Research map / Label-efficient and robust learning / Recognition and long-tail
Uncertainty Calibration with Energy Based Instance-wise Scaling in the Wild Dataset
Mijoo Kim, Junseok Kwon
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 |
|---|---|---|---|
| use energy-based instance-wise temperature scaling instead of IRM/SPLINE | post-hoc calibration · CIFAR10 (SVHN tuning), averaged over corruption severities | ECE: 5.78 -> 4.47 (DenseNet201 vs IRM) | Improved |
| calibrate with energy score instead of plain temperature scaling on ImageNet-R | DenseNet201 · ImageNet-R | ECE: 24.35 -> 15.22 | Improved |
| add DAC preprocessing to the proposed method (ablation) | Ours · CIFAR100-C | ECE: 11.92 -> 8.99 | Improved |
| add DAC preprocessing to the proposed method (ablation) | Ours · CIFAR10-C | ECE: 5.98 -> 7.81 | Hurt |
| tune without semantic OOD data (ablation) | Ours · CIFAR100 (SVHN tuning), avg ECE | ECE: 7.35 -> 10.45 (VGG19) | Hurt |
| tune without semantic OOD data (ablation) | Ours · CIFAR10 (SVHN tuning), avg ECE | ECE: 4.47 -> 5.44 (DenseNet201) | Hurt |
| switch tuning OOD dataset from SVHN to Texture (ablation) | Ours · CIFAR10 DenseNet201 | ECE: 4.47 -> 4.51 | No clear effect |
| apply instance-wise calibration to OOD detection | DenseNet201 · ImageNet-1k / Texture | AU-ROC: 81.43 -> 82.25 | Improved |
| apply method to ViT-B-16 (already well-calibrated transformer) | ViT-B-16 · ImageNet-C avg | ECE: 8.73 -> 8.45 | Improved |
Closest papers by meaning
- Improving model calibration with accuracy versus uncertainty optimization (NeurIPS 2020 · similarity 0.90)
- Better Uncertainty Calibration via Proper Scores for Classification and Beyond (NeurIPS 2022 · similarity 0.90)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.89)
- A Benchmark Study on Calibration (ICLR 2024 · similarity 0.89)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.88)
- Improving Accuracy and Calibration via Differentiated Deep Mutual Learning (CVPR 2025 · similarity 0.87)
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.
