Research map / Label-efficient and robust learning / Recognition and long-tail
Confidence-Aware Learning for Deep Neural Networks
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 |
|---|---|---|---|
| Add Correctness Ranking Loss to the classifier | DenseNet · CIFAR-100 | accuracy: 75.39±0.29 -> 76.82±0.26 | Helped |
| Add Correctness Ranking Loss to the classifier | DenseNet · CIFAR-10 | AURC: 5.15±0.35 -> 4.92±0.14 | Helped |
| Add Correctness Ranking Loss to the classifier | DenseNet · CIFAR-100 | AURC: 71.75±0.89 -> 61.77±1.07 | Helped |
| Add Correctness Ranking Loss to the classifier | ResNet110 · SVHN OOD detection with TinyImageNet out-of-distribution data | FPR-95%TPR: 29.65±2.40 -> 5.89±0.70 | Helped |
| Use CRL confidence estimates for least-confidence active-learning sampling | ResNet18 · CIFAR-10, active-learning stage 2 | accuracy: 80.40% -> 80.6% | Helped |
| Use CRL confidence estimates for least-confidence active-learning sampling | ResNet18 · CIFAR-100 active learning | accuracy: 70.4% and 0.43% accuracy gap | Helped |
| Use negative entropy rather than maximum class probability as the CRL confidence function (ablation) | CRL · CIFAR-10, ResNet110 | accuracy: 94.11±0.20 -> 94.24±0.11 | Helped |
| Use margin rather than maximum class probability as the CRL confidence function (ablation) | CRL · CIFAR-10, ResNet110 | accuracy: 94.11±0.20 -> 93.83±0.10 | Hurt |
| Use negative entropy rather than maximum class probability as the CRL confidence function (ablation) | CRL · CIFAR-10, DenseNet | AURC: 4.92±0.14 -> 4.95±0.30 | Hurt |
| Use margin rather than maximum class probability as the CRL confidence function (ablation) | CRL · CIFAR-10, DenseNet | AURC: 4.92±0.14 -> 5.26±0.23 | Hurt |
| Use negative entropy rather than maximum class probability as the CRL confidence function (ablation) | CRL · CIFAR-100, VGG-16 | AURC: 77.33±1.15 -> 70.19±1.53 | Helped |
| Use margin rather than maximum class probability as the CRL confidence function (ablation) | CRL · CIFAR-100, VGG-16 | AURC: 77.33±1.15 -> 75.91±0.76 | Helped |
Closest papers by meaning
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.87)
- Learning for Single-Shot Confidence Calibration in Deep Neural Networks Through Stochastic Inferences (CVPR 2019 · similarity 0.86)
- Towards Trustworthy Deep Learning (AAAI 2024 · similarity 0.86)
- Pointwise Information Measures as Confidence Estimators in Deep Neural Networks: A Comparative Study (ICML 2025 · similarity 0.85)
- VNN: Verification-Friendly Neural Networks with Hard Robustness Guarantees (ICML 2024 · similarity 0.85)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.85)
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.