Research map / Label-efficient and robust learning / Recognition and long-tail
Parameterized Temperature Scaling for Boosting the Expressive Power in Post-Hoc Uncertainty Calibration
Christian Tomani, Daniel Cremers, Florian Buettner
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 non-linear neural-network temperature instead of linear model (ablation) | PTS · ImageNet (5 models) | ECE: 1.36 -> 1.28 | Improved |
| replace ETS (3 fixed temperatures) with prediction-specific temperatures (PTS) | ETS · CIFAR-100 VGG19 | ECE: 2.11 -> 0.82 | Improved |
| replace TS with PTS | TS · ImageNet MobileNetV2 | ECE: 5.91 -> 1.43 | Improved |
| PTS versus ETS on simple settings | PTS · CIFAR-10 (ECE_KDE) | ECE: 1.82 -> 1.49 | No clear effect |
| increase hidden-layer size of the temperature network (ablation) | PTS · ImageNet MobileNetV2 | ECE: additional improvement in ECE of 52% | Improved |
| ensemble of 3 temperatures instead of single temperature (ablation) | TS · ImageNet MobileNetV2 | ECE: improved ECE (50% for MobileNetV2 trained on ImageNet) | Improved |
| reduce validation set to 10% of the standard size (ablation) | PTS · ImageNet (5 models, normalized ECE) | ECE relative to PTS: ETS 0.40, TS 0.71, IRM 0.83 relative to PTS at 10% | Hurt |
| vary number of ECE bins from 5 to 20 (ablation) | PTS · ImageNet (5 models) | ECE | No clear effect |
| use non-parametric IROvA instead of TS-based calibrators on small validation sets (ablation) | IROvA · ImageNet MobileNetV2 | ECE: IROvA 6.84 vs PTS 1 at 10% validation | Hurt |
Closest papers by meaning
- Consistency-Guided Temperature Scaling Using Style and Content Information for Out-of-Domain Calibration (AAAI 2024 · similarity 0.91)
- On the Limitations of Temperature Scaling for Distributions with Overlaps (ICLR 2024 · similarity 0.88)
- On Temperature Scaling and Conformal Prediction of Deep Classifiers (ICML 2025 · similarity 0.87)
- A Benchmark Study on Calibration (ICLR 2024 · similarity 0.87)
- T-CIL: Temperature Scaling using Adversarial Perturbation for Calibration in Class-Incremental Learning (CVPR 2025 · similarity 0.87)
- HoTS: Homophily-Aware Temperature Scaling for Graph Neural Network Calibration (NeurIPS 2026 · 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.
