BriefGPT.xyz
Mar, 2021
关于过拟合两层神经切向核模型的泛化能力
On the Generalization Power of Overfitted Two-Layer Neural Tangent Kernel Models
HTML
PDF
Peizhong Ju, Xiaojun Lin, Ness B. Shroff
TL;DR
本文研究具有ReLU激活函数且没有偏差项的两层神经网络的神经切向核（NTK）模型的min（L2）-norm过拟合解的泛化性能，并显示随着神经元数目p的增加，测试误差表现出不同于具有简单傅里叶或高斯特征的过度参数化线性模型的“双峰现象”的特征。
Abstract
In this paper, we study the
generalization performance
of min $\ell_2$-norm
overfitting solutions
for the neural tangent kernel (NTK) model of a two-layer
→