검색 상세

Bridging Domain Spaces via Vicinal Space for Unsupervised Domain Adaptation

  • 나재민 (JAEMIN NA, 아주대학교 일반대학원)

초록/요약

In this thesis, the research focuses on leveraging domain spaces for unsupervised domain adaptation (UDA). The primary objective is to explore approaches that effectively utilize the intermediate spaces between the source and target domains. The goal is to overcome the limitations of traditional direct domain adaptation methods, which are inherently constrained in their ability to handle large discrepancies between domains. The first chapter introduces a novel approach that constructs intermediate domain spaces with distinct characteristics using fixed ratio-based mixup. To enhance domain invariant representations, we incorporate confidence-based learning techniques, including bidirectional matching and self-penalization. The effectiveness of each component is demonstrated through thorough analysis, while competitive performance is observed on three standard benchmarks compared to other UDA methods. In the second chapter, we present a more advanced method tailored to bridging domains while considering the uncertainty of model predictions. We extend the fixed-ratio-based mixup to operate at the feature level, adaptively determining the layer for mixup based on prediction uncertainty. Furthermore, we enhance our complementary learning by adjusting augmentation intensity using an adaptive confidence threshold. Extensive experiments validate the superiority of our proposed methods across public benchmarks, including single- source and multi-source scenarios. The final chapter sheds light on the problem of equilibrium collapse, where source labels dominate over target labels in the predictions of the vicinal space. To address this issue, we propose an instance-wise minimax strategy that minimizes the entropy of highly uncertain instances in the vicinal space. We divide the vicinal space into two subspaces and mitigate inter-domain discrepancy by minimizing their distance. Thorough ablation studies provide insights into the proposed method, demonstrating comparable performance to state-of-the-art approaches in standard unsupervised domain adaptation benchmarks. Overall, this thesis offers groundbreaking insights and approaches that leverage domain spaces for unsupervised domain adaptation, leading to significant advancements in the field. The proposed approaches are not only effective but also highly competitive, as demonstrated through comprehensive evaluations across diverse benchmarks and scenarios. The findings contribute valuable knowledge to the field of unsupervised domain adaptation, offering new perspectives and techniques to address the challenges associated with domain gaps. By demonstrating the effectiveness and competitiveness of the proposed approaches, this thesis paves the way for further advancements in unsupervised domain adaptation research. Keywords: Unsupervised domain adaptation, single/multi-source domain adaptation, deep neural network, transfer learning.

more

목차

1 Introduction 1
1.1 Thesis Outline 2
2 Bridging Domains via Vicinal Space 4
2.1 Overview 4
2.2 Background 7
2.2.1 Semi-supervised Learning 7
2.2.2 Unsupervised Domain Adaptation 8
2.3 Methodology 9
2.3.1 Fixed Ratio-based Mixup 10
2.3.2 Confidence-based Learning 11
2.3.3 Consistency Regularization 12
2.3.4 Training Procedure 13
2.4 Experiment 14
2.4.1 Setups 14
2.4.2 Ablation studies and discussions 15
2.4.3 Comparison with the state-of-the-art methods 20
2.5 Discussion 21
3 Uncertainty Calibration for Domain Bridging 22
3.1 Overview 22
3.2 Background 26
3.2.1 Semi-supervised Learning 26
3.2.2 Unsupervised Domain Adaptation 27
3.2.3 Uncertainty-based Methods 29
3.3 Methodology 29
3.3.1 Fixed Ratio-based Mixup 30
3.3.2 Confidence-based Learning 32
3.3.3 Uncertainty-aware Learning 33
3.3.4 Bidirectional Fixed-Matching 34
3.3.5 Consistency Regularization 35
3.3.6 Training Procedure 35
3.4 Experiment 36
3.4.1 Setups 37
3.4.2 Ablation studies and discussions 38
3.4.3 Comparison with the state-of-the-art methods 44
3.5 Discussion 47
4 Contrastive and Consensus Vicinal Space 49
4.1 Overview 49
4.2 Background 52
4.2.1 Unsupervised domain adaptation 52
4.2.2 Mixup augmentation 53
4.2.3 Consistency training 53
4.3 Methodology 54
4.3.1 Preliminaries 54
4.3.2 EMP-Mixup 55
4.3.3 Contrastive Views and Labels 57
4.3.4 Label Consensus 59
4.4 Experiment 60
4.4.1 Experimental Setups 61
4.4.2 Comparison with the State-of-the-Art Methods 61
4.4.3 Ablation Studies and Discussions 63
4.5 Discussion 66
5 Conclusion and Discussion 68
5.1 Future Directions 68
Bibliography 70

more
반출 Meta View 목록