Research map / Label-efficient and robust learning / Recognition and long-tail
Quantile‑Adaptive Temperature Scaling for Confidence Calibration
Omprakash Chakraborty, Leo Fillioux, Ismail Ayed, Jose Dolz
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 piecewise linear temperature (K=10) instead of linear (ablation) | QaTS · CIFAR-100-LT R-100 (ViT-B) | ECE (%): 8.37 -> 8.76 | Hurt |
| use piecewise linear temperature (K=10) instead of linear (ablation) | QaTS · CIFAR-100-C (ViT-B) | ECE (%): 9.89 -> 10.64 | Hurt |
| use piecewise linear temperature (K=2) instead of linear (ablation) | QaTS · CIFAR-100-C (ViT-B) | ECE (%): 9.89 -> 10.14 | Hurt |
| use piecewise linear temperature (K=4) instead of linear (ablation) | QaTS · CIFAR-100 clean (ViT-B) | ECE (%): 2.26 -> 2.24 | No clear effect |
| use piecewise linear temperature (K=4) instead of linear (ablation) | QaTS · CIFAR-100-LT R-100 (ViT-B) | ECE (%): 8.37 -> 8.68 | Hurt |
| apply QaTS on top of Brier-loss training | Brier-trained RN-50 · CIFAR-100 | ECE (%): 5.54 -> 2.72 | Improved |
| replace FeatClip with QaTS (accuracy preservation) | FeatClip · 161 dataset-architecture combinations | accuracy: QaTS higher accuracy in 94% of settings | Improved |
| replace FeatClip with QaTS (segmentation) | FeatClip · VOC-2012 | mIoU: 69.78 -> 69.51 | Improved |
| replace FeatClip with QaTS (text) | FeatClip · NewsGroup-20 | accuracy: 84.9 -> 84.7 | Improved |
| use QaTS instead of FeatClip under mild long-tail | FeatClip · CIFAR-100-LT R-10, RN-50 | ECE (%): 6.62 -> 3.76 | Improved |
| use QaTS instead of TS | TS · CIFAR-100 (RN-50) | ECE (%): 5.32 -> 2.56 | Improved |
| use QaTS instead of FeatClip on ViT-S CIFAR-10 | FeatClip · CIFAR-10 (ViT-S) | ECE (%): 0.62 -> 0.51 | Improved |
Closest papers by meaning
- Consistency-Guided Temperature Scaling Using Style and Content Information for Out-of-Domain Calibration (AAAI 2024 · similarity 0.89)
- On the Limitations of Temperature Scaling for Distributions with Overlaps (ICLR 2024 · similarity 0.87)
- HoTS: Homophily-Aware Temperature Scaling for Graph Neural Network Calibration (NeurIPS 2026 · similarity 0.86)
- On Temperature Scaling and Conformal Prediction of Deep Classifiers (ICML 2025 · similarity 0.86)
- T-CIL: Temperature Scaling using Adversarial Perturbation for Calibration in Class-Incremental Learning (CVPR 2025 · similarity 0.86)
- GETS: Ensemble Temperature Scaling for Calibration in Graph Neural Networks (ICLR 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.
