Research map / Learning theory and algorithms / Prediction and decision losses
A Large-Scale Study of Probabilistic Calibration in Neural Network Regression
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 |
|---|---|---|---|
| Post-hoc quantile recalibration with empirical CDF map (Rec-EMP) and conformal prediction (CQR) versus regularization | MIX-NLL, MIX-CRPS, SQR-CRPS · 57 tabular datasets | PCE | Helped |
| Regularization objectives (PCE-KDE, QR, Trunc) added to neural regression training | MIX-NLL · 57 tabular datasets | PCE | Helped |
| Regularization added to base model (higher lambda) | MIX-NLL, MIX-CRPS, SQR-CRPS · 57 tabular datasets | CRPS | Mixed |
| Linear calibration map (Rec-LIN) instead of KDE map in quantile recalibration | MIX models · 57 tabular datasets | NLL | Hurt |
| Smooth KDE calibration map (Rec-KDE) instead of linear map | MIX-NLL quantile recalibration · 57 tabular datasets | NLL | Helped |
| Rec-EMP recalibration on MIX-NLL yields infinite NLL | MIX-NLL · 57 tabular datasets | NLL | Hurt |
| Rec-KDE recalibration on MIX-CRPS base model | MIX-CRPS · 57 tabular datasets | CRPS | Hurt |
| Combining regularization with a post-hoc method (ablation) | post-hoc methods (recalibration/CQR) · 57 tabular datasets | PCE | No clear effect |
| Post-hoc calibration fit on training data instead of held-out calibration set (ablation) | MIX-NLL, SQR-CRPS · 57 tabular datasets | PCE | Hurt |
| Post-hoc calibration fit on training data instead of calibration set (ablation) | MIX-NLL, SQR-CRPS · 57 tabular datasets | CRPS | No clear effect |
| Increase number of quantiles from 64 in SQR model (ablation) | SQR-CRPS · 57 tabular datasets | PCE, CRPS, sharpness | Helped |
| Single Gaussian instead of 3-component mixture (ablation) | MIX-NLL · 57 tabular datasets | CRPS, NLL, sharpness | Hurt |
Closest papers by meaning
- Self-Calibrating Conformal Prediction (NeurIPS 2024 · similarity 0.90)
- Generalized Venn and Venn-Abers Calibration with Applications in Conformal Prediction (ICML 2025 · similarity 0.90)
- Reassessing How to Compare and Improve the Calibration of Machine Learning Models (ICLR 2025 · similarity 0.88)
- Approximating Full Conformal Prediction for Neural Network Regression with Gauss-Newton Influence (ICLR 2025 · similarity 0.87)
- Tilt and Average : Geometric Adjustment of the Last Layer for Recalibration (ICML 2024 · similarity 0.87)
- On Temperature Scaling and Conformal Prediction of Deep Classifiers (ICML 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.