BriefGPT.xyz
Apr, 2016
DisturbLabel: 在损失层上对CNN进行正则化
DisturbLabel: Regularizing CNN on the Loss Layer
HTML
PDF
Lingxi Xie, Jingdong Wang, Zhen Wei, Meng Wang, Qi Tian
TL;DR
本文提出了DisturbLabel算法，通过在每次迭代中随机替换部分标签为不正确的值，使神经网络模型训练不会出现过拟合，并在几个流行的图像识别数据集上展示了有竞争力的识别结果。
Abstract
During a long period of time we are combating
over-fitting
in the
cnn training
process with model
regularization
, including weight decay,
→