BriefGPT.xyz
May, 2018
探究过度参数化在神经网络泛化中的作用
Towards Understanding the Role of Over-Parametrization in Generalization of Neural Networks
HTML
PDF
Behnam Neyshabur, Zhiyuan Li, Srinadh Bhojanapalli, Yann LeCun, Nathan Srebro
TL;DR
本研究提出了基于单元能力的复杂度度量，为两层ReLU网络提供了更紧密的泛化界限，这可能有助于解释神经网络过参数化的泛化改进现象。同时，我们还提出了一个匹配的Rademacher复杂性下限，该下限优于之前神经网络的容量下限。
Abstract
Despite existing work on ensuring
generalization
of
neural networks
in terms of scale sensitive
complexity measures
, such as norms, margin
→