Domain-Symmetric Networks for Adversarial Domain Adaptation

By Yabin Zhang, Hui Tang, K. Jia, Mingkui Tan. 2019

354 citations recorded by the source corpus.

Unsupervised domain adaptation aims to learn a model of classifier for unlabeled samples on the target domain, given training data of labeled samples on the source domain. Impressive progress is made recently by learning invariant features via domain-adversarial training of deep networks. In spite of the recent progress, domain adaptation is still limited in achieving the invariance of feature distributions at a finer category level. To this end, we propose in this paper a new domain adaptation method called Domain-Symmetric Networks (SymNets). The proposed SymNet is based on a symmetric design of source and target task classifiers, based on which we also construct an additional classifier that shares with them its layer neurons. To train the SymNet, we propose a novel adversarial learning objective whose key design is based on a two-level domain confusion scheme, where the category-level confusion loss improves over the domain-level one by driving the learning of intermediate network features to be invariant at the corresponding categories of the two domains. Both domain discrimination and domain confusion are implemented based on the constructed additional classifier. Since target samples are unlabeled, we also propose a scheme of cross-domain training to help learn the target classifier. Careful ablation studies show the efficacy of our proposed method. In particular, based on commonly used base networks, our SymNets achieve the new state of the art on three benchmark domain adaptation datasets.

Subfields: Artificial Intelligence, Infectious Diseases, Radiology, Nuclear Medicine and Imaging, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Infectious Diseases

In Radiology, Nuclear Medicine and Imaging

In Safety, Robustness & Interpretability

Citations in the corpus

Repurposing Pretrained Models for Robust Out-of-domain Few-Shot Learning (2021, background): “GAN-based DA approaches require having access to the large amount of unlabeled data from both the source and target datasets (Tzeng et al., 2017; Zhang et al., 2019; Wilson & Cook, 2020).”
A Comprehensive Survey on Transfer Learning (2019, background): “[159] adopted a symmetric design for the source and the target classifiers.”
HGNet: Hybrid Generative Network for Zero-Shot Domain Adaptation (2020, background): “Concretely, it attempts to borrow knowledge from well-annotated modality (source domain) to solve classiﬁcation task on target domain without any label information [30, 32, 35].”
Gradually Vanishing Bridge for Adversarial Domain Adaptation (2020, methodology): “Symnets [48] makes a symmetric design of source and target task classifiers, in which both domain classifiers produce classification outputs.”
Domain Generalization Using a Mixture of Multiple Latent Domains (2019, methodology): “Therefore, we introduce the entropy loss Lent (Grandvalet and Bengio 2005), which is used in some domain adaptation methods (Long et al. 2016; Zhang et al. 2019) to train a more discriminative model for target samples by encouraging low-density separation between object categories.”
BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation (2021, methodology): “Recent methods aim to improve the domain adaptability of deep neural networks via adversarial training [35, 62].”
Under review as a conference paper at ICLR 2020 many domain adaptation methods (2019, background): “Several other domain adaptation papers in adversarial learning using conditional learning Long et al. (2018), domain-symmetric Zhang et al. (2019), and collaborative Zhang et al. (2018b) methods have been introduced.”
Reducing Domain Gap by Reducing Style Bias (2021, methodology): “We demonstrate that SagNets are not only effective when used alone, but also able to complement popular UDA methods such as DANN [11], JAN [35], CDAN [34], and SymNet [51] to make further improvements.”
Cross-Domain Gradient Discrepancy Minimization for Unsupervised Domain Adaptation (2021, background): “Later, some studies [25, 32, 44, 15] suggest to align conditional distributions in feature space to achieve accurate alignment at the category-level.”
Reliable Weighted Optimal Transport for Unsupervised Domain Adaptation (2020, background): “Domain-Symmetric Networks [50] learns invariant features in the domain adaptation.”
Unsupervised Domain Adaptation via Structurally Regularized Deep Clustering (2020, methodology): “To enforce the cluster assumption, conditional entropy minimization [20, 32] is widely used in the UDA community [11, 44, 45, 50, 51, 60, 64, 68].”
Minimum Class Confusion for Versatile Domain Adaptation (2019, background): “], CDAN [27] proposes to align features in a conditional adversarial manner. CyCADA [18] adapts features in both pixel and feature level. TADA [54] proposes a transferable attention mechanism. SymNet [61] introduces a symmetric classier, and DTA [23] learns discriminative features with adversarial d”
Sequential Unsupervised Domain Adaptation through Prototypical Distributions (2020, methodology): “We include methods that are based on adversarial learning: GtA [6], DANN [29], SymNets [8] ADDA [30], MADA [7], and CDAN [9].”
FixBi: Bridging Domain Spaces for Unsupervised Domain Adaptation (2020, neutral): “9 SymNets [46] 90.”