Research map / Label-efficient and robust learning / Domain adaptation and generalization
Domain-Specific Batch Normalization for Unsupervised Domain Adaptation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace shared BN layers with domain-specific BN in both training stages | MSTN and CPUA · VisDA-C validation | accuracy (Avg, %): 65.0 -> 80.2; 66.6 -> 76.2 | Helped |
| Add DSBN in both stages instead of BN in both stages (ablation) | MSTN · VisDA-C validation | accuracy (Avg, %): 63.4 -> 80.2 | Helped |
| Add DSBN in Stage 1 while retaining BN in Stage 2 (ablation) | MSTN · VisDA-C validation | accuracy (Avg, %): 63.4 -> 71.1 | Helped |
| Add DSBN in Stage 2 while retaining BN in Stage 1 (ablation) | MSTN · VisDA-C validation | accuracy (Avg, %): 63.4 -> 73.2 | Helped |
| Add DSBN in both stages instead of BN in both stages (ablation) | CPUA · VisDA-C validation | accuracy (Avg, %): 63.4 -> 76.2 | Helped |
| Add DSBN in Stage 1 while retaining BN in Stage 2 (ablation) | CPUA · VisDA-C validation | accuracy (Avg, %): 63.4 -> 69.4 | Helped |
| Add DSBN in Stage 2 while retaining BN in Stage 1 (ablation) | CPUA · VisDA-C validation | accuracy (Avg, %): 63.4 -> 71.3 | Helped |
| Add Stage 2 DSBN training after Stage 1 DSBN pseudo-labeling (ablation) | MSTN · VisDA-C validation | accuracy gain from Stage 1 to Stage 2 (percentage points): +7.9 | Helped |
| Use BN rather than DSBN in Stage 2 (ablation) | MSTN · VisDA-C validation | accuracy gain from Stage 1 to Stage 2 (percentage points): -1.2 | Hurt |
| Add Stage 2 DSBN training after Stage 1 DSBN pseudo-labeling (ablation) | CPUA · VisDA-C validation | accuracy gain from Stage 1 to Stage 2 (percentage points): +4.3 | Helped |
| Use BN rather than DSBN in Stage 2 (ablation) | CPUA · VisDA-C validation | accuracy gain from Stage 1 to Stage 2 (percentage points): -2.6 | Hurt |
| Repeat Stage 2 training and update target pseudo-labels from the preceding iteration (ablation) | MSTN with DSBN · VisDA-C | accuracy (%): 80.2 -> 81.4 -> 82.2 -> 82.7 | Helped |
Closest papers by meaning
- Reducing Divergence in Batch Normalization for Domain Adaptation (AAAI 2025 · similarity 0.92)
- Cross-Domain Collaborative Normalization via Structural Knowledge (AAAI 2022 · similarity 0.92)
- Test-Time Domain Adaptation by Learning Domain-Aware Batch Normalization (AAAI 2024 · similarity 0.91)
- Enhancing Domain Adaptation through Prompt Gradient Alignment (NeurIPS 2024 · similarity 0.87)
- Leveraging Normalization Layer in Adapters with Progressive Learning and Adaptive Distillation for Cross-Domain Few-Shot Learning (AAAI 2024 · similarity 0.87)
- Domain-Aware Fine-Tuning: Enhancing Neural Network Adaptability (AAAI 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.