Research map / Label-efficient and robust learning / Recognition and long-tail
H-Calibration: Rethinking Classifier Recalibration With Probabilistic Error-Bounded Objective
Wenjian Huang, Guiping Cao, Jiahao Xia, Jingkun Chen, Hao Wang, Jianguo Zhang
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 |
|---|---|---|---|
| recalibrate with h-calibration instead of no calibration | ResNet110 · CIFAR100 | ECE (top-label): 0.1848 -> 0.0118 | Improved |
| use h-calibration instead of temperature scaling | ResNet110 · CIFAR100 | ECE (top-label): 0.0187 -> 0.0118 | Improved |
| use h-calibration instead of Spline calibration | ResNet101pre on CARS · CARS | ECE (top-label): 0.0109 -> 0.0052 | Improved |
| use h-calibration instead of Spline calibration | ResNet50(NTS) on BIRDS · BIRDS | ECE (top-label): 0.0103 -> 0.0096 | Improved |
| use h-calibration instead of SoftBin | ResNet110 · CIFAR10 | ECE (top-label): 0.0030 -> 0.0043 | Hurt |
| use h-calibration instead of ScaBin | DenseNet40 · CIFAR10 | ECE (top-label): 0.0020 -> 0.0073 | Hurt |
| use h-calibration instead of TS+LECE | WideResNet32 · CIFAR100 | ECE (top-label): 0.0088 -> 0.0149 | Hurt |
| use h-calibration instead of GP | WideResNet32 · CIFAR10 | ECE (top-label): 0.0013 -> 0.0015 | Mixed |
| use h-calibration instead of Spline calibration | ResNet101 · CARS | ECE (top-label): 0.0050 -> 0.0082 | Hurt |
| use h-calibration instead of temperature scaling | ResNet101pre · CARS | ECE (top-label): 0.0297 -> 0.0052 | Improved |
Closest papers by meaning
- Meta-Cal: Well-controlled Post-hoc Calibration by Ranking (ICML 2021 · similarity 0.92)
- A Consistent and Differentiable Lp Canonical Calibration Error Estimator (NeurIPS 2022 · similarity 0.91)
- Reassessing How to Compare and Improve the Calibration of Machine Learning Models (ICLR 2025 · similarity 0.90)
- A Benchmark Study on Calibration (ICLR 2024 · similarity 0.89)
- Tilt and Average : Geometric Adjustment of the Last Layer for Recalibration (ICML 2024 · similarity 0.88)
- Divide et Calibra: Multiclass Local Calibration via Vector Quantization (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.
