Research map / Label-efficient and robust learning / Domain adaptation and generalization
Multi-Prompt Alignment for Multi-Source Unsupervised Domain Adaptation
Haoran Chen, Xintong Han, Zuxuan Wu, Yu-Gang Jiang
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 |
|---|---|---|---|
| Use pair-specific prompts, denoise them with an autoencoder, and align reconstructed prompts | MPA · ImageCLEF Avg | top-1 accuracy (%): 89.4 -> 91.7 | Improved |
| Use pair-specific prompts, denoise them with an autoencoder, and align reconstructed prompts | MPA · Office-Home Avg | top-1 accuracy (%): 74.1 -> 75.4 | Improved |
| Use pair-specific prompts, denoise them with an autoencoder, and align reconstructed prompts | MPA · DomainNet Avg | top-1 accuracy (%): 52.0 -> 54.1 | Improved |
| Tune the learned latent subspace instead of retraining MPA for a later target | MPA continuous adaptation · DomainNet Avg, continuous adaptation | top-1 accuracy (%): 53.9 -> 53.6 | Hurt |
| Replace MFSAN's ImageNet-pretrained backbone with CLIP's image encoder (ablation) | MFSAN · ImageCLEF Avg | top-1 accuracy (%): 89.4 -> 89.1 | Hurt |
| Use domain-specific manual prompt templates instead of the simple class-name prompt (ablation) | MPA · Office-Home Avg | top-1 accuracy (%): 75.4 -> 74.1; 75.4 -> 75.2 | Hurt |
| Raise the pseudo-label confidence threshold from 0.3 to 0.8 (ablation) | MPA · Office-Home Avg | top-1 accuracy (%): 75.4 -> 74.7 | Hurt |
| Increase prompt token lengths from M1=M2=8 to M1=M2=20 (ablation) | MPA · Office-Home Avg | top-1 accuracy (%): 75.1 -> 75.6 | Improved |
| Remove L1 prompt-alignment loss (ablation) | MPA · Office-Home Avg | top-1 accuracy (%): 75.4 -> 74.7 | Hurt |
| Remove L1 prompt-alignment loss (ablation) | MPA · Office-Home Clipart | top-1 accuracy (%): 54.9 -> 53.4 | Hurt |
| Remove autoencoder reconstruction loss (ablation) | MPA · Office-Home Avg | top-1 accuracy (%): 75.4 -> 74.9 | Hurt |
| Remove domain-specific prompt tokens (ablation) | MPA · Office-Home Avg | top-1 accuracy (%): 75.4 -> 74.4 | Hurt |
Closest papers by meaning
- Enhancing Domain Adaptation through Prompt Gradient Alignment (NeurIPS 2024 · similarity 0.93)
- Prompt-Based Distribution Alignment for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.93)
- Moment Matching for Multi-Source Domain Adaptation (ICCV 2019 · similarity 0.91)
- Co-regularized Alignment for Unsupervised Domain Adaptation (NeurIPS 2018 · similarity 0.90)
- Domain-Agnostic Mutual Prompting for Unsupervised Domain Adaptation (CVPR 2024 · similarity 0.89)
- Preserving Clusters in Prompt Learning for Unsupervised Domain Adaptation (CVPR 2025 · similarity 0.89)
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.
