Research map / Adversarial robustness and security / Certified and empirical robustness
Human Uncertainty Makes Classification More Robust
Joshua C. Peterson, Ruairidh M. Battleday, Thomas L. Griffiths, Olga Russakovsky
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 |
|---|---|---|---|
| train with human soft labels instead of one-hot | CNN classifiers · CIFAR10 | accuracy: 83.5 -> 84.5 | Improved |
| train with human soft labels instead of one-hot | CNN classifiers · ImageNet-Far | accuracy: 49.4 -> 51.4 | Improved |
| train with human soft labels instead of one-hot | CNN classifiers · CIFAR10 | crossentropy: 0.7 -> 0.5 | Improved |
| train with human soft labels instead of one-hot | CNN classifiers · ImageNet-Far | crossentropy: 2.9 -> 1.8 | Improved |
| train with human soft labels instead of one-hot | CNN classifiers · CIFAR10H holdout | second-best accuracy: 5% better on average | Improved |
| use image-level human soft labels instead of class-level soft labels (ablation) | ResNet fine-tuned on CIFAR10H · CIFAR10H holdout | crossentropy: 0.42 -> 0.35 | Improved |
| use image-level human soft labels instead of mixup (ablation) | PyramidNet fine-tuned · CIFAR10H holdout | crossentropy: 0.49 -> 0.28 | Improved |
| fine-tune with human soft labels instead of ground-truth hard labels (ablation) | Shake-Shake · CIFAR10H holdout | crossentropy: 0.51 -> 0.26 | Improved |
| sample hard labels from human distribution instead of using soft targets (ablation) | ResNet preact · CIFAR10.1v4 | crossentropy: 0.49 -> 0.42 | Improved |
| sample hard labels from human distribution instead of using soft targets (ablation) | ResNet · CIFAR10H holdout | crossentropy: 0.35 -> 0.35 | No clear effect |
| fine-tune with human soft labels | VGG under FGSM attack · CIFAR10 test | crossentropy: 7.9 -> 4.1 | Improved |
| fine-tune with human soft labels | eight CNNs under FGSM attack · CIFAR10 test | accuracy: VGG 7% -> 8%, PyramidNet 22% -> 19% | Mixed |
Closest papers by meaning
- Using Pre-Training Can Improve Model Robustness and Uncertainty (ICML 2019 · similarity 0.88)
- Robustness Meets Uncertainty: Evidential Adversarial Training for Robust Selective Classification (ECCV 2026 · similarity 0.88)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.87)
- Adversarial Robustness Limits via Scaling-Law and Human-Alignment Studies (ICML 2024 · similarity 0.86)
- Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement (CVPR 2024 · similarity 0.86)
- Doppelgangers and Adversarial Vulnerability (CVPR 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.
