BriefGPT.xyz
Jul, 2023
使用浅层单变量ReLU网络的噪声插值学习
Noisy Interpolation Learning with Shallow Univariate ReLU Networks
HTML
PDF
Nirmit Joshi, Gal Vardi, Nathan Srebro
TL;DR
研究了使用最小范数两层ReLU网络进行有噪单变量回归时的渐近过拟合行为，发现对于L1损失和p<2的任何Lp损失缓解了过拟合，但对于p≥2却是灾难性的。
Abstract
We study the
asymptotic
overfitting
behavior of
interpolation
with minimum norm ($\ell_2$ of the weights) two-layer
→