the challenge with limited labeled source domain samples. Specifically, due to the label scarcity problem, the projected clustering is first conducted on both the source and target domains, so that the discriminative structures of data could be exploited elegantly. Then label propagation is adopted to propagate the labels from those limited labeled source samples to the whole unlabeled data progressively, so that the cluster labels are revealed correctly. Finally, we jointly align the marginal and conditional distributions to mitigate the cross-domain mismatching problem, and optimize those three procedures iteratively. However, it is nontrivial to incorporate the above three procedures into a unified optimization framework seamlessly since some variables to be optimized are implicitly involved in their formulas, thus they could not benefit to each other. Remarkably, we prove that the projected clustering and conditional distribution alignment could be reformulated into other formulations, thus the implicit variables are embedded in different optimization steps. As such, the variables related to those three quantities could be optimized in a unified optimization framework and benefit to each other, and improve the recognition performance obviously. Extensive experiments have verified that our approach could deal with the challenge in the SLSA-DA setting, and achieve the best performances across different real-world cross-domain visual recognition tasks. Our preliminary Matlab code is available at https://github.com/WWLoveTransfer/SLSA-DA/ .