BriefGPT.xyz
Jun, 2022
标签平滑的礼物：通过辅助分类器和标签噪声实现自适应标签平滑的强健训练
ALASCA: Rethinking Label Smoothing for Deep Learning Under Label Noise
HTML
PDF
Jongwoo Ko, Bongsoo Yi, Se-Young Yun
TL;DR
该论文提出了一种名为ALASCA的方法，该方法通过自适应标签平滑和辅助分类器的结合，提供了一种高效的抵抗标签噪声的健壮特征提取器，从而解决了深度学习中存在的关于噪声标签问题的挑战。
Abstract
As label noise, one of the most popular distribution shifts, severely degrades deep
neural networks
' generalization performance, robust training with
noisy labels
is becoming an important task in modern deep lear
→