BriefGPT.xyz
Apr, 2021
输出正则化为什么可以缓解标签噪声问题探究
An Exploration into why Output Regularization Mitigates Label Noise
HTML
PDF
Neta Shoham, Tomer Avidor, Nadav Israel
TL;DR
本研究探讨利用输出正则化来缓解标签噪声对于损失函数的影响,发现当正则化系数趋于无穷大时,这些损失函数变得对称且稳健。因此,正则化系数可以被看作控制对称性和噪声鲁棒性的超参数。
Abstract
label noise
presents a real challenge for
supervised learning
algorithms. Consequently, mitigating
label noise
has attracted immense resea
→