Research map / Label-efficient and robust learning / OOD and anomaly detection
Revisiting Confidence Estimation: Towards Reliable Failure Prediction
Fei Zhu, Xu-Yao Zhang, Zhen Cheng, Cheng‐Lin Liu
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 mixup | ResNet110 baseline · CIFAR-10 | AUROC: 90.13 -> 86.21 | Hurt |
| train with label smoothing | ResNet110 baseline · CIFAR-10 | AUROC: 90.13 -> 81.41 | Hurt |
| train with focal loss | ResNet110 baseline · CIFAR-10 | AURC: 9.52 -> 9.89 | Hurt |
| post-hoc temperature scaling on validation set | ResNet baseline · CIFAR-10 | AUROC: 90.76 -> 89.15 | Hurt |
| temperature scaling fit on test set | ResNet baseline · CIFAR-10 | AUROC: 90.76 -> 90.88 | No clear effect |
| train with LogitNorm | ResNet110 baseline · CIFAR-100 | AUROC: 84.91 -> 79.56 | Hurt |
| use ODIN score | ResNet110 baseline · CIFAR-10 | AUROC: 90.13 -> 79.45 | Hurt |
| use energy score | ResNet110 baseline · CIFAR-10 | AUROC: 90.13 -> 84.72 | Hurt |
| apply FMFP flat-minima training (SAM+SWA) | ResNet110 baseline · CIFAR-10 | AUROC: 90.13 -> 94.07 | Improved |
| apply FMFP flat-minima training (SAM+SWA) | WRN-28-10 baseline · CIFAR-100 | AURC: 46.84 -> 36.99 | Improved |
| apply FMFP flat-minima training (SAM+SWA) | ResNet110 baseline · CIFAR-10 | ECE: 3.83 -> 0.45 | Improved |
| apply FMFP flat-minima training (SAM+SWA) | ResNet-50 baseline · Tiny-ImageNet | AURC: 119.01 -> 98.43 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OpenMix — OpenMix: Reviving Known Knowledge for Discovering Novel Visual Categories in an Open World (CVPR 2021) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rethinking Confidence Calibration for Failure Prediction (ECCV 2022 · similarity 0.96)
- A Call to Reflect on Evaluation Practices for Failure Detection in Image Classification (ICLR 2023 · similarity 0.89)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.87)
- Interpretable Failure Detection with Human-Level Concepts (AAAI 2025 · similarity 0.86)
- Typicalness-Aware Learning for Failure Detection (NeurIPS 2024 · similarity 0.86)
- Pointwise Information Measures as Confidence Estimators in Deep Neural Networks: A Comparative Study (ICML 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.
