Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

More is Better: Deep Domain Adaptation with Multiple Sources

Sicheng Zhao, Hui Chen, Hu Huang, Pengfei Xu, Guiguang Ding

TL;DR

By modeling multiple sources with distributions $p_i(\mathbf{x},\mathbf{y})$ and a target distribution $P_T(\mathbf{x},\mathbf{y})$, the survey organizes deep MDA methods into latent space transformation, intermediate domain generation, and task classifier refinement. It discusses matching strategies and special settings such as federated and source-free MDA, providing datasets, benchmarks, and practical guidelines. Empirically, domain alignment improves target accuracy over source-only baselines across multiple datasets, though substantial gaps to oracle performance remain, indicating room for improvement. The work offers a roadmap for designing robust MDA systems and outlines promising directions such as multi-modal MDA, test-time adaptation, and theory-informed analysis.

Abstract

In many practical applications, it is often difficult and expensive to obtain large-scale labeled data to train state-of-the-art deep neural networks. Therefore, transferring the learned knowledge from a separate, labeled source domain to an unlabeled or sparsely labeled target domain becomes an appealing alternative. However, direct transfer often results in significant performance decay due to domain shift. Domain adaptation (DA) aims to address this problem by aligning the distributions between the source and target domains. Multi-source domain adaptation (MDA) is a powerful and practical extension in which the labeled data may be collected from multiple sources with different distributions. In this survey, we first define various MDA strategies. Then we systematically summarize and compare modern MDA methods in the deep learning era from different perspectives, followed by commonly used datasets and a brief benchmark. Finally, we discuss future research directions for MDA that are worth investigating.

More is Better: Deep Domain Adaptation with Multiple Sources

TL;DR

By modeling multiple sources with distributions and a target distribution , the survey organizes deep MDA methods into latent space transformation, intermediate domain generation, and task classifier refinement. It discusses matching strategies and special settings such as federated and source-free MDA, providing datasets, benchmarks, and practical guidelines. Empirically, domain alignment improves target accuracy over source-only baselines across multiple datasets, though substantial gaps to oracle performance remain, indicating room for improvement. The work offers a roadmap for designing robust MDA systems and outlines promising directions such as multi-modal MDA, test-time adaptation, and theory-informed analysis.

Abstract

In many practical applications, it is often difficult and expensive to obtain large-scale labeled data to train state-of-the-art deep neural networks. Therefore, transferring the learned knowledge from a separate, labeled source domain to an unlabeled or sparsely labeled target domain becomes an appealing alternative. However, direct transfer often results in significant performance decay due to domain shift. Domain adaptation (DA) aims to address this problem by aligning the distributions between the source and target domains. Multi-source domain adaptation (MDA) is a powerful and practical extension in which the labeled data may be collected from multiple sources with different distributions. In this survey, we first define various MDA strategies. Then we systematically summarize and compare modern MDA methods in the deep learning era from different perspectives, followed by commonly used datasets and a brief benchmark. Finally, we discuss future research directions for MDA that are worth investigating.
Paper Structure (14 sections, 4 figures, 1 table)

This paper contains 14 sections, 4 figures, 1 table.

Table of Contents

  1. Background and Motivation
  2. Problem Definition
  3. Deep MDA Taxonomy
  4. MDA Alignment Methodologies
  5. Latent Space Transformation
  6. Intermediate Domain Generation
  7. Task Classifier Refinement
  8. MDA Matching Strategies
  9. Domain Pairing Strategy
  10. Feature Extractors' Weight Sharing Strategy
  11. Domain/Sample Weighting Strategy
  12. Other Specific MDA Settings
  13. Applications, Datasets, and Benchmark
  14. Conclusion and Future Directions

Figures (4)

  • Figure 1: Examples of domain shift in the single-source scenario. The models trained on the labeled source domain do not perform well when directly transferring to the target domain.
  • Figure 2: Examples of domain shift in the multi-source scenario. Simply combining multiple source domains into one source and directly performing SDA does not guarantee better performance compared to just using the best individual source.
  • Figure 3: Taxonomy of deep MDA methods.
  • Figure 4: A widely employed MDA framework. The solid arrows, dashed dot arrows, and bold arrows indicate the training of latent space transformation, intermediate domain generation, and task classifier refinement, respectively. The dashed arrows indicate the reference process.