Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Calibrated and Sharp Uncertainties in Deep Learning via Density Estimation
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 |
|---|---|---|---|
| distribution recalibration (density estimation of recalibrator) added on top of base model | Bayesian linear regression, Bayesian NN, deep ensembles · UCI regression (8 datasets) | check score (CHK) | Helped |
| distribution recalibration added to base model | Bayesian linear regression · UCI regression (yacht) | CHK: 2.43 -> 0.35 | Helped |
| distribution recalibration added to base model | deep ensemble · UCI regression (mpg) | MAE: 7.66 -> 9.01 | Hurt |
| distribution recalibration added to base model | Bayesian neural network · UCI regression (mpg) | CHK: 1.19 -> 1.08 | Helped |
| neural network distribution recalibrator added to CNN classifier | CNN (ResNet50) image classifiers · MNIST, SVHN, CIFAR10 | calibration error: MNIST 0.3166 -> 0.1030; SVHN 0.5975 -> 0.2674; CIFAR10 0.5848 -> 0.1091 | Helped |
| neural network distribution recalibrator added to CNN classifier | CNN (ResNet50) image classifiers · MNIST, SVHN, CIFAR10 | accuracy: MNIST 0.9952 -> 0.9951; SVHN 0.9508 -> 0.9509; CIFAR10 0.9179 -> 0.9163 | No clear effect |
| Platt scaling (softmax regression recalibration) added to CNN classifier | CNN (ResNet50) image classifiers · MNIST, SVHN, CIFAR10 | calibration error: MNIST 0.3166 -> 0.2212; SVHN 0.5975 -> 0.3278; CIFAR10 0.5848 -> 0.2233 | Helped |
| distribution recalibration vs Platt scaling on CNN classifiers (ablation) | CNN (ResNet50) image classifiers · MNIST, SVHN, CIFAR10 | calibration error: Ours lower than Platt on all three datasets | Helped |
Closest papers by meaning
- Better Uncertainty Calibration via Proper Scores for Classification and Beyond (NeurIPS 2022 · similarity 0.91)
- Reassessing How to Compare and Improve the Calibration of Machine Learning Models (ICLR 2025 · similarity 0.90)
- A Benchmark Study on Calibration (ICLR 2024 · similarity 0.88)
- Calibration Bottleneck: Over-compressed Representations are Less Calibratable (ICML 2024 · similarity 0.87)
- Self-Calibrating Vicinal Risk Minimisation for Model Calibration (CVPR 2024 · similarity 0.87)
- Uncertainty Weighted Gradients for Model Calibration (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.