Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Effective Robustness against Natural Distribution Shifts for Models with Different Training Data
Zhouxing Shi, Nicholas Carlini, Ananth Balashankar, Ludwig Schmidt, Cho-Jui Hsieh, Alex Beutel, Yao Qin
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 two ID test sets (multi-ID) instead of one | effective robustness fit · CIFAR-10.2 | R^2: 0.981 -> 0.996 | Improved |
| use two ID test sets instead of one | effective robustness fit · CIFAR-10.2 | MAE (%): 2.22 -> 0.95 | Improved |
| use two ID test sets instead of one | ImageNet models · CIFAR-10.2 | effective robustness (%): 3.91 -> 0.56 | Improved |
| use two ID test sets instead of one | fit for LAION vs ImageNet · ImageNet-R | R^2: 0.216 -> 0.982 | Improved |
| use two ID test sets instead of one | LAION models · ImageNet-like OOD average | effective robustness (%): 5.96 -> -0.00 | Improved |
| use two ID test sets instead of one | YFCC models · ImageNet-like OOD average | effective robustness (%): 2.59 -> 0.77 | Improved |
| switch the single ID test set from ImageNet to YFCC | single-ID effective robustness · YFCC vs ImageNet models | effective robustness | Mixed |
| use two ID test sets instead of one | ImageNet models (37) · ImageNet-R | effective robustness (%): -9.45 -> -0.54 | Improved |
| include mixed-data (ImageNet+YFCC) models in the baseline fit (ablation) | multi-ID fit · ObjectNet | MAE (%): 1.73 -> 0.95 | Improved |
| include mixed-data (ImageNet+LAION) models in the baseline fit (ablation) | multi-ID fit · ImageNet-Sketch | MAE (%): 2.62 -> 2.03 | Improved |
| use two ID test sets instead of one | OpenCLIP LAION · ImageNet-like OOD average | MAE (%): 22.59 -> 2.63 | Improved |
| use two ID test sets instead of one | SLIP YFCC · ImageNet-like OOD average | effective robustness (%): 10.56 -> 3.19 | Mixed |
Closest papers by meaning
- Foundation Model-oriented Robustness: Robust Image Model Evaluation with Pretrained Models (ICLR 2024 · similarity 0.87)
- OODRobustBench: a Benchmark and Large-Scale Analysis of Adversarial Robustness under Distribution Shift (ICML 2024 · similarity 0.87)
- Robust Fine-tuning of Zero-shot Models via Variance Reduction (NeurIPS 2024 · similarity 0.87)
- LAION-C: An Out-of-Distribution Benchmark for Web-Scale Vision Models (ICML 2025 · similarity 0.86)
- Unexplored Faces of Robustness and Out-of-Distribution: Covariate Shifts in Environment and Sensor Domains (CVPR 2024 · similarity 0.86)
- Aggregation Hides Out-of-Distribution Generalization Failures from Spurious Correlations (NeurIPS 2025 · 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.
