Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Contrastive Adapters for Foundation Model Group Robustness
Michael Zhang, Christopher Re
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 |
|---|---|---|---|
| Replace zero-shot inference with contrastive adapter training | CLIP ResNet-50 · Waterbirds | worst-group accuracy: 36.6 -> 83.7 | Improved |
| Replace zero-shot inference with contrastive adapter training | CLIP ResNet-50 · CelebA | worst-group accuracy: 74.0 -> 90.0 | Improved |
| Replace zero-shot inference with contrastive adapter training | CLIP ResNet-50 · BREEDS Living-17 | worst-group accuracy: 6.0 -> 62.0 | Improved |
| Replace zero-shot inference with contrastive adapter training | CLIP ResNet-50 · CIFAR-10.02 | worst-group accuracy: 39.1 -> 60.7 | Improved |
| Train an ERM adapter instead of using zero-shot inference | CLIP ResNet-50 · CelebA | worst-group accuracy: 74.0 -> 36.1 | Hurt |
| Train an ERM adapter instead of using zero-shot inference | CLIP ResNet-50 · Waterbirds | worst-group accuracy: 36.6 -> 60.8 | Improved |
| Train an ERM adapter instead of using zero-shot inference | CLIP ResNet-50 · BREEDS Living-17 | worst-group accuracy: 6.0 -> 70.7 | Improved |
| Train an ERM adapter instead of using zero-shot inference | CLIP ResNet-50 · CIFAR-10.02 | worst-group accuracy: 39.1 -> 68.8 | Improved |
| Use contrastive adapter training rather than ERM adapter training | CLIP ViT-L/14 · CelebA | worst-group accuracy: 36.7 -> 84.6 | Improved |
| Train a linear probe instead of using zero-shot inference | CLIP ResNet-50 · CelebA | worst-group accuracy: 74.0 -> 11.9 | Hurt |
| Add a batch normalization layer between adapter input and output layers (ablation) | Adapter · Not specified | group robustness | Improved |
| Add a residual connection to the adapter (ablation) | Adapter · Not specified | group robustness | No clear effect |
Closest papers by meaning
- Prompting is a Double-Edged Sword: Improving Worst-Group Robustness of Foundation Models (ICML 2024 · similarity 0.87)
- The Group Robustness is in the Details: Revisiting Finetuning under Spurious Correlations (NeurIPS 2024 · similarity 0.87)
- LR0.FM: Low-Res Benchmark and Improving robustness for Zero-Shot Classification in Foundation Models (ICLR 2025 · similarity 0.87)
- Project-Probe-Aggregate: Efficient Fine-Tuning for Group Robustness (CVPR 2025 · similarity 0.86)
- Foundation Model-oriented Robustness: Robust Image Model Evaluation with Pretrained Models (ICLR 2024 · similarity 0.86)
- Rank-Constrained Adaptation for Reliable Real-World Performance (NeurIPS 2026 · similarity 0.86)
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.
