Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Improving the Reliability for Confidence Estimation
Haoxuan Qu, Yanchao Li, Lin Geng Foo, Jason Kuen, Jiuxiang Gu, Jun 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 |
|---|---|---|---|
| add virtual training and testing set construction for label imbalance only (ablation) | SLURP · KITTI Eigen test (depth confidence, trained on KITTI) | AUSE-Absrel: 4.36 -> 3.84 | Improved |
| add virtual set construction for OOD inputs only (ablation) | SLURP · KITTI Eigen test (depth confidence, trained on KITTI) | AUSE-Absrel: 4.36 -> 3.90 | Improved |
| combine both set constructions (full framework) (ablation) | SLURP · KITTI Eigen test (depth confidence, trained on KITTI) | AUSE-Absrel: 4.36 -> 3.62 | Improved |
| add full framework | SLURP · Cityscapes test (depth confidence, trained on KITTI) | AUSE-Absrel: 10.90 -> 5.32 | Improved |
| add full framework instead of Resampling+Mixup | SLURP · Cityscapes test (depth confidence, trained on KITTI) | AUSE-Absrel: 9.64 -> 5.32 | Improved |
| add Mixup (ablation) | SLURP · KITTI Eigen test (depth confidence, trained on KITTI) | AUSE-Absrel: 4.36 -> 4.15 | Improved |
| add Reweight (ablation) | SLURP · Cityscapes test (depth confidence, trained on KITTI) | AUSE-Absrel: 10.90 -> 10.41 | No clear effect |
| use meta-learning (second-order virtual train/test) instead of joint training (ablation) | SLURP + sets · Cityscapes test (depth confidence, fine-tuned on Cityscapes) | AUSE-RMSE: 2.47 -> 0.60 | Improved |
| use meta-learning instead of joint training (ablation) | SLURP + sets · Cityscapes test (depth confidence, fine-tuned on Cityscapes) | AUROC: 0.867 -> 0.933 | Improved |
| add full framework | TCP · MNIST test (confidence estimation) | AUPR-Error: 45.89 -> 47.05 | Improved |
| add full framework | TCP · CIFAR-10 test (confidence estimation) | FPR-95%-TPR: 44.94 -> 44.69 | Improved |
| add OOD-only construction (ablation) | TCP · MNIST test (confidence estimation) | FPR-95%-TPR: 3.33 -> 3.33 | No clear effect |
Closest papers by meaning
- Towards More Reliable Confidence Estimation (TPAMI 2023 · similarity 0.98)
- Confidence Estimation Using Unlabeled Data (ICLR 2023 · similarity 0.88)
- Quantitative Evaluation of Confidence Measures in a Machine Learning World (ICCV 2017 · similarity 0.87)
- Learning to Predict Stereo Reliability Enforcing Local Consistency of Confidence Maps (CVPR 2017 · similarity 0.87)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.85)
- Pointwise Information Measures as Confidence Estimators in Deep Neural Networks: A Comparative Study (ICML 2025 · similarity 0.84)
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.
