Research map / Architectures and efficiency / RNNs and normalization
Sample-Dependent Adaptive Temperature Scaling for Improved Calibration
Tom Joy, Francesco Pinto, Ser-Nam Lim, Philip H.S. Torr, Puneet K. Dokania
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 |
|---|---|---|---|
| Replace a single global temperature with a sample-dependent temperature predicted from VAE feature-space likelihoods | ResNet50 trained with cross-entropy · CIFAR10 | ECE: 1.41 ± 0.43 -> 1.13 ± 0.60 | Improved |
| Replace a single global temperature with a sample-dependent temperature | WideResNet28-10 trained with cross-entropy · CIFAR100 | ECE: 3.76 ± 0.29 -> 2.95 ± 0.41 | Improved |
| Replace a single global temperature with a sample-dependent temperature | ResNet50 trained with cross-entropy · Tiny-ImageNet | ECE: 7.64 ± 1.53 -> 4.51 ± 1.76 | Improved |
| Replace a single global temperature with a sample-dependent temperature | ResNet50 trained with cross-entropy · CIFAR10-C | ECE: 12.72 ± 0.64 -> 10.86 ± 1.88 | Improved |
| Replace a single global temperature with a sample-dependent temperature | WideResNet28-10 trained with cross-entropy · CIFAR100-C | ECE: 12.28 ± 1.14 -> 12.17 ± 0.10 | Improved |
| Replace a single global temperature with a sample-dependent temperature | ResNet50 · CIFAR-10.1 | ECE: 4.57 ± 0.32 -> 3.67 ± 1.41 | Improved |
| Apply adaptive temperature scaling instead of vanilla temperature scaling | ResNet50 · CIFAR10 | accuracy: 93.13 ± 1.97 -> 93.13 ± 1.97 | No clear effect |
| Apply adaptive temperature scaling instead of vanilla temperature scaling | ResNet50 · CIFAR100 | AURRA-C: 93.62 ± 0.06 -> 94.03 ± 0.18 | Improved |
| Apply adaptive temperature scaling instead of vanilla temperature scaling | ResNet50 · Tiny-ImageNet | AURRA-C: 79.58 ± 2.06 -> 81.26 ± 0.49 | Improved |
| Use per-sample temperatures rather than one global temperature | CIFAR100-C models · CIFAR100-C | ECE: 12.28 ± 1.14 -> 12.17 ± 0.10; 11.78 ± 0.91 -> 11.69 ± 0.74 | Mixed |
Closest papers by meaning
- On the Limitations of Temperature Scaling for Distributions with Overlaps (ICLR 2024 · similarity 0.90)
- Consistency-Guided Temperature Scaling Using Style and Content Information for Out-of-Domain Calibration (AAAI 2024 · similarity 0.90)
- On Temperature Scaling and Conformal Prediction of Deep Classifiers (ICML 2025 · similarity 0.89)
- GETS: Ensemble Temperature Scaling for Calibration in Graph Neural Networks (ICLR 2025 · similarity 0.87)
- T-CIL: Temperature Scaling using Adversarial Perturbation for Calibration in Class-Incremental Learning (CVPR 2025 · similarity 0.87)
- A Benchmark Study on Calibration (ICLR 2024 · 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.
