BriefGPT.xyz
May, 2018
深度神经分类器的偏差减少不确定性估计
Boosting Uncertainty Estimation for Deep Neural Classifiers
HTML
PDF
Yonatan Geifman, Guy Uziel, Ran El-Yaniv
TL;DR
本研究提出了一种基于模型历史快照的算法，用于在非贝叶斯深度神经分类中，有选择地估计高度自信点的不确定性，这解决了从已训练网络中提取不确定信号的已知方法所带来的偏差估计问题，研究表明所提出的算法比所有已知方法的不确定性估计结果更加准确。
Abstract
We consider the problem of
uncertainty estimation
in the context of (non-Bayesian)
deep neural classification
. All current methods are based on extracting uncertainty signals from a trained network optimized to s
→