Research map / Learning theory and algorithms / Prediction and decision losses
A Consistent and Differentiable Lp Canonical Calibration Error Estimator
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 |
|---|---|---|---|
| add L1 KDE-XE calibration regularizer to cross-entropy | DenseNet-40 · DermaMNIST canonical L1 ECE-KDE | canonical L1 ECE-KDE: 0.573 -> 0.366 | Helped |
| add L1 KDE-XE calibration regularizer to cross-entropy | DenseNet-40 · Kather canonical L1 ECE-KDE | canonical L1 ECE-KDE: 0.244 -> 0.114 | Helped |
| add L1 KDE-XE calibration regularizer to cross-entropy | ResNet-110 (SD) · CIFAR-10 canonical L1 ECE-KDE | canonical L1 ECE-KDE: 0.132 -> 0.178 | Hurt |
| add L1 KDE-XE calibration regularizer to cross-entropy | ResNet-110 · CIFAR-10 canonical L1 ECE-KDE | canonical L1 ECE-KDE: 0.133 -> 0.126 | Helped |
| add L1 KDE-XE regularizer vs MMCE regularizer | DenseNet-40 on DermaMNIST · canonical L1 ECE-KDE | canonical L1 ECE-KDE: 0.514 -> 0.366 | Helped |
| add L1 KDE-XE regularizer | ResNet-110 (SD) · Kather accuracy | accuracy: 0.870 -> 0.914 | Helped |
| add L1 KDE-XE regularizer | Wide-ResNet-28-10 · Kather accuracy | accuracy: 0.933 -> 0.921 | Hurt |
| add post-hoc temperature scaling to KDE-XE (ablation) | L1 KDE-XE · canonical L1 ECE-KDE | canonical L1 ECE-KDE: 0.311 -> 0.289 (Kather ResNet-110); 0.467 -> 0.516 (DermaMNIST ResNet-110) | Mixed |
| add post-hoc temperature scaling to cross-entropy | XE · DermaMNIST canonical L1 ECE-KDE | canonical L1 ECE-KDE: 0.579 -> 0.602 | Hurt |
| add KDE-XE regularizer | XE training · CIFAR-10 training time | seconds per epoch: 51.8 -> 53 | No clear effect |
Closest papers by meaning
- Calibrated and Sharp Uncertainties in Deep Learning via Density Estimation (ICML 2022 · similarity 0.90)
- Reassessing How to Compare and Improve the Calibration of Machine Learning Models (ICLR 2025 · similarity 0.89)
- The Minimax Rate of Second-Order Calibration (NeurIPS 2026 · similarity 0.88)
- Divide et Calibra: Multiclass Local Calibration via Vector Quantization (NeurIPS 2026 · similarity 0.88)
- Truthful Calibration Errors for Multi-Class Prediction (NeurIPS 2026 · similarity 0.88)
- LaSCal: Label-Shift Calibration without target labels (NeurIPS 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.