Abstract:<p>Learning a cross-domain representation from labeled source domains to unlabeled target domains is an important research problem in representation learning. Despite the success of traditional adversarial methods, they proposed to align features from each domain only while neglecting the importance of labels, when fooling a special domain discriminator network. Thus, the discriminator of these approaches merely distinguishes whether the generated features are in-domain or not, which may lead to less class-discriminative features. In this paper, by considering the joint distributions of features and labels in both domains, we present Feature and Label Adversarial Networks (FLAN). As a result, FLAN can generate more discriminative features in both domains. Experimental results on standard unsupervised domain adaptation benchmarks have demonstrated that FLAN can outperform the state-of-art domain invariant representation learning methods.</p>

What problem does this paper attempt to address?

The problem that this paper attempts to solve is in unsupervised domain adaptation, how to generate more discriminative features by simultaneously considering the joint distribution of features and labels, thereby reducing the data distribution difference between the source domain and the target domain. Traditional adversarial methods mainly focus on aligning the features of each domain while ignoring the importance of labels, which may lead to the generated features being not discriminative enough in classification. Therefore, this paper proposes Feature and Label Adversarial Networks (FLAN), aiming to improve the classification performance on the target domain by using the joint distribution information of features and labels. Specifically, the main contributions of FLAN include: 1. **Proposing a generative adversarial framework**: Directly learn the shared feature embedding while considering feature and label information. The key improvement is that the discriminator makes full use of feature and label information, so that rich information can be transmitted to the generator to learn more discriminative features. 2. **Exploring pseudo - labels of the target domain**: Although there is no labeled data in the target domain, by introducing entropy minimization, the classifier is encouraged to better fit the specific structure of the target domain, thereby generating high - quality pseudo - labels. 3. **Efficient end - to - end solution**: The entire framework can be efficiently solved by back - propagation. Experimental results show that FLAN outperforms the existing state - of - the - art algorithms on three domain - adaptation datasets. Through these improvements, FLAN can more effectively reduce the data distribution difference between the source domain and the target domain in unsupervised domain adaptation tasks, thereby improving the classification performance on the target domain.

Domain adaptation with feature and label adversarial networks

Federated Domain Adaptation via Pseudo-label Refinement

Trust-aware Conditional Adversarial Domain Adaptation with Feature Norm Alignment.

Joint Domain Alignment and Discriminative Feature Learning for Unsupervised Deep Domain Adaptation

Joint Feature and Labeling Function Adaptation for Unsupervised Domain Adaptation

Domain-Invariant Adversarial Learning for Unsupervised Domain Adaption.

Hierarchical Domain Adaptation with Local Feature Patterns

Cross-domain feature enhancement for unsupervised domain adaptation

A Label Proportions Estimation Technique for Adversarial Domain Adaptation in Text Classification

Adversarial-Learned Loss for Domain Adaptation

Domain-Adversarial Training of Neural Networks

AFAN: Augmented Feature Alignment Network for Cross-Domain Object Detection

Deep Joint Discriminative Feature Learning and Class-Aware Domain Alignment for Unsupervised Domain Adaptation

Domain Adaptation Using Pseudo Labels

Multiple Source Domain Adaptation with Adversarial Training of Neural Networks

Partial Adversarial Domain Adaptation

Discriminative Partial Domain Adversarial Network

A New Method of Image Classification Based on Domain Adaptation

Domain Adaptation for Remote Sensing Image Semantic Segmentation: An Integrated Approach of Contrastive Learning and Adversarial Learning

Learning explicitly transferable representations for domain adaptation

Domain generalization based on domain-specific adversarial learning