Research map / Learning theory and algorithms / Sample complexity
Understanding the Under-Coverage Bias in Uncertainty Estimation
Yu Bai, Song Mei, Huan Wang, Caiming Xiong
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 |
|---|---|---|---|
| increase model capacity from linear to MLP-3-512 | quantile regression · Community, alpha=0.9 | coverage (%): 88.63 -> 63.09 | Hurt |
| increase MLP width from 64 to 512 | MLP quantile regression · STAR, alpha=0.9 | coverage (%): 83.14 -> 69.71 | Hurt |
| increase MLP width from 64 to 512 | MLP quantile regression · MEPS_19, alpha=0.9 | coverage (%): 85.46 -> 78.55 | Hurt |
| increase MLP width from 64 to 512 | MLP quantile regression · Bike, alpha=0.9 | coverage (%): 88.75 -> 87.67 | Hurt |
| freeze representation layers of width-512 MLP, train last layer only (ablation) | MLP-3-512 · Community, alpha=0.9 | coverage (%): 63.09 -> 87.85 | Improved |
| freeze representation layers of width-512 MLP, train last layer only (ablation) | MLP-3-512 · MEPS_21, alpha=0.9 | coverage (%): 73.09 -> 89.15 | Improved |
| use linear model on datasets with small d/n | quantile regression · Bike, MEPS_19, MEPS_20, MEPS_21 | coverage (%): 89.64, 90.09, 90.06, 89.99 | No clear effect |
| use linear model on dataset with highest d/n of about 1/16 | linear quantile regression · Community, alpha=0.9 | coverage (%): 88.63 vs 90.00 | Hurt |
| increase kappa=d/n from 0.01 to 0.5 (fewer training samples) | linear quantile regression · MEPS_20 true labels | coverage (%): 89.83 -> 84.75 | Hurt |
| increase kappa=d/n from 0.01 to 0.5 (fewer training samples) | linear quantile regression · MEPS_20 pseudo-labels | coverage (%): 90.05 -> 83.70 | Hurt |
| replace true labels by pseudo-labels from fitted linear model | linear quantile regression · MEPS_20 | coverage (%): 84.75 vs 83.70 at kappa 0.5 | No clear effect |
| increase quantile level alpha at fixed kappa | quantile regression · simulated Gaussian linear model | coverage (%) | Hurt |
Closest papers by meaning
- Relaxed Quantile Regression: Prediction Intervals for Asymmetric Noise (ICML 2024 · similarity 0.90)
- The Bias-Variance Tradeoff in Data-Driven Optimization: A Local Misspecification Perspective (NeurIPS 2025 · similarity 0.85)
- Rectifying Conformity Scores for Better Conditional Coverage (ICML 2025 · similarity 0.85)
- Efficient Analytic Uncertainty Quantification for Multimodal Regression (NeurIPS 2026 · similarity 0.85)
- QuEst: Enhancing Estimates of Quantile-Based Distributional Measures Using Model Predictions (ICML 2025 · similarity 0.85)
- Sub-Gaussian Confidence Intervals for Heavy-Tailed Data: Characterizing the Limits of Inference (NeurIPS 2026 · similarity 0.85)
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.
