BriefGPT.xyz
Nov, 2018
深度学习中的单模型不确定性
Frequentist uncertainty estimates for deep learning
HTML
PDF
Natasa Tagasovska, David Lopez-Paz
TL;DR
本文提出了一种基于深度神经网络单模型的aleatoric和epistemic不确定性估计方法,分别为Simultaneous Quantile Regression(SQR)和Orthonormal Certificates(OCs),这些方法无需集成或重新训练深层模型就能达到竞争性能。
Abstract
We provide frequentist estimates of aleatoric and
epistemic uncertainty
for
deep neural networks
. To estimate
aleatoric uncertainty
we pro
→