Unsupervised domain adaptation with adversarial distribution adaptation network
- PDF / 988,494 Bytes
- 13 Pages / 595.276 x 790.866 pts Page_size
- 51 Downloads / 222 Views
(0123456789().,-volV)(0123456789(). ,- volV)
ORIGINAL ARTICLE
Unsupervised domain adaptation with adversarial distribution adaptation network Qiang Zhou1 • Wen’an Zhou1 • Shirui Wang1 • Ying Xing2 Received: 3 June 2020 / Accepted: 4 November 2020 Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Adversarial domain adaptation is a powerful approach to transfer the knowledge of the label-rich source domain to the label-scarce target domain by mitigating domain shifts across distributions. Existing domain adaptation methods align either the marginal distribution with a single-domain discriminator or conditional distributions with multiple-domain discriminators. However, aligning both marginal (global) and conditional (local) distributions should be considered for domain adaptation. This paper proposes a novel adversarial distribution adaptation network (ADAN) to jointly reduce both the global and local distribution discrepancies between different domains for learning domain-invariant representations. ADAN utilizes a single-domain discriminator to adapt the global distribution between two domains, and source decision boundaries to align the local distributions between sub-domains. Furthermore, we extend our ADAN as improved ADAN (iADAN), in which we utilize a feature norm term to regularize the task-specific features to improve model generalization. Extensive experimental results show that our method outperforms other state-of-the-art domain adaptation methods on Office-Home and ImageCLEF-DA datasets. Keywords Domain adaptation Adversarial training Image classification
1 Introduction Deep neural networks have made great progress in computer vision with the help of numerous labeled samples [9, 10]. However, collecting numerous labeled samples can be expensive and time-consuming. Data from different but related domains can be available for target tasks [22]. Domain adaptation tackles this problem by transferring knowledge from a label-rich source domain to a label& Wen’an Zhou [email protected] Qiang Zhou [email protected] Shirui Wang [email protected] Ying Xing [email protected] 1
School of Computer Science, Beijing University of Posts and Telecommunications, Beijing 100876, China
2
School of Automation, Beijing University of Posts and Telecommunications, Beijing 100876, China
scarce target domain, and domain adaptation aims at learning a discriminative model for both source and target domains by reducing the dataset shift [11, 19, 27] between training (source) and test (target) datasets. Domain adaptation has been extensively studied in recent years [5, 6, 13, 15, 21, 28, 32], unlike the common assumption in theoretic learning model, such as standard PAC model [29], source training samples are not drawn from the same distribution as the target test samples in the domain adaptation setting, so the model trained on the source labeled data cannot be directly implemented on the target domain. Domain adaptation methods aim at learning domain-invariant representations by reducing t
Data Loading...
