Research map / Label-efficient and robust learning / Domain adaptation and generalization
Progressive Distribution Bridging: Unsupervised Adaptation for Large-scale Pre-trained Models via Adaptive Auxiliary Data
Weinan He, Yixin Zhang, Zilei Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add progressive auxiliary data (alpha 0.1, 0.4, 0.7) in order vs reversed order (ablation) | PDB · DomainNet | avg accuracy (%): 62.82 -> 63.10 | Improved |
| use progressive styles instead of target-style data only (ablation) | PDB · DomainNet | avg accuracy (%): 62.80 -> 63.10 | Improved |
| add retrieved auxiliary data to target-only training (ablation) | PDB · DomainNet | avg accuracy (%): 62.21 -> 63.10 | Improved |
| replace labeled UDA source domain with retrieved adaptive auxiliary data (ablation) | PDB · DomainNet | avg accuracy (%): 62.80 -> 63.10 | Improved |
| add Style filter to T2T retrieval (ablation) | cascaded semantic filters · DomainNet (alpha=0.1) | avg accuracy (%): 60.80 -> 61.09 | Improved |
| combine Style, CM and GMM filters (ablation) | cascaded semantic filters · DomainNet (alpha=0.1) | avg accuracy (%): 60.80 -> 61.53 | Improved |
| add a fourth alpha step (1.0) (ablation) | PDB · DomainNet (->Q) | accuracy (%): 22.82 -> 22.89 | No clear effect |
| increase auxiliary samples per category beyond Nmax = 100 (ablation) | PDB · DomainNet | accuracy | No clear effect |
| use multi-step progressive alpha vs single alpha 0.1 (ablation) | PDB · DomainNet (->Q) | accuracy (%): 21.10 -> 22.82 | Improved |
| use PDB instead of UFT POUF-FT | CLIP · Office-Home | avg accuracy (%): 86.1 -> 87.8 | Improved |
| use PDB instead of labeled-source UDA methods | CLIP · DomainNet | avg accuracy (%): 63.7 -> 63.1 | No clear effect |
| use PDB instead of UFT POUF-FT on poor-initial-performance domain | CLIP · DomainNet (->Q) | accuracy (%): 18.8 -> 22.8 | Improved |
Closest papers by meaning
- A Closer Look at the Few-Shot Adaptation of Large Vision-Language Models (CVPR 2024 · similarity 0.88)
- Prompt-Based Distribution Alignment for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.88)
- Source-Free Domain Adaptation with Frozen Multimodal Foundation Model (CVPR 2024 · similarity 0.88)
- Preserving Clusters in Prompt Learning for Unsupervised Domain Adaptation (CVPR 2025 · similarity 0.88)
- ProKeR: A Kernel Perspective on Few-Shot Adaptation of Large Vision-Language Models (CVPR 2025 · similarity 0.87)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
