BriefGPT.xyz
Feb, 2020
为什么深度残差网络比深度前馈网络具有更好的泛化性能？——— 一种神经切向核的视角
Why Do Deep Residual Networks Generalize Better than Deep Feedforward Networks? -- A Neural Tangent Kernel Perspective
HTML
PDF
Kaixuan Huang, Yuqing Wang, Molei Tao, Tuo Zhao
TL;DR
该论文从神经切向核的角度研究了深度学习中深层残差网络相对于前馈网络的泛化性能优势，并给出了数字结果支持。
Abstract
Deep residual networks (
resnets
) have demonstrated better
generalization
performance than deep
feedforward networks
(FFNets). However, the
→