Research map / Learning theory and algorithms / Kernels and regression theory
Quantile regression with ReLU Networks: Estimators and minimax rates
Oscar Hernan Madrid Padilla, Wesley Tansey, Yanzhen Chen
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 |
|---|---|---|---|
| use quantile (check) loss instead of squared error loss | ReLU network · Scenario 1, n=10000, tau=0.50 | squared error from true quantile: 0.94 -> 0.01 | Improved |
| use quantile loss instead of squared error loss under heavy-tailed errors | ReLU network · Scenario 5 (t(3) errors), n=10000, tau=0.50 | squared error from true quantile: 30.60 -> 0.17 | Improved |
| use quantile loss instead of squared error loss | ReLU network · Scenario 4, n=10000, tau=0.50 | squared error from true quantile: 0.05 -> 0.07 | Hurt |
| use quantile loss instead of squared error loss in small sample | ReLU network · Scenario 2, n=100, tau=0.50 | squared error from true quantile: 0.28 -> 2.21 | Hurt |
| use ReLU quantile network instead of quantile spline on smooth low-dimensional function | Quantile Spline · Scenario 2, n=10000, tau=0.50 | squared error from true quantile: 0.00 -> 0.03 | Hurt |
| use quantile network instead of quantile forests | Quantile Forests · Scenario 5, n=10000, tau=0.50 | squared error from true quantile: 3.70 -> 0.17 | Improved |
| use quantile network instead of quantile spline with discontinuities and heteroscedastic errors | Quantile Spline · Scenario 3, n=10000, tau=0.50 | squared error from true quantile: 0.07 -> 0.03 | Improved |
| use quantile network instead of quantile forests at extreme quantile | Quantile Forests · Scenario 1, n=10000, tau=0.95 | squared error from true quantile: 13.89 -> 0.03 | Improved |
| increase sample size from n=100 to n=10000 | Quantile Network · Scenario 1, tau=0.50 | squared error from true quantile: 0.46 -> 0.01 | Improved |
| use quantile loss instead of squared error loss for multivariate response | ReLU network · Scenario 6 (multivariate t3), n=10000 | mean squared error: 0.89 -> 0.01 | Improved |
| use quantile loss instead of squared error loss for multivariate response in small sample | ReLU network · Scenario 7 (Laplace), n=100 | mean squared error: 0.24 -> 1.67 | Hurt |
| use Quantile Network instead of Geometric Quantile | Geometric Quantile · Scenario 6, n=100 | mean squared error: 0.54 -> 0.49 | No clear effect |
Closest papers by meaning
- Deep Nonparametric Quantile Regression under Covariate Shift (JMLR 2024 · similarity 0.91)
- Nonparametric Quantile Regression with ReLU-Activated Recurrent Neural Networks (NeurIPS 2025 · similarity 0.90)
- Nonparametric Estimation of Non-Crossing Quantile Regression Process with Deep ReQU Neural Networks (JMLR 2024 · similarity 0.90)
- Distributed High-Dimensional Quantile Regression: Estimation Efficiency and Support Recovery (ICML 2024 · similarity 0.86)
- Relaxed Quantile Regression: Prediction Intervals for Asymmetric Noise (ICML 2024 · similarity 0.85)
- Simulation-Based Inference with Quantile Regression (ICML 2024 · 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.
