AI Paper Atlas

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

JMLR 2022 · Learning theory and algorithms › Kernels and regression theory · approach: Statistical perspectives on generalization

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 changedWhereMeasuredEffect
use quantile (check) loss instead of squared error lossReLU network · Scenario 1, n=10000, tau=0.50squared error from true quantile: 0.94 -> 0.01Improved
use quantile loss instead of squared error loss under heavy-tailed errorsReLU network · Scenario 5 (t(3) errors), n=10000, tau=0.50squared error from true quantile: 30.60 -> 0.17Improved
use quantile loss instead of squared error lossReLU network · Scenario 4, n=10000, tau=0.50squared error from true quantile: 0.05 -> 0.07Hurt
use quantile loss instead of squared error loss in small sampleReLU network · Scenario 2, n=100, tau=0.50squared error from true quantile: 0.28 -> 2.21Hurt
use ReLU quantile network instead of quantile spline on smooth low-dimensional functionQuantile Spline · Scenario 2, n=10000, tau=0.50squared error from true quantile: 0.00 -> 0.03Hurt
use quantile network instead of quantile forestsQuantile Forests · Scenario 5, n=10000, tau=0.50squared error from true quantile: 3.70 -> 0.17Improved
use quantile network instead of quantile spline with discontinuities and heteroscedastic errorsQuantile Spline · Scenario 3, n=10000, tau=0.50squared error from true quantile: 0.07 -> 0.03Improved
use quantile network instead of quantile forests at extreme quantileQuantile Forests · Scenario 1, n=10000, tau=0.95squared error from true quantile: 13.89 -> 0.03Improved
increase sample size from n=100 to n=10000Quantile Network · Scenario 1, tau=0.50squared error from true quantile: 0.46 -> 0.01Improved
use quantile loss instead of squared error loss for multivariate responseReLU network · Scenario 6 (multivariate t3), n=10000mean squared error: 0.89 -> 0.01Improved
use quantile loss instead of squared error loss for multivariate response in small sampleReLU network · Scenario 7 (Laplace), n=100mean squared error: 0.24 -> 1.67Hurt
use Quantile Network instead of Geometric QuantileGeometric Quantile · Scenario 6, n=100mean squared error: 0.54 -> 0.49No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy