BriefGPT.xyz
Jul, 2020
浅层网络1-路径范数的高效近端映射
Efficient Proximal Mapping of the 1-path-norm of Shallow Networks
HTML
PDF
Fabian Latorre, Paul Rolland, Nadav Hallak, Volkan Cevher
TL;DR
研究了浅层神经网络的1-path-范数，发现它可以通过闭合形式的近端算子进行计算，且可以提供网络李普希茨常数的上界，是训练鲁棒性强的网络的有力工具。与L1-范数以及基于层的李普希茨常数的制约相比，1-path-范数的近端映射具有更好的鲁棒性-准确性平衡。
Abstract
We demonstrate two new important properties of the
1-path-norm
of
shallow neural networks
. First, despite its non-smoothness and non-convexity it allows a closed form
→