Multi-Adversarial Domain Adaptation

By Zhongyi Pei, Zhangjie Cao, Mingsheng Long, Jianmin Wang. 2018

959 citations recorded by the source corpus.

Recent advances in deep domain adaptation reveal that adversarial learning can be embedded into deep networks to learn transferable features that reduce distribution discrepancy between the source and target domains. Existing domain adversarial adaptation methods based on single domain discriminator only align the source and target data distributions without exploiting the complex multimode structures. In this paper, we present a multi-adversarial domain adaptation (MADA) approach, which captures multimode structures to enable fine-grained alignment of different data distributions based on multiple domain discriminators. The adaptation can be achieved by stochastic gradient descent with the gradients computed by back-propagation in linear-time. Empirical evidence demonstrates that the proposed model outperforms state of the art methods on standard domain adaptation datasets.

Subfields: Artificial Intelligence, Representation & Self-Supervised Learning

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 Representation & Self-Supervised Learning

Citations in the corpus

Prototype-Guided Continual Adaptation for Class-Incremental Unsupervised Domain Adaptation (2022, methodology): “Unsupervised domain adaptation (UDA) seeks to improve the performance on an unlabeled target domain by leveraging a label-rich source domain via knowledge transfer [5,6,10,14,17,34,51,59].”
Class-Incremental Domain Adaptation (2020, background): “een long-standing. A line of work called Unsupervised Domain Adaptation (UDA) [2,3,5,21,25,31,22,51] has emerged that oers an elegant solution to the domain-shift problem. In UDA, the usual practice [10,37] is to obtain a labeled source dataset and unlabeled targets samples, to perform adaptation un”
Contrastive Adaptation Network for Unsupervised Domain Adaptation (2019, methodology): “For a comparison under optimal parameter setting, we cite the performance of MADA, RevGrad, MCD and ADR reported in their corresponding papers [28, 31, 32, 10].”
Efficient Test-Time Model Adaptation without Forgetting (2022, methodology): “Deep neural networks (DNNs) have achieved excellent performance in many challenging tasks, including image clas-siﬁcation (He et al., 2016; Tan & Le, 2019), video recognition (Wang et al., 2018; Chen et al., 2021), and many other areas (Choi et al., 2018; Fan et al., 2020; Xu et al., 2021).”
Characterizing and Avoiding Negative Transfer (2018, neutral): “In zero-shot transfer learning2 [8, 21] where there is no labeled target data (nl = 0), only using unlabeled target data would result in a weak random model and thus NTC is unlikely to be satisfied.”
Unsupervised adversarial domain adaptation for acoustic scene classification (2018, background): “reverse gradient of the domain classiﬁcation to optimize the parameters that produce the latent representation for the classiﬁcation. A similar concept has been adopted in many subsequent works, e.g. [19]. Later, an adversarial domain adaptation approach is presented in [9], where the training proce”
A Brief Review of Domain Adaptation (2020, background): “Multi-Adversarial Domain Adaptation (MADA)[62] was proposed to reduce the shift in the joint distribution between domains and enable fine-grained alignment of different class distributions by constructing multiple class-wise domain discriminators.”
Unsupervised Domain Adaptation via Structurally Regularized Deep Clustering (2020, methodology): “Comparisons with existing methods on ImageCLEF-DA [1] using ResNet-50 [7] as the base network are reported in Table B, where results of existing methods are quoted from their respective papers or the work of [13, 16].”
Adversarial Domain Adaptation with Domain Mixup (2019, methodology): “Other works optimize the performance on target domain by capturing complex multimode structures (Pei et al. 2018; Long et al. 2018), exploring task-specific decision boundaries (Saito et al.”
Domain Aggregation Networks for Multi-Source Domain Adaptation (2019, background): “, 2018b), class-dependent adversarial domain classifier (Pei et al., 2018) and Known Unknown Discrimination (Schoenauer-Sebag et al.”
Domain-Symmetric Networks for Adversarial Domain Adaptation (2019, methodology): “6 MADA [18] 90.”
Deep Subdomain Adaptation Network for Image Classification (2020, methodology): “Second, comparing DSAN with MADA [15] and CDAN [14], DSAN also takes less time to converge.”
Reliable Weighted Optimal Transport for Unsupervised Domain Adaptation (2020, methodology): “Later, several extensions were proposed, such as Multi-Adversarial Domain Adaption (MADA) [28], and Conditional Domain Adversarial Networks (CDANs) [22] enable the alignment of multi-modal distributions.”
Attending to Discriminative Certainty for Domain Adaptation (2019, methodology): “We follow similar setting in [47, 12, 34] and plot t-SNE embeddings of the dataset in the Figure 3.”
On Learning Invariant Representations for Domain Adaptation (2019, background): “Theorem 2.1 is the foundation of many recent works on unsupervised domain adaptation via learning invariant representations (Ajakan et al., 2014; Ganin et al., 2016; Zhao et al., 2018b; Pei et al., 2018; Zhao et al., 2018a).”