Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
CLEAR: Calibrated Learning for Epistemic and Aleatoric Risk
Ilia Azizi, Juraj Bodik, Jakob Heiss, Bin Yu
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 |
|---|---|---|---|
| combine aleatoric and epistemic uncertainty with two calibrated parameters (CLEAR) instead of PCS alone | PCS ensemble · 17 regression datasets | quantile loss: 15.8% improvement | Improved |
| add epistemic uncertainty and tune lambda on top of residual-based aleatoric estimate (ablation) | ALEATORIC-R · 17 regression datasets | quantile loss: 9.4% | Improved |
| fit aleatoric quantiles on residuals instead of directly on targets (ablation) | bootstrapped CQR · 17 regression datasets | quantile loss: 34.4% -> 9.4% gap to CLEAR | Improved |
| replace UACQR with CLEAR | UACQR-S / UACQR-P · 17 regression datasets | NCIW, AISL, width: CLEAR outperforms both on 14 of 17 datasets | Improved |
| replace UACQR with CLEAR on insurance | UACQR-S / UACQR-P · insurance | NCIW: -15.5% (vs UACQR-S), -18.6% (vs UACQR-P) | Hurt |
| apply CLEAR to Deep Ensembles | Deep Ensembles · 17 regression datasets | NCIW: +28.57% | Improved |
| apply CLEAR to SQR | SQR · 17 regression datasets | NCIW: +13.36% | Improved |
| apply CLEAR to conformally calibrated SQR | SQR-conformal · 17 regression datasets | PICP: -0.15% | Hurt |
| learn both gamma1 and lambda instead of fixing lambda=1 or gamma1=1 (ablation) | CLEAR · 17 regression datasets | interval quality | Mixed |
| use CLEAR instead of PCS or CQR with 2 features | Ames Housing · Ames Housing, 2 features | quantile loss: 3,818 -> 3,131 | Improved |
| use CLEAR instead of PCS with all features | Ames Housing · Ames Housing, all features | quantile loss: 1,922 -> 1,923 | No clear effect |
| reduce training data to 20% so lambda grows (ablation) | CLEAR · Ames Housing, all features, 20% data | lambda: 14.45 -> 100 | Mixed |
Closest papers by meaning
- Rethinking Aleatoric and Epistemic Uncertainty (ICML 2025 · similarity 0.89)
- Provable Uncertainty Decomposition via Higher-Order Calibration (ICLR 2025 · similarity 0.88)
- Evaluating Epistemic Uncertainty: Beyond OOD Detection and Active Learning (NeurIPS 2026 · similarity 0.88)
- Epistemic Uncertainty Quantification For Pre-Trained Neural Networks (CVPR 2024 · similarity 0.87)
- Uncertainty Quantification for Deep Regression using Contextualised Normalizing Flows (NeurIPS 2025 · similarity 0.87)
- Is Epistemic Uncertainty Faithfully Represented by Evidential Deep Learning Methods? (ICML 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.
