Research map / Image generation / Fast diffusion sampling
DDB: Diffusion Driven Balancing to Address Spurious Correlations
Aryan Yazdan Parast, Basim Azam, Naveed Akhtar
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 pruning of generated samples (ablation) | DDB · Waterbirds | worst-group accuracy: 91.28 -> 93.0 | Improved |
| add pruning of generated samples (ablation) | DDB · CelebA | worst-group accuracy: 81.7 -> 85.8 | Improved |
| add pruning of generated samples (ablation) | DDB · MetaShift | worst-group accuracy: 80.6 -> 81.2 | Improved |
| add textual inversion with 30 samples (ablation) | DDB · Waterbirds | worst-group accuracy: 88.9 -> 93 | Improved |
| add textual inversion with 10 samples (ablation) | DDB · Waterbirds | worst-group accuracy: 88.9 -> 90.5 | Improved |
| increase textual inversion samples from 30 to 40 (ablation) | DDB · Waterbirds | worst-group accuracy: 93 -> 90.3 | Hurt |
| textual inversion reduces number of pruned samples (0 to 30 samples) (ablation) | DDB · Waterbirds | pruned samples: 1018 -> 737 | Improved |
| add 2000 NonBlond Female instead of 1000 (with 2000 Blond Male) (ablation) | DDB · CelebA | worst-group accuracy: 85.6 -> 87.1 | Improved |
| add 3000 Blond Male instead of 2000 (with 1000 NonBlond Female) (ablation) | DDB · CelebA | worst-group accuracy: 85.6 -> 86.6 | Improved |
| add 4000 Blond Male instead of 3000 (with 2000 NonBlond Female) (ablation) | DDB · CelebA | worst-group accuracy: 86.8 -> 86.7 | No clear effect |
| retrain on DDB augmented data instead of plain ERM | ERM ResNet-50 · Waterbirds | worst-group accuracy: 74.6 -> 93.0 | Improved |
| retrain on DDB augmented data instead of plain ERM | ERM ResNet-50 · CelebA | average accuracy: 96.0 -> 87.3 | Hurt |
Closest papers by meaning
- Directional Label Diffusion Model for Learning from Noisy Labels (CVPR 2025 · similarity 0.87)
- Slight Corruption in Pre-training Data Makes Better Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Taming Diffusion for Dataset Distillation with High Representativeness (ICML 2025 · similarity 0.87)
- An Efficient Framework for Enhancing Discriminative Models via Diffusion Techniques (AAAI 2025 · similarity 0.87)
- When Model Knowledge meets Diffusion Model: Diffusion-assisted Data-free Image Synthesis with Alignment of Domain and Class (ICML 2025 · 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.
