Research map / Adversarial robustness and security / Certified and empirical robustness
Benchmarking Neural Network Robustness to Common Corruptions and Perturbations
Read the paperCheck 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 |
|---|---|---|---|
| Fine-tune the full ResNet-50 for five epochs on CLAHE-processed images | ResNet-50 · ImageNet-C | mCE: 76.7% to 74.5% | Helped |
| Use a multiscale architecture | ImageNet classifiers · ImageNet-C | mCE: 76.7% to 73.6% (MSDNet); 76.7% to 73.3% (Multigrid) | Helped |
| Use feature aggregation instead of a ResNet-50 | ImageNet classifiers · ImageNet-C | mCE: 76.7% to 73.4% (DenseNet-121); 76.7% to 68.2% (ResNeXt-50) | Helped |
| Use a larger feature-aggregating model | DenseNet-121 and ResNeXt-50 · ImageNet-C | mCE: 73.4% to 66.4% (DenseNet); 68.2% to 62.2% (ResNeXt) | Helped |
| Use feature aggregation instead of a ResNet-50 | ImageNet classifiers · ImageNet-P | mFR: 58.0% to 56.4% (DenseNet-121); 58.0% to 52.4% (ResNeXt-50) | Helped |
| Use feature aggregation instead of a ResNet-50 | ImageNet classifiers · ImageNet-P | mT5D: 78.3% to 76.8% (DenseNet-121); 78.3% to 74.2% (ResNeXt-50) | Helped |
| Use a larger feature-aggregating model | DenseNet-121 and ResNeXt-50 · ImageNet-P | mFR: 56.4% to 46.9% (DenseNet); 52.4% to 43.2% (ResNeXt) | Helped |
| Use a larger feature-aggregating model | DenseNet-121 and ResNeXt-50 · ImageNet-P | mT5D: 76.8% to 69.5% (DenseNet); 74.2% to 65.9% (ResNeXt) | Helped |
| Train on typical and stylized ImageNet images | ResNet-50 · ImageNet-C | mCE: 76.7% to 69.3% | Helped |
| Fine-tune with Adversarial Logit Pairing | ResNet-50 · Tiny ImageNet-P | relative mFP: 41% relative decrease | Helped |
| Fine-tune with stability training | ResNet-50 · ImageNet-C | mCE | Hurt |
| Fine-tune with stability training | ResNet-50 · ImageNet-P validation | mFR: 58% to 57% | Helped |
Closest papers by meaning
- Benchmarking Adversarial Robustness on Image Classification (CVPR 2020 · similarity 0.92)
- ImageNet-E: Benchmarking Neural Network Robustness via Attribute Editing (CVPR 2023 · similarity 0.91)
- RobustBench: a standardized adversarial robustness benchmark (NeurIPS 2021 · similarity 0.90)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.89)
- Improving robustness to corruptions with multiplicative weight perturbations (NeurIPS 2024 · similarity 0.88)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.88)
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.