Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Uncertainty Quantification and Deep Ensembles
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Average 30 mixup-trained models without calibration | single model · CIFAR10 1000 samples | ECE (%): 7.31 -> 13.9 | Hurt |
| Apply temperature scaling after pooling (pool-then-calibrate) (ablation) | vanilla pooling · CIFAR10 1000 samples | ECE (%): 13.9 -> 4.9 | Helped |
| Apply temperature scaling after pooling (ablation) | vanilla pooling · CIFAR100 5000 samples | ECE (%): 17.8 -> 2.1 | Helped |
| Apply temperature scaling after pooling (ablation) | vanilla pooling · Imagenette 1000 samples | ECE (%): 18.2 -> 3.5 | Helped |
| Pool then calibrate instead of calibrating each model then averaging (ablation) | temp-scaled ensemble · CIFAR100 full data | ECE (%): 13.2 -> 3.4 | Helped |
| Ensembling models when individuals are over-confident | individual unscaled models · CIFAR100 full data | ECE (%): 9.8 -> 5.9 | Helped |
| Ensembling models when individuals are already calibrated | temp-scaled models · CIFAR100 full data | ECE (%): 2.1 -> 13.2 | Hurt |
| Pool-then-calibrate instead of vanilla ensemble | deep ensemble · CIFAR-10C Gaussian noise | ECE (%): 15.13 -> 1.65 | Helped |
| Pool-then-calibrate instead of vanilla ensemble | deep ensemble · CIFAR10 vs CIFAR100 OOD | entropy median gap: 0.359 -> 0.521 | Helped |
| Add mixup augmentation (ablation) | single model · CIFAR10 1000 samples | test accuracy (%): 63.73 -> 66.45 | Helped |
| Add mixup augmentation (ablation) | single model · CIFAR10 1000 samples | ECE (%): 20.7 -> 7.03 | Helped |
| Use median or trimmed pooling instead of mean (ablation) | pool-then-calibrate · multiple datasets | calibration metrics | No clear effect |
Closest papers by meaning
- Combining Ensembles and Data Augmentation Can Harm Your Calibration (ICLR 2021 · similarity 0.90)
- Better Uncertainty Calibration via Proper Scores for Classification and Beyond (NeurIPS 2022 · similarity 0.90)
- Mix-n-Match : Ensemble and Compositional Methods for Uncertainty Calibration in Deep Learning (ICML 2020 · similarity 0.89)
- Calibrated and Sharp Uncertainties in Deep Learning via Density Estimation (ICML 2022 · similarity 0.89)
- On the Limitations of Temperature Scaling for Distributions with Overlaps (ICLR 2024 · similarity 0.87)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 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.